Sales Forecasting Methods, Models, and Techniques: A Practical Breakdown for B2B Teams

Disclaimer: This article was created in collaboration with Airspeed, who sponsored this piece. Airspeed is an AI revenue execution platform using AI agents across GTM workflows. I’ve explored and implemented the platform directly, and personally recommend both the team and technology. All insights remain independent and focused on practical value for the enablement and revenue community.

Quick answer

Sales forecasting uses different forms of evidence to estimate how much an organisation is likely to sell within a defined period. Some approaches rely on human judgement, some use historical performance or conversion rates, some assess the opportunities currently in the pipeline, and others apply statistical or probabilistic models.

The main approaches covered in this guide fall into four broad families:

• Judgement-based forecasting uses the experience and assessment of executives, managers, or sellers. It is particularly useful when historical data is limited, the organisation is entering a new market, or individual opportunities depend on context that a standard calculation may not capture. Its main weakness is that judgement can be inconsistent, optimistic, or influenced by internal pressure.
• Historical and conversion-based forecasting uses previous historical sales data, current run rate, lead volumes, and historical conversion rates to estimate future performance. It works best in stable, repeatable sales environments where past behaviour remains relevant, but becomes less dependable when the product, market trends, customer mix, or sales motion changes materially.
• Pipeline- and opportunity-based forecasting builds the estimate from the active deals in the CRM, using evidence such as stage, value, probability, deal age, time in stage, and expected close date. It is well suited to deal-driven B2B sales, although its reliability depends heavily on accurate opportunity data, consistent stage definitions, credible close dates, and realistic probability assumptions.
• Statistical and probabilistic forecasting uses formal models to identify historical patterns, estimate relationships between variables, calculate deal-level probabilities, or represent a range of possible outcomes for future revenue. This family includes time-series models, regression, multivariable models, and Monte Carlo simulation. These approaches can provide more structured and scalable estimates, but they require sufficient data, careful validation, and ongoing monitoring.

No single family is universally best, and 100% accurate sales forecasting is nigh impossible. The right choice depends on what the organisation is forecasting, the sales motion, forecast horizon, available data, deal volume, sales-cycle length, and level of uncertainty involved. In practice, many B2B organisations compare or combine several forecast views rather than relying on one estimate alone.

A method describes the broad forecasting logic, while a model is the calculation, probability structure, or algorithm used to produce the estimate. Techniques support the collection, review, adjustment, and communication of the forecast, while the methodology and process define how the wider forecasting system is designed and operated.

If you want the full breakdown, the rest of this guide goes deeper.

How to choose the right B2B sales forecasting approach – comparison table

The best approach depends on the question you need to answer, the evidence available, and the nature of the sales motion. The table below provides a useful starting point.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Several approaches may fit the same organisation because they answer different questions. A historical baseline might show what normally happens, a pipeline forecast might show what the current deals could produce, and seller judgement might add account context that neither calculation captures.

Where several views are useful, the organisation can compare or combine them within a wider forecasting methodology.

Introduction

I started looking more closely at sales forecasting while researching the best sales forecasting software available to B2B teams.

At first, the task looked fairly straightforward: compare the platforms, understand their capabilities, and see how they help organisations produce a more reliable and accurate forecast.

But the deeper I went, the clearer it became that the sales forecasting software sits on top of a more fundamental question: how is the forecast being produced in the first place?

Sales forecasting is often discussed as though it were one standard calculation. In reality, organisations forecast in very different ways.

Some start with historical performance. Others build the number upwards from the opportunities currently in the pipeline. Some rely heavily on the judgement of sellers and managers, while others use statistical models to identify patterns across multiple variables.

These approaches won’t always produce the same answer because they use different evidence and assumptions. A method that projects recent sales performance forward is answering a different question from one that evaluates the likelihood of individual opportunities closing or estimates a range of possible outcomes.

The way terminology is also used in this area creates a lot of confusion among teams, organisations, and even seasoned sales leaders.

Terms such as method, model, technique, and methodology are often used interchangeably, even though they can refer to different parts of the forecasting system. Forecast categories such as commit and best case are also sometimes presented as forecasting methods, when they are better understood as ways of collecting, organising, and reviewing a forecast.

This guide focuses specifically on B2B sales forecasting: how commercial teams estimate what they are likely to sell within a defined period.

It will touch on pipeline and revenue forecasting where those distinctions matter, but it won’t attempt to cover demand forecasting, inventory planning, or the wider corporate financial forecasting process.

We’ll cover the main B2B sales forecasting methods, models, and techniques, how each one works, where it performs well, and how to choose the right approach for your sales environment.

Sales forecasting methods, models, techniques, methodologies, and processes: what is the difference?

Before comparing the individual approaches, we need a shared language.

Forecasting literature, professional bodies, and sales platforms don’t use these terms consistently. In some sources, method and model mean almost the same thing. Some sales platforms even use model to describe how a forecast is configured and displayed rather than the mathematical logic behind it.

The distinctions I’ll be presenting below, based on my research, should therefore be read as practical working definitions for this guide, not as universally accepted rules.

Sales forecasting method

A sales forecasting method is the broad way an organisation estimates future sales.

It establishes where the forecast starts and which type of evidence carries the most weight. This is consistent with the way established forecasting literature discusses qualitative and quantitative methods according to the available data and the revenue projections that can reasonably be made about the future. [1]

Historical forecasting, for example, starts with past sales data. Opportunity-stage forecasting starts with the position of active deals in the sales process. Sales force composite forecasting starts with the assessments submitted by individual sellers.

Other examples include run-rate forecasting, lead- or funnel-based forecasting, and length-of-sales-cycle forecasting.

Sales forecasting model

A sales forecasting model is the calculation, probability logic, algorithm, or statistical structure used to turn the available inputs into an estimate.

This definition aligns closely with The Institute of Business Forecasting and Planning (IBF), which describes a forecasting model as a quantitative method, algorithm, or equation used to prepare a forecast. [2] Forecasting literature also distinguishes between model families that use past values of the forecast variable, models that use external predictors, and approaches that combine both. [1]

Examples include:

• stage-weighted probability;
• regression;
• time-series analysis;
• multivariable probability scoring;
• and probabilistic simulation.

A model will often sit inside a broader method.

An opportunity-based forecasting method, for example, might use fixed probabilities assigned to each sales stage. A more mature organisation might replace those fixed percentages with probabilities calculated from historical conversion data or a multivariable predictive model.

The broad method remains opportunity-based, but the model used to calculate the forecast has changed.

Sales forecasting technique

A sales forecasting technique is a specific practice used to collect, refine, inspect, reconcile, or communicate a forecast.

This operating layer may include:

• rep forecast submissions;
• manager overrides;
• scenario weighting;
• forecast notes;
• comparisons between forecast snapshots;
• and structured forecast reviews.

Commit, best case, upside, and pipeline are better described as forecast categories within this operating layer. Salesforce defines them as categories linked to the position of an opportunity in the sales cycle and uses them to organise forecast roll-ups. [3]

They help teams organise and communicate degrees of confidence, but they don’t provide an independent estimation logic in the same way as historical forecasting, regression, or weighted pipeline forecasting.

Rep submissions and manager adjustments serve a similar operational purpose. A submission captures a seller’s or manager’s estimate at a particular point in time, while an adjustment changes the forecast roll-up without changing the underlying opportunity data. [4][5]

The same applies to a forecast call. It may be an important part of the forecasting process, but the meeting itself isn’t a forecasting method.

Sales forecasting methodology

A sales forecasting methodology is the organisation’s complete forecasting system.

It brings together the methods and models used to produce the forecast with the rules, roles, data, categories, and governance that make the approach consistent across the organisation.

The methodology is closely related to the sales forecasting process, but the two terms aren’t quite identical. The methodology defines the overall system and principles. The process is the recurring sequence of activities through which that system is operated – from collecting inputs and submitting forecasts to reviewing, adjusting, communicating, and evaluating them.

The Institute of Business Forecasting and Planning (IBF) uses the term forecasting process for this broader organisational layer. It describes it as the complete set of decisions governing the information required, where it comes from, who participates, which forecasting approach is followed, and how performance is evaluated. [6]

In a B2B sales organisation, the methodology defines:

• what the organisation is forecasting and over what period;
• which methods and models it uses;
• the data, stage definitions, and probabilities behind the forecast;
• what each forecast category means;
• how sellers, managers, RevOps, finance, and leadership contribute;
• when forecasts are submitted, inspected, adjusted, and communicated;
• and how accuracy, bias, slippage, and changes between submissions are reviewed.

A company can choose a perfectly accurate sales forecasting method and still operate a weak methodology.

The underlying calculation may be sound, but the forecast will remain unreliable if CRM data is poor, stages are applied inconsistently, close dates aren’t maintained, managers use different evidence standards, or overrides aren’t documented and evaluated.

A forecasting method produces an estimate. A forecasting methodology defines the wider system that makes that estimate consistent, inspectable, and useful. The forecasting process is how that system operates in a sequence of steps.

The boundaries between these terms aren’t absolute. In practice, sources frequently use them interchangeably, and some forecasting approaches sit across more than one category.

Throughout this guide, I’ll use forecasting approach as the broader umbrella term. I’ll then use method, model, technique, methodology, and process consistently according to the working definitions above.

The main families of B2B sales forecasting approaches

The forecasting approaches covered in this guide can be organised into four broad families.

These families are intended to make a complex subject easier to navigate. They shouldn’t be read as five perfectly equivalent or universally accepted technical classifications. Some mainly contain recognised forecasting methods. Others bring together types of models, ways of representing probability and uncertainty, or approaches that combine several different forecast views.

The lack of perfect symmetry reflects the reality of forecasting rather than a flaw in the structure. Forecasting literature itself uses several overlapping classification systems, and more recent research cautions against treating any one taxonomy as definitive. [1][8]

1. Judgement-based forecasting

Judgement-based forecasting relies primarily on human assessment, experience, domain knowledge, and contextual information.

Rather than starting with a fixed statistical relationship, the forecast is formed through the judgement of executives, managers, sellers, or other subject-matter experts. This can be particularly valuable when historical evidence is limited, the organisation is entering a new market, or current conditions differ materially from anything seen before. [1]

The methods covered in this family are:

• executive judgement forecasting;
• and sales force composite forecasting.

The distinction between them is mainly where the judgement originates. Executive judgement usually develops the estimate from the top down, while sales force composite forecasting builds it from the individual assessments of sellers and then rolls those estimates upwards through the organisation.

2. Historical and conversion-based forecasting

Historical and conversion-based forecasting estimates future sales from patterns the organisation has already observed.

The starting point may be performance during an earlier comparable period, the current pace of sales, or the rates at which leads and opportunities have historically progressed from one stage of the funnel to the next.

The methods covered in this family are:

• historical forecasting;
• run-rate forecasting;
• and lead- or funnel-based forecasting.

These methods are generally straightforward to calculate and explain. Their central assumption is that previous performance or conversion behaviour remains sufficiently relevant to help estimate what is likely to happen next.

That makes them useful in stable and repeatable environments, but less dependable when the business model, market, product mix, lead quality, or sales motion has changed materially.

3. Pipeline- and opportunity-based forecasting

Pipeline- and opportunity-based forecasting builds the estimate from the deals currently being pursued.

Instead of beginning with aggregate sales history, it looks at evidence connected to individual opportunities, such as:

• deal value;
• sales stage;
• estimated close probability;
• opportunity age;
• time spent in the current stage;
• and expected close timing.

The methods covered in this family are:

• opportunity-stage forecasting;
• weighted pipeline forecasting;
• and length-of-sales-cycle forecasting.

These approaches are closely related, but they don’t answer exactly the same question.

Opportunity-stage forecasting uses a deal’s position in the sales process to assign or infer how likely it is to close. Weighted pipeline forecasting turns that probability into an expected sales contribution by applying it to the value of the opportunity. Length-of-sales-cycle forecasting adds a timing perspective by considering how quickly or slowly the deal is progressing compared with the organisation’s typical sales cycle.

In practice, the three are often used together rather than as completely separate alternatives.

4. Statistical and probabilistic forecasting

Statistical and probabilistic forecasting uses formal models to identify patterns, estimate relationships, assess the likelihood of individual deals closing, or represent a range of possible outcomes.

Unlike the previous families, this group mainly contains types of models and modelling approaches, rather than only broad business methods.

It includes:

• time-series models;
• regression models;
• multivariable forecasting models;
• and Monte Carlo simulation.

Time-series models use patterns observed in sales data over time, such as trends, seasonality, and recurring fluctuations.

Regression models estimate how sales are related to one or more selected variables that may influence the result, such as price, headcount, marketing activity, pipeline coverage, or economic conditions.

Multivariable models combine several signals at the same time to estimate an outcome or probability. For example, an opportunity model might consider deal size, sales stage, opportunity age, engagement activity, and previous conversion patterns when estimating whether a deal is likely to close.

Monte Carlo simulation repeatedly models different combinations of uncertain inputs to produce a distribution, or range, of plausible results rather than one fixed estimate.

It is also worth clarifying what probabilistic means here. It isn’t one specific type of model. It refers to forecasts that make uncertainty visible through outputs such as:

• close probabilities that reflect how often similar deals actually close;
• prediction intervals showing a range within which the final result is expected to fall;
• scenario ranges representing different possible outcomes;
• or complete forecast distributions showing both the possible results and their relative likelihood.

A time-series model, an opportunity model, and a simulation can all produce probabilistic outputs, even though their underlying mechanics are different. [8]

5. Hybrid forecasting

Hybrid forecasting combines several methods, models, or sources of human judgement.

It is therefore not one method with a standard calculation. It describes how organisations compare or combine different forecast views to build a more robust estimate.

A B2B organisation might, for example, compare:

• a bottom-up forecast submitted by sellers;
• a stage-weighted pipeline forecast;
• a historical conversion baseline;
• a statistical model;
• and a leadership adjustment based on current market or account context.

Research into forecast combination has repeatedly found that different forecasts can contain different useful information and that combining them can improve robustness and, in many circumstances, accuracy. [7]

This doesn’t mean every output should be averaged mechanically. The value of a hybrid approach comes from understanding why the forecasts disagree, what evidence each one captures, and which assumptions deserve the most weight in the circumstances.

The cross-cutting operating layer

Alongside these five families sits a separate operating layer.

This includes forecast categories such as pipeline, best case, and commit, as well as rep submissions, hierarchy roll-ups, manager adjustments, forecast notes, forecast calls, and comparisons between forecast snapshots. [3][4][5]

These elements can have a major influence on the quality of the forecast, but they aren’t additional forecasting methods.

They help the organisation:

• collect judgement from sellers and managers;
• organise opportunities into different confidence categories;
• consolidate individual forecasts through the management hierarchy;
• inspect and adjust the overall estimate;
• document the reasoning behind changes;
• compare how the forecast evolves from one point in time to another;
• and communicate the resulting forecast.

Keeping this layer separate prevents forecast-management practices from being confused with the underlying logic used to produce the estimate.

These five families provide a practical way to organise the methods, model families, and combination approaches examined in the following sections. While they offer a clear structure for understanding the main forecasting approaches, the boundaries between them are naturally fluid, and some approaches span more than one family.

The purpose of this overall framework is to help you understand how the principal approaches differ, relate to one another, and can be applied in practice.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

1. Judgement-based forecasting

Judgement-based forecasting relies heavily on human interpretation.

The forecast is developed through the knowledge and assessment of executives, managers, sellers, or other people who understand the business, market, or opportunities being forecast.

Using a judgement approach is especially useful when:

• historical data is limited,
• the organisation is forecasting something new,
• market or business conditions have changed materially,
• or individual opportunities have characteristics that historical averages and aggregate models can’t fully capture.

A model can identify patterns in the data it receives. A person may also know that a new competitor has entered the market, a major customer has changed its buying process, a regulation is about to affect demand, or a strategically important deal is more fragile than its CRM data suggests.

Judgement allows the forecast to incorporate information that may be relevant but difficult to represent in a standard calculation. Research therefore treats human judgement as an important part of forecasting, while also recognising that it can introduce inconsistency and bias.[9]

The quality of that judgement directly affects its usefulness.

An unsupported opinion from a senior leader is still an unsupported opinion. Judgement becomes more dependable when the organisation gives people a clear way to:

• identify the evidence behind the estimate,
• state the assumptions being made,
• compare the situation with relevant past examples,
• consider alternative outcomes,
• document why the forecast has changed,
• and review previous judgements against what eventually happened.

This turns intuition into a more structured forecasting process.

Executive judgement forecasting

Executive judgement forecasting uses the experience, market knowledge, and commercial assessment of senior leaders to estimate future sales.

It doesn’t have one standard formula. Leaders develop the forecast by bringing together the information available, agreeing on the assumptions that matter most, and translating those assumptions into an expected result.

Depending on the situation, that evidence may include:

• historical performance from a similar product, market, or territory,
• results from comparable organisations or market segments,
• market size and expected growth,
• customer and partner intelligence,
• competitor activity,
• planned pricing and positioning,
• sales headcount and capacity,
• expected seller ramp time,
• pipeline creation assumptions,
• average deal value,
• buying behaviour,
• economic or regulatory conditions,
• and leadership’s understanding of strategic accounts.

The forecast may begin with a historical analogy. For example, leaders launching a product in France might examine what happened when the organisation previously entered Germany or Spain.

Of course, an analogy like this needs to be examined rather than accepted at face value. The team should ask where the markets are genuinely comparable, where they differ, and how those differences affect the forecast.

Research into structured analogies has found that experts tend to produce better forecasts when they deliberately identify relevant comparisons, assess how similar they are, and examine their outcomes, rather than relying on whichever past example first comes to mind. [10]

How leaders develop the estimate

Executive judgement usually combines evidence with a set of commercial assumptions.

For a new territory, leaders might estimate sales by considering:

• how many sellers will be hired,
• when they will start,
• how quickly they will become productive,
• how much pipeline each seller can generate,
• expected conversion rates,
• average contract value,
• available partner support,
• and the likely time required to establish the brand.

Those assumptions can then be converted into a forecast.

The calculation may be simple, but the difficult part is deciding whether the inputs are realistic.

Executive judgement forecasting example

Imagine that a B2B software company plans to enter a new European market.

It has no direct sales history in the country, so leaders use performance from previous market entries, interviews with local partners, existing customer demand, and sales capacity assumptions.

They develop the following base case:

• five account executives,
• $250,000 in annual bookings per fully productive seller,
• and average first-year productivity of 70% while the team ramps.

The initial forecast would be: Five sellers × $250,000 × 70% = $875,000

That figure is the result of the assumptions, rather than an independently proven outcome.

Leaders might therefore develop three scenarios:

• Downside scenario: Hiring takes longer, sellers ramp slowly, and partner-sourced pipeline is weaker than expected.
• Base scenario: Hiring and ramp follow the expected plan, producing $875,000 in bookings.
• Upside scenario: The company signs an early anchor customer, partner contribution is stronger, and sellers ramp faster.

Scenario construction makes the uncertainty visible, and it also gives the organisation a clearer view of which conditions would need to be true for each outcome to occur.

This is particularly important when forecasting new products. With little or no historical data available, research recommends using an assumption-led approach that shows how different assumptions produce a range of possible outcomes. [11]

When executive judgement forecasting is useful

Executive judgement is best suited to situations where historical evidence is weak but informed commercial knowledge is available.

Typical applications include:

• new businesses,
• new products,
• new territories,
• unfamiliar customer segments,
• major changes in pricing or go-to-market strategy,
• periods of market disruption,
• and longer-range strategic forecasting.

It can also add value when a small number of large events could materially change the result.

For example, a company entering a regulated market may know that the timing of one approval, partnership, or major contract will determine a large part of the year’s revenue. A historical average may struggle to represent that situation.

Executive judgement is less useful as the only forecasting method in a stable, high-volume environment with extensive historical data. In those conditions, quantitative methods can provide a stronger starting point, with judgement used to interpret or adjust the result when relevant information isn’t reflected in the model.

Strengths of executive judgement forecasting

Executive judgement can:

• incorporate new information before it appears in historical data,
• account for strategic decisions and market changes,
• draw on experience from comparable situations,
• support forecasting when little direct evidence exists,
• capture factors that are difficult to quantify,
• and help the organisation explore several plausible scenarios.

It can also force leaders to discuss what they believe will drive the result.

When those assumptions are explicit, the forecast becomes more useful than a number alone. The organisation can track whether hiring, ramp, demand, pricing, conversion, and other critical assumptions are developing as expected.

Limitations and risks

The flexibility of executive judgement is also its main weakness.

Different leaders may interpret the same evidence differently. Their estimates may be influenced by commercial targets, internal expectations, personal experience, or the outcome they would prefer to see.

Common risks include:

• Optimism: Leaders may overestimate demand, seller productivity, conversion, or the speed at which a new initiative will gain traction.
• Anchoring: The first number discussed, such as the annual target or an earlier forecast, may influence later estimates too heavily.
• Overconfidence: Leaders may underestimate the range of possible outcomes and place too much trust in their ability to predict what will happen.
• Political pressure: The forecast may be shaped by the number the board, investors, or senior management expects rather than by the most likely commercial outcome.
• Reliance on seniority: Teams may accept an estimate because it comes from the most senior person in the room, even when another participant has stronger evidence.
• Selective analogies: Leaders may choose a previous launch or market entry that supports their preferred conclusion while overlooking less successful comparisons.

Research into judgemental forecasting has documented optimism, anchoring, and overreaction as recurring sources of forecasting bias. [12]

To reduce the impact of these biases, organisations need to make the reasoning behind the forecast explicit and open to challenge.

More disciplined organisations require leaders to explain significant adjustments, record the assumptions behind them, and review whether those interventions improved the forecast once actual results are available. [13]

Information required

Executive judgement forecasting works best when leaders have access to:

• a clearly defined forecast period and outcome,
• relevant market and customer intelligence,
• historical results from genuinely comparable situations,
• product, pricing, and go-to-market plans,
• sales capacity and ramp assumptions,
• expected pipeline and conversion performance,
• known strategic opportunities and risks,
• alternative scenarios,
• and a written record of the assumptions behind the estimate.

The organisation should also preserve previous forecasts and compare them with actual results to create a continuous learning cycle. Over time, leaders can identify which assumptions proved reliable, where judgement improved the forecast, and where confidence repeatedly exceeded the available evidence.

Sales force composite forecasting

Sales force composite forecasting develops a bottom-up estimate from the individual forecasts submitted by sellers.

Each seller estimates what they expect to sell within a defined period across their territory, accounts, or active opportunities. Those estimates are then combined at team, segment, regional, or business-unit level. [14]

Sales managers review the submissions, challenge the supporting evidence, make adjustments where appropriate, and roll the resulting forecast upwards through the organisation.

In its simplest form:

The actual process is usually more involved because the organisation needs to decide what sellers should submit, which evidence should support their estimates, and how managers should review and adjust them.

How sales force composite forecasting works

The process often follows this sequence:

1. Each seller reviews their territory, accounts, and active opportunities.
2. The seller submits an estimate for what they expect to close during the forecast period.
3. A manager reviews the estimate and the opportunity-level evidence behind it.
4. Seller estimates are consolidated into a team forecast.
5. Team forecasts are rolled up by segment, region, or business unit.
6. Sales leadership reviews the combined forecast and applies any justified adjustments.
7. The submitted forecast is compared with later forecast snapshots and actual results.

Modern CRM systems support this workflow through forecast submissions, hierarchy roll-ups, adjustments, notes, and point-in-time snapshots. [4][5]

Those workflow elements help collect and inspect seller judgement, representing part of the cross-cutting operating layer rather than becoming separate forecasting methods.

What sellers use to develop their estimates

The method works because sellers may know things about individual accounts that a central model can’t easily see.

That information may include:

• strength of the customer’s business problem,
• access to the economic buyer,
• support from an internal champion,
• quality of stakeholder engagement,
• procurement and legal progress,
• confirmed budget,
• competitive position,
• customer urgency,
• agreed next steps,
• political dynamics inside the account,
• and whether the stated close date remains credible.

A seller may therefore judge two opportunities in the same CRM stage very differently.

Both deals might appear to have the same value and close probability in the system. One has executive sponsorship, an agreed decision process, and procurement already involved. The other has gone quiet for three weeks and still depends on an unconfirmed budget.

The aggregate data may make them look similar. The seller’s account knowledge suggests otherwise.

Sales force composite forecasting example

Imagine that five account executives submit the following quarterly forecasts:

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

The initial bottom-up team forecast is therefore $500,000.

During the manager review, Seller C explains that $40,000 of their forecast depends on a contract being approved before the buyer’s financial year-end.

The manager knows that legal review has only started and believes $20,000 of that amount is more likely to move into the following quarter.

The manager adjusts the team forecast from $500,000 to $480,000, records the reason, and submits the revised figure to regional leadership.

The underlying seller submissions still matter, showing how the total was built and where the manager’s view differs.

Where sales force composite forecasting works well

Sales force composite forecasting is particularly useful in deal-driven B2B environments where sellers manage a defined group of accounts or opportunities.

It tends to fit:

• mid-market and enterprise sales,
• territory-based sales teams,
• account-based sales motions,
• complex opportunities involving several stakeholders,
• renewals and expansions where account owners understand customer health,
• and businesses where a limited number of deals can materially affect the period.

Its value declines in extremely high-volume, transactional environments where individual seller insight adds little beyond the aggregate conversion data.

Strengths of sales force composite forecasting

Sales force composite forecasting can:

• use information held by the people closest to the buyer,
• reflect differences between individual opportunities,
• create accountability for seller submissions,
• provide a detailed bottom-up view of expected sales,
• reveal where seller and manager judgement differ,
• and help leaders identify the deals and assumptions driving the overall forecast.

It can also improve forecast conversations.

Rather than debating only whether the total feels right, managers can examine the accounts, evidence, risks, and dependencies behind it.

Where distortion can enter

A bottom-up forecast is only as reliable as the judgement and incentives behind the individual submissions.

Distortion can enter in several ways:

• Seller optimism – A seller may overestimate the strength of the opportunity, accept the buyer’s stated timing too readily, or underestimate the work still required to close the deal.
• Sandbagging – A seller may deliberately submit a conservative estimate so they can reduce scrutiny, protect themselves from disappointment, or improve the chance of outperforming the forecast later.
• Strategic under-commitment – Even without deliberate sandbagging, sellers may exclude credible opportunities because they believe that committing to them will create additional pressure or management attention.
• Inconsistent evidence standards – One seller may require confirmed budget, buyer access, and a clear decision process before including a deal. Another may include opportunities based mainly on positive conversations. The resulting submissions may look comparable even though they reflect very different standards.
• Leadership pressure – Managers may push sellers towards the number required to hit the target, which can turn the submission into a negotiation rather than an honest estimate of the likely outcome.
• Uneven manager calibration – Some managers challenge every deal aggressively. Others accept seller judgement with little inspection. Some habitually reduce submissions, while others add optimistic adjustments. This creates inconsistency as forecasts move through the hierarchy.

Making sales force composite forecasting more dependable

The method becomes more reliable when the organisation defines what evidence should support a seller’s forecast by giving sellers and managers a shared standard for deciding whether an opportunity belongs in the forecast.

A more disciplined process should include:

• clear submission definitions,
• consistent opportunity evidence,
• documented reasons for material adjustments,
• manager review focused on deal evidence rather than pressure,
• comparison of seller and manager accuracy over time,
• calibration across teams and regions,
• and regular review of submitted forecasts against actual outcomes.

Managers should also distinguish between a target and a forecast.

The target represents what the organisation wants or needs to achieve. The forecast represents the most credible current estimate of what is likely to happen.

When those two numbers differ, changing the forecast doesn’t solve the underlying performance gap.

A method and an operating workflow

Sales force composite forecasting has two connected roles.

As a recognised forecasting method, it creates a bottom-up estimate by combining the judgements of individual sellers.

As an operating workflow, it depends on submissions, manager review, organisational roll-ups, notes, and justified adjustments.

Keeping those roles distinct prevents the surrounding practices from being presented as additional forecasting methods.

Pipeline, best case, and commit are forecast categories. Rep submissions collect judgement. Manager adjustments layer another assessment onto the forecast. Forecast calls provide a forum for reviewing the evidence.

They all shape how sales force composite forecasting is managed, but the underlying method remains the same: individual seller estimates are combined to create an organisational forecast.

2. Historical and conversion-based forecasting

Historical and conversion-based forecasting estimates future sales from patterns the business has already observed.

Rather than asking leaders or sellers to assess what they think will happen, these approaches start with measurable evidence such as:

• sales from previous periods,
• the current pace of performance,
• lead volumes,
• conversion rates,
• average deal values,
• and the time normally required to move through the funnel.

They are usually easy to understand, fast to calculate, and useful as benchmarks against which other forecasts can be tested.

This doesn’t make them unsophisticated.

Forecasting research routinely uses simple and seasonal baseline methods because they provide transparent points of comparison. [1] Research comparing simple and complex forecasting methods also shows that greater complexity does not automatically improve forecast accuracy. A well-matched simple approach may perform as well as, or better than, a more complex model. [15]

The real question is whether the historical pattern remains relevant.

A forecast based on last year’s sales may work well if the product, customer mix, pricing, sales capacity, and market have remained broadly consistent. It becomes much less dependable when those conditions change.

The same principle applies to conversion rates. A historical lead-to-customer rate only remains useful if the organisation continues to attract similar leads through a comparable funnel.

Within this family, three practical approaches are particularly relevant to B2B sales forecasting:

1. Historical forecasting uses sales from a previous comparable period as its starting point, sometimes applying an expected rate of growth or decline.
2. Run-rate forecasting extends the current pace of sales across the remainder of the forecast period.
3. Lead- or funnel-based forecasting estimates future opportunities or sales from lead volumes, conversion rates, average deal values, and the time normally required to move between stages.

Each approach draws on a different type of observed pattern: historical forecasting looks at previous results, run rate looks at current pace, and funnel forecasting looks at how volume converts through the sales process.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Historical forecasting

Historical forecasting uses sales from an earlier comparable period as the baseline for estimating future performance.

In its simplest form, the organisation assumes that the next period will resemble the previous one.

A business that generated $800,000 last quarter might therefore begin with $800,000 as its forecast for the next quarter.

Most organisations add some form of adjustment. They may expect sales to increase because the team has added capacity, prices have risen, or demand is growing. Alternatively, they may forecast a decline because of customer churn, reduced headcount, weaker market conditions, or the loss of a major channel.

A practical formula is:

If the expected change is negative, the same formula applies.

Choosing the right historical baseline

The most important word in historical forecasting is comparable.

The previous calendar period isn’t always the most relevant reference point. The right comparison depends on the rhythm of the business.

Common baselines include:

• the previous month,
• the previous quarter,
• the same month or quarter in the previous year,
• the average of several comparable periods,
• or the equivalent seasonal period from an earlier year.

A period-on-period comparison uses the immediately preceding period.

For example, a team may use Q2 sales as the baseline for Q3. This works best when performance is relatively even throughout the year, and there is no major seasonal effect.

A year-on-year comparison uses the same period from the previous year.

For example, the team may use Q3 last year as the baseline for Q3 this year. This is usually more useful when demand follows a recurring annual pattern.

Imagine that an organisation consistently generates more sales in the final quarter because customers use the remainder of their annual budgets. Comparing Q4 with Q3 would ignore that seasonal uplift. Comparing Q4 with Q4 from the previous year provides a more relevant starting point.

Forecasting literature describes this logic through seasonal baseline methods, which use the most recent observation from the equivalent season rather than the immediately preceding period. [1]

Historical forecasting example

Imagine that a mature B2B software company generated $900,000 in new bookings during Q3 last year.

The business expects Q3 performance to increase by 8% this year because:

• the sales team has slightly more capacity,
• the company has introduced a modest price increase,
• and demand has remained stable.

The forecast would be: $900,000 × (1 + 8%) = $972,000

The calculation is straightforward, yet the more difficult question is whether the 8% adjustment is credible. The team should be able to explain what is expected to produce that increase.

For example, if seller capacity is expected to increase bookings volume by around 5% and average pricing rises by 3%, the combined effect would be approximately 8.15%, assuming the two effects compound and other conditions remain stable.

In practice, the relationship may not be that precise. A 5% increase in seller capacity does not necessarily translate into a 5% increase in bookings, so the forecast should be supported by evidence from productivity, pipeline, conversion rates, ramp time, and demand.

If the 8% was selected only because it matches the annual target, the forecast is much weaker.

When historical forecasting works well

Historical forecasting works best when the business contains enough continuity for earlier results to remain informative.

Typical conditions include:

• a mature and repeatable sales motion,
• a stable product portfolio,
• a broadly consistent customer mix,
• predictable sales capacity,
• limited changes to pricing,
• recurring or repeatable demand,
• and a market that isn’t changing dramatically.

It can be particularly useful for contracted recurring revenue, renewals, established products, and higher-volume sales motions where individual deals don’t dominate the result.

It also provides a helpful sense check against other forecasting methods.

A pipeline forecast may suggest that the team will grow bookings by 35% next quarter, while the historical baseline shows growth remaining between 5% and 10% across the previous eight quarters.

That gap should prompt leaders to examine what has changed, which opportunities are expected to drive the increase, and whether the evidence supports such a large departure from recent performance.

Accounting for seasonality

Seasonality occurs when sales follow a recurring pattern linked to the time of year, budget cycles, customer behaviour, industry events, or operating calendars.

A business may experience:

• stronger sales at the end of each quarter,
• lower activity during the summer,
• increased renewals at the start of the year,
• higher demand around annual planning periods,
• or concentration around customer budget deadlines.

In those environments, using the immediately previous period can produce a misleading comparison.

A seasonal business should compare like with like. December may need to be compared with previous Decembers rather than with November. Q4 may need to be compared with earlier fourth quarters rather than with Q3.

The organisation can also examine several years of seasonal performance instead of relying on one comparison. This reduces the risk that one unusual year becomes the entire basis for the forecast.

One-off events and unusual periods

Historical results sometimes contain events that aren’t likely to repeat.

These may include:

• one unusually large contract,
• a temporary pricing promotion,
• a major partner referral,
• a short-term market disruption,
• a delayed deal that happened to land in the period,
• or a customer expansion that won’t recur.

Suppose last year’s $900,000 Q3 result included a one-off $250,000 contract.

Using the full $900,000 as the baseline assumes that the company will replace that contract before producing any underlying growth. That may be possible, but it needs to be recognised explicitly.

The organisation might instead separate the recurring or repeatable portion of the result:

Adjusted baseline = $900,000 – $250,000 = $650,000

It can then forecast the repeatable business and assess the likelihood of replacing the exceptional deal separately.

This produces a more honest baseline than treating every dollar of historical performance as equally repeatable.

Where historical forecasting becomes unreliable

Historical forecasting assumes some degree of continuity, which is something that weakens when the business or its environment changes materially.

Common sources of disruption include:

• a major pricing change,
• the launch or removal of a product,
• entry into a new market,
• a change in customer segment,
• a new sales channel,
• substantial changes in sales capacity,
• a redesigned sales process,
• an acquisition,
• regulatory change,
• or a significant economic or competitive shift.

These events can create a structural break – a point at which the relationships seen in the historical data no longer represent the business in the same way.

For example, a company may have grown steadily while selling a $10,000 product to small businesses. If it moves upmarket and begins selling $100,000 contracts through a longer enterprise sales cycle, its previous volume, conversion, and timing patterns may no longer provide a credible basis for the forecast.

Historical data hasn’t become useless, but it simply needs to be interpreted in light of the change.

In those situations, the organisation may use historical forecasting as one benchmark while relying more heavily on judgement, pipeline evidence, scenarios, or a model that reflects the new operating conditions.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Run-rate forecasting

Run-rate forecasting projects the current pace of sales through the remainder of a defined period.

It focuses on the question: if sales continue at their current rate, where will the period finish?

A common formula is:

The same calculation can also be expressed as:

The unit of time might be a day, week, month, or quarter, depending on what the organisation is forecasting.

Monthly run-rate example

Imagine that a business has generated $240,000 after 12 of the month’s 20 working days.

Average sales per working day are: $240,000 ÷ 12 = $20,000

Extending that pace across the full month gives: $20,000 × 20 = $400,000

The monthly run-rate forecast is therefore $400,000.

This calculation assumes that the remaining eight working days will perform at roughly the same pace as the first 12.

Quarterly run-rate example

Imagine that a sales team has generated $600,000 halfway through the quarter.

The elapsed proportion is 50%, so the forecast would be: $600,000 ÷ 50% = $1.2 million

The same result can be reached by doubling the sales generated during the first half of the quarter.

The calculation is clear. Whether the answer is useful depends on how evenly sales normally arrive throughout the quarter.

The assumptions behind run rate

Run-rate forecasting makes a straight-line assumption, which is that the pace observed so far provides a reasonable guide to the pace that will follow.

That assumption is more credible when:

• sales volume is reasonably high,
• individual transactions are relatively small,
• performance is spread evenly through the period,
• the sales process is short,
• and no major change is expected during the remaining time.

A subscription business acquiring a steady volume of low-value customers may produce a meaningful weekly or monthly run rate.

A complex enterprise sales team closing six large deals per quarter probably won’t.

When run-rate forecasting is useful

Run rate is particularly useful for:

• short-term forecasting,
• high-volume or transactional environments,
• relatively even sales patterns,
• newly created teams with limited operating history,
• recently launched commercial motions,
• and rapid reasonableness checks.

It can give leaders an immediate baseline before more detailed pipeline, funnel, or modelled analysis becomes available.

For example, a new product-led growth motion may have only three months of history. The organisation doesn’t yet have enough data to establish a dependable annual pattern, but its recent acquisition pace can still provide an early directional estimate.

Run rate can also expose large gaps between the current pace and the target.

If a team is halfway through the quarter with $400,000 closed against a $1.2 million target, its straight-line run rate is $800,000. Leadership can immediately see that the current pace leaves a $400,000 gap.

The run rate doesn’t explain whether the team can close that gap, yet it makes the gap visible.

When run rate can mislead

Run rate becomes unreliable when sales don’t arrive evenly through time.

This is common in B2B sales.

Many teams generate a disproportionate amount of revenue near the end of the month or quarter. Procurement deadlines, manager pressure, discount approvals, and customer budget cycles can all create late-period concentration.

A team that historically closes 60% of its quarterly business in the final month will look behind pace during the first two months. A straight-line projection would systematically understate the likely result.

The opposite can also happen.

A team may begin the quarter with an unusually large renewal or enterprise contract. Extending that early pace across the whole quarter can create a forecast the remaining pipeline has little chance of supporting.

Run rate can therefore mislead in:

• seasonal businesses,
• enterprise and long-cycle sales,
• highly concentrated portfolios,
• end-of-quarter-heavy sales motions,
• periods following product launches,
• and periods influenced by one unusually large contract.

Imagine that a team closes a $500,000 deal during the first week of a quarter in which normal quarterly bookings are $1 million.

A weekly straight-line projection would produce an enormous forecast. The arithmetic would be correct, but the underlying assumption would be indefensible because that deal isn’t representative of the normal weekly pace.

Adjusting the run rate

A more useful run-rate analysis may account for the way performance normally develops through the period.

Suppose historical evidence shows that a team usually generates:

• 20% of quarterly bookings in the first month,
• 30% in the second month,
• and 50% in the final month.

If the team has generated $240,000 by the end of month one, a seasonally adjusted projection would be: $240,000 ÷ 20% = $1.2 million

A simple straight-line projection based on one of three months would produce: $240,000 × three = $720,000

The difference is substantial.

The adjusted estimate may still be wrong, but it reflects the organisation’s normal intra-quarter pattern rather than assuming that every month contributes equally.

A baseline, not a guaranteed outcome

Run rate is most useful when treated as a baseline or sense check.

It tells the organisation what will happen if the current pace continues. It doesn’t prove that the pace will continue, explain what will drive the remaining sales, or account automatically for changes in mix, seasonality, or deal concentration.

A dependable forecast should therefore compare run rate with other evidence, such as:

• historical period shape,
• current pipeline data,
• expected renewals,
• funnel conversion,
• sales capacity,
• and known one-off events.

The value of run rate is its speed and transparency. Its weakness is the amount of business context it leaves out.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Lead- or funnel-based forecasting

Lead- or funnel-based forecasting estimates future opportunities or sales from the volume entering the commercial funnel and the proportion that historically converts through each stage. [16]

Instead of beginning with previous revenue or the current sales pace, it begins further upstream.

A simplified formula is:

In practice, the calculation may include several conversion points.

For example:

A more detailed model can use every meaningful transition in the funnel.

How funnel-based forecasting works

Imagine that an organisation expects to generate 1,000 leads next quarter.

Historically:

• 10% of leads become qualified opportunities,
• 25% of qualified opportunities become customers,
• and the average deal value is $8,000.

The forecast would be: 1,000 × 10% × 25% × $8,000 = $200,000

The model implies:

• 1,000 leads,
• 100 qualified opportunities,
• 25 customers,
• and $200,000 in sales.

This gives the organisation more than an end number. It shows the operating assumptions required to produce that result.

If lead volume drops, the forecast changes. If conversion improves, the forecast changes. If average deal value falls, the forecast changes.

That makes funnel forecasting useful for diagnosing the drivers behind the estimate.

Defining the funnel

The quality of the forecast depends on how clearly the funnel is defined.

A typical B2B funnel might include:

• lead,
• marketing-qualified lead,
• sales-qualified lead,
• qualified opportunity,
• proposal,
• closed-won customer.

The exact labels matter less than the definitions behind them.

If one team treats every webinar attendee as a marketing-qualified lead while another requires a defined level of intent and fit, their conversion rates aren’t comparable.

The organisation should define:

• what qualifies an entry into each stage,
• which event marks conversion,
• how duplicate or recycled leads are handled,
• when a lead leaves the funnel,
• and which time period or cohort the rate represents.

Without those rules, the forecast may look mathematically consistent while combining inconsistent behaviours.

Lead sources and conversion differences

Leads from different sources rarely behave in the same way.

An inbound demo request may convert differently from:

• a paid social lead,
• a content download,
• an event attendee,
• a partner referral,
• a product sign-up,
• or an outbound prospecting response.

Suppose an organisation generates 1,000 leads:

• 400 high-intent inbound leads convert to opportunities at 20%,
• while 600 content leads convert at 4%.

Applying one blended 10.4% conversion rate may reproduce the historical total, but it can hide a major change in source mix.

If the next quarter contains more content leads and fewer demo requests, the blended historical rate will overstate the likely number of opportunities.

A more dependable forecast calculates expected conversion separately by source:

Forecasted opportunities = leads from each source × the relevant source conversion rate

The results can then be combined.

This preserves differences in lead quality instead of averaging them away. [17]

Stage-to-stage conversion

A funnel can also hide where conversion is changing.

Two organisations may both convert 2.5% of leads into customers, but the reasons can be completely different.

One may convert many leads into opportunities but struggle to close them. Another may qualify leads aggressively and produce fewer opportunities, but close a higher proportion of those that pass qualification.

Stage-to-stage analysis helps identify the difference.

For example:

• 1,000 leads,
• 20% become sales-qualified leads,
• 50% of those become opportunities,
• 25% of opportunities close,
• and average deal value is $8,000.

The forecast is:

1,000 × 20% × 50% × 25% × $8,000 = $200,000

If opportunity-to-win conversion falls from 25% to 15%, the forecast drops to $120,000 even if lead volume remains unchanged.

That is why volume alone isn’t enough; the organisation needs to understand how that volume moves through the funnel.

Average deal value and product mix

Average deal value is another source of hidden variation.

A business may have an overall average contract value of $12,000, but that figure could combine:

• a $4,000 small-business product,
• a $15,000 mid-market product,
• and a $60,000 enterprise package.

If the expected product mix shifts towards smaller contracts, applying the historical blended average will overstate revenue.

The same issue applies to customer segments, regions, pricing plans, new business, renewals, and expansions.

Where those groups behave differently, the forecast should calculate them separately and then combine the results.

Time lags through the funnel

Conversion rates describe how much business moves through the funnel.

They don’t automatically tell you when that business will become revenue.

Suppose the organisation expects 1,000 leads next quarter and the historical calculation implies $200,000 in eventual sales. If the average journey from lead creation to closed-won is 120 days, much of that revenue may not arrive during the same quarter.

A useful funnel forecast therefore needs to consider the time required to move:

• from lead to qualification,
• from qualification to opportunity,
• from opportunity to proposal,
• and from proposal to close.

The forecast horizon should match the maturity of the leads being analysed.

Leads generated this month may support a forecast for the following quarter rather than the current one. Existing opportunities that originated in earlier periods may provide most of the current-quarter result.

Ignoring this lag can make demand-generation activity look more immediately productive than it is.

Cohort behaviour

A cohort is a group of leads that entered the funnel during the same period or under similar conditions.

For example, the organisation might track all leads created in January and measure how many became opportunities or customers after 30, 60, 90, and 120 days.

This gives a more accurate view than calculating conversion from all leads and all closed deals in the same calendar month.

The leads created in March aren’t necessarily the leads that became customers in March. Combining the two can distort the relationship between input and output.

Cohort analysis helps the organisation understand:

• how conversion develops over time,
• when a cohort is mature enough to assess,
• whether recent cohorts are performing differently,
• and how long new lead volume will take to affect sales.

It also prevents the team from judging recent cohorts too early.

A lead cohort may appear to convert poorly after 30 days even though the normal sales cycle is 90 days. The issue may be timing rather than quality.

Where funnel-based forecasting works best

Lead- or funnel-based forecasting works best when the organisation has a measurable, repeatable acquisition motion.

Typical applications include:

• inbound sales,
• product-led growth,
• small and medium-sized business sales,
• high-volume acquisition,
• demand-generation-led motions,
• and repeatable mid-market funnels with enough volume to stabilise conversion rates.

It is particularly useful when marketing activity has a clear and measurable relationship with pipeline and revenue creation.

The method is less suited to low-volume enterprise sales where a small number of accounts, long buying cycles, and deal-specific circumstances dominate the result.

In those environments, generating ten additional leads may tell the organisation very little about in-period revenue. The identity, quality, buying situation, and progression of each opportunity matter far more than aggregate lead volume.

When blended conversion rates mislead

A blended conversion rate combines different populations into one average.

That can be useful as a headline benchmark, but it becomes misleading when the underlying groups behave differently.

Common causes include:

• different lead channels,
• different customer segments,
• changes in lead quality,
• shifts in product mix,
• regional differences,
• inconsistent stage definitions,
• and substantial variation in conversion time.

Imagine that enterprise leads convert at 5% with an average deal value of $80,000, while small-business leads convert at 20% with an average deal value of $5,000.

An overall average conversion rate and deal value can produce a forecast that represents neither group properly.

The problem becomes more serious when the mix changes.

A forecast calibrated during a quarter dominated by high-intent partner referrals may fail when the next quarter relies heavily on lower-intent paid leads.

The historical rate is mathematically correct, but it is no longer representative.

Making funnel-based forecasting more dependable

A stronger funnel forecast should:

• define each stage consistently,
• segment conversion by source, product, market, and customer type where relevant,
• use mature cohorts,
• account for the normal time between stages,
• monitor changes in lead quality,
• separate new business, renewals, and expansion where their behaviour differs,
• and compare forecasted conversion with actual outcomes.

The organisation should also review whether conversion assumptions remain stable.

A rate based on the previous 24 months may contain plenty of data, but it may react too slowly to a recent shift in demand or lead quality. A rate based on the past four weeks may be more current, but also much noisier.

There is no universally correct window.

The organisation needs enough history for the rate to be credible, while keeping it recent enough to represent the current funnel.

Such a trade-off is central to all historical and conversion-based forecasting.

These methods create useful estimates when the past provides a relevant guide to the future. When the business changes, the forecast needs to change with it.

3. Pipeline- and opportunity-based forecasting

Pipeline- and opportunity-based forecasting builds the forecast primarily from active sales opportunities.

Instead of starting with previous sales, current run rate, or top-of-funnel volume, these approaches focus on what the existing pipeline is likely to produce during the forecast period.

As a result, they are particularly relevant to B2B sales teams where named opportunities, deal values, sales stages, and expected close dates are tracked in a CRM system.

The basic logic is that since the organisation already has a set of potential deals, if it can assess how likely each deal is to close, when it is likely to close, and how much it is worth, it can turn the open pipeline into a sales forecast.

The calculation may be straightforward; however, the inputs rarely are.

Pipeline-based forecasts depend heavily on clear and consistently applied sales stages, reliable opportunity values, credible close dates, accurate and timely CRM updates, historically meaningful probability assumptions, and enough opportunities for probability-based estimates to become useful at portfolio level.

When those conditions are present, pipeline forecasting provides a structured way to connect current selling activity with expected results.

When they aren’t, the forecast may look precise while resting on weak evidence.

Three closely related approaches are particularly relevant in this category:

• Opportunity-stage forecasting estimates the likelihood of a deal closing from its current position in the sales process.
• Weighted pipeline forecasting applies a probability to each opportunity’s value and combines those expected contributions into a forecast.
• Length-of-sales-cycle forecasting uses deal age, time in stage, and progression against normal sales-cycle patterns to assess likelihood or timing.

These approaches are often used as layers within the same forecasting process. Opportunity-stage forecasting provides an initial estimate of likelihood. Weighted pipeline converts that probability into expected sales value. Length-of-sales-cycle forecasting then tests whether the deal’s progression and expected timing support that estimate.

Opportunity-stage forecasting

Opportunity-stage forecasting estimates the likelihood of closing an opportunity from the stage it has reached in the sales process.

A typical B2B sales process might include stages such as: qualification → discovery → solution validation → proposal → commercial negotiation, → and contract review.

The underlying assumption is that opportunities become more likely to close as they progress through those stages.

An opportunity in early discovery may carry a 15% probability. A deal in proposal may carry 50%, while a contract under final review may carry 80%.

CRM systems often connect stages with default probabilities and forecast categories. When an opportunity moves to a new stage, its probability and contribution to the forecast may change accordingly. [18]

How stage probabilities are determined

There are several ways to assign a probability to each stage.

Fixed stage probabilities

The organisation assigns a standard percentage to every stage.

For example:

• Qualification: 10%
• Discovery: 20%
• Solution validation: 40%
• Proposal: 60%
• Negotiation: 80%
• Contract review: 90%

This is easy to implement and explain, and it also creates a consistent starting point across sellers and teams.

The weakness is that the percentages may be little more than assumptions.

A company may assign 60% to Proposal because the number feels appropriate, not because 60% of comparable opportunities have historically closed.

Historically calibrated probabilities

A stronger approach calculates the percentage of opportunities that have historically closed from each stage.

Suppose the organisation finds that:

• 18% of opportunities reaching Discovery eventually close,
• 42% of opportunities reaching Solution validation close,
• 63% of opportunities reaching Proposal close,
• and 81% of opportunities reaching Negotiation close.

Those figures can replace arbitrary percentages with evidence from the organisation’s own pipeline.

Historical calibration still needs care. The results may differ substantially by customer segment, product, region, deal size, lead source, new business versus expansion, and seller population.

An enterprise opportunity in Proposal may behave differently from a small-business opportunity in the same nominal stage. Applying one percentage to both may hide more than it reveals.

Manual probability assignment

Some organisations allow sellers or managers to change the probability manually.

This can add useful context.

A seller may know that a deal in Negotiation has lost executive sponsorship, entered a formal procurement delay, or encountered an unexpected competitor. The standard stage probability may no longer reflect the real position.

That said, manual probabilities can also introduce inconsistency. One seller may use 70% to mean “the buyer appears positive”, while another reserves 70% for a deal with agreed commercials, confirmed budget, and a documented approval path.

The organisation therefore needs clear evidence standards behind any manual adjustment.

Opportunity-stage forecasting example

Imagine that a team has five active opportunities:

Opportunity-stage forecasting indicates that Opportunities D and E are more likely to close than Opportunities A and B.

At this point, the method has assigned or inferred likelihood. It hasn’t yet translated that likelihood into forecasted sales value.

That happens when the organisation applies the probability to the value of each opportunity, which moves the calculation into weighted pipeline forecasting.

Eventual win probability and in-period close probability

One distinction is far more important than it seems.

The probability that an opportunity will eventually close isn’t necessarily the same as the probability that it will close during the current forecast period.

Consider a $300,000 opportunity in final negotiation.

The team may believe there is an 80% chance of eventually winning the deal. However, if procurement normally takes another six weeks and the quarter ends in ten days, the probability of closing it this quarter may be much lower.

Both assessments can be valid:

• 80% probability of eventually closing,
• 30% probability of closing this quarter.

A sales forecast needs the second probability when estimating in-period results.

This is where many pipeline forecasts become inflated because they treat a healthy opportunity as though it must close within the date currently recorded in the CRM.

Predictive forecasting systems increasingly distinguish manual confidence from the likelihood of revenue landing within the forecast period, using historical close rates and current pipeline progression to assess timing risk. [19]

Stage entry and exit criteria

Stage forecasting becomes more dependable when each stage represents observable progress.

A stage name alone isn’t enough.

The organisation should define:

• what must be true before an opportunity enters the stage,
• which buyer actions or evidence are required,
• what work the seller must complete,
• what prevents the deal from progressing,
• and what must happen before it can leave the stage.

For example, Proposal shouldn’t simply mean that the seller has sent a document.

A stronger definition may require:

• confirmed business requirements,
• an agreed solution scope,
• access to the relevant decision-makers,
• validated budget,
• and buyer agreement to review a formal proposal.

These criteria create a shared standard, and they also make historical stage-conversion rates more meaningful because opportunities placed in the same stage have met comparable conditions.

Stage discipline and stage inflation

Stage inflation occurs when an opportunity moves forward without enough evidence.

A seller may advance a deal because:

• a forecast review is approaching,
• the pipeline looks too light,
• leadership expects more late-stage coverage,
• the buyer expressed general interest,
• or the seller is optimistic about what will happen next.

The opportunity now appears stronger in the CRM. Its probability increases, and it may enter a more confident forecast category.

However, nothing material has changed with the buyer.

This distorts both the current forecast and the historical data used to calibrate future probabilities. If weak opportunities routinely enter Proposal, the organisation’s Proposal conversion rate will decline. The business may then respond by lowering the stage probability, even though inconsistent stage use caused the problem.

Stage discipline is therefore part of the forecasting method, not a separate administrative concern.

Why deals in the same stage can behave differently

A stage provides a useful summary, but it remains a coarse measure of opportunity health.

Two opportunities may both sit in Negotiation while carrying very different levels of risk.

The first deal may have:

• a confirmed decision date,
• an active executive sponsor,
• agreed commercial terms,
• legal review in progress,
• and a clear next meeting involving both sides.

The second may have:

• an unresponsive contact,
• no access to the economic buyer,
• an unconfirmed budget,
• a close date moved forward twice,
• and no buyer activity during the previous three weeks.

The stage is the same. The underlying evidence isn’t.

That is why stage probabilities often need to be segmented, challenged, or strengthened with other signals such as deal age, stakeholder engagement, qualification evidence, and close-date movement.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Weighted pipeline forecasting

The previous method focused on assigning or inferring a probability for each opportunity.

Weighted pipeline forecasting takes the next step. It applies that probability to the value of the deal to estimate its expected contribution to the overall forecast.

The standard calculation is:

For each opportunity, the organisation multiplies the value of the deal by its estimated probability of closing. It then adds those weighted amounts together.

Oracle NetSuite describes weighted forecasts using this same logic: the probability is multiplied by the projected amount of each transaction. Salesforce similarly calculates expected revenue from opportunity amount and probability. [18][20]

This creates a useful distinction between the two methods: opportunity-stage forecasting assigns or infers probability, while weighted pipeline forecasting converts that probability into expected sales value.

In practice, the two often operate as parts of the same calculation. An organisation may assign a probability according to opportunity stage and immediately apply it to the deal value.

For example, a $100,000 opportunity in a stage carrying a 60% probability contributes $60,000 to the weighted forecast.

The probability doesn’t always have to come from the opportunity stage. As discussed above, organisations may also derive it from:

• historical conversion data,
• seller or manager judgement,
• forecast-category rules,
• or a predictive model using several opportunity signals.

The calculation remains the same. What changes is how much confidence the organisation can place in the probability behind it.

A 60% probability calibrated from several years of relevant outcomes carries a different evidential weight from a seller entering 60% because the deal feels positive. Both produce the same weighted value, but the reliability of that value may be very different.

Weighted pipeline example

Imagine that a team has five open opportunities:

The weighted forecast is:

$70,000 + $48,000 + $24,000 + $15,000 + $8,000 = $165,000

The team has $330,000 in unweighted pipeline, but its expected-value forecast is $165,000.

That figure acknowledges that the organisation is unlikely to win every open deal.

Understanding expected value

Weighted pipeline uses portfolio logic.

Imagine a $1 million opportunity with a 50% probability of closing.

Its expected contribution to the forecast is: $1 million × 50% = $500,000

The organisation won’t receive half of the contract.

The deal will most likely produce one of two outcomes:

• approximately $1 million if it closes,
• or $0 if it doesn’t.

The $500,000 represents the average expected contribution across many comparable situations.

Suppose an organisation repeatedly carried 100 independent $1 million opportunities, each with a well-calibrated 50% probability. It would expect to win approximately half of them, producing around $50 million.

The individual opportunities don’t close fractionally. The portfolio result can still approach the expected value.

As you can imagine, this logic becomes weaker when the portfolio contains only a few large deals.

If the quarter depends on two $1 million opportunities, both weighted at 50%, the forecast may show $1 million. The actual result could just as probably be $0, $1 million, or $2 million.

The weighted number is mathematically valid, but it may communicate more stability than the pipeline actually contains. Just take a look at that big fat zero from the previous paragraph.

Probability calibration

A probability is calibrated when its assigned likelihood broadly matches actual outcomes over time.

If the organisation labels a large group of comparable opportunities as 70%, approximately 70% of them should eventually close within the defined forecast horizon.

If only 35% close, the probability is too optimistic.

If 90% close, the estimate may be too conservative.

The organisation should therefore compare forecast probabilities with actual results and examine calibration by stage, segment, product, region, deal-size band, seller, and forecast horizon.

This requires enough closed opportunities to produce a meaningful pattern. Predictive opportunity-scoring systems impose minimum historical data requirements for the same reason: a probability model needs enough won and lost outcomes from which to learn. [19]

False precision

A weighted pipeline can create a highly specific number from highly uncertain inputs.

A forecast of $1,743,250 looks precise.

It may have been calculated from:

• stale opportunity amounts,
• stages that haven’t been updated,
• close dates selected to keep deals in the quarter,
• default probabilities that were never calibrated,
• and opportunities with little recent buyer engagement.

The arithmetic is correct, but the result still isn’t dependable.

This is the main danger of weighted pipeline forecasting. It converts weak assumptions into a number that looks objective.

The organisation should therefore inspect the inputs before accepting the output.

Stale opportunities and inaccurate stages

An opportunity may remain open long after its commercial momentum has disappeared.

Common examples include deals where:

• the buyer has stopped responding,
• the next step is missing,
• the opportunity hasn’t moved for several months,
• the seller is reluctant to close it as lost,
• or the stage still reflects a conversation that happened weeks ago.

If these opportunities remain in the pipeline, they continue to contribute weighted value.

A $200,000 proposal-stage opportunity at 60% contributes $120,000 to the forecast, even if the buyer hasn’t responded for six weeks.

The forecast doesn’t know that unless the probability, stage, or supporting data changes.

Weak close-date discipline

A pipeline forecast usually includes opportunities according to their expected close dates.

For this reason, we can argue that the close date is one of the most consequential fields in the CRM as well as one of the easiest to misuse.

Sellers may select the last day of the quarter by default, move deals forward one period at a time, keep an expired date in place, or choose a date that reflects the target rather than the buyer’s process.

Repeated close-date movement is valuable evidence. It suggests that the organisation has previously misunderstood the deal’s timing.

A forecast should therefore distinguish between:

• a close date supported by a mutual decision process,
• an internally selected target date,
• and a date that has already slipped several times.

Treating all three as equivalent weakens the forecast.

Concentrated enterprise pipelines

Weighted pipeline works most naturally when the organisation carries enough opportunities for gains and losses to balance across the portfolio.

This often makes it useful in repeatable small and medium-sized business (SMB) and mid-market motions.

Enterprise pipelines behave differently.

A small number of large opportunities may determine the entire quarter. Each deal may involve a different buying group, procurement process, legal review, security assessment, and political context.

In those environments, the weighted forecast can still provide a useful baseline, but it shouldn’t replace deal-level inspection, seller and manager judgement, or scenario analysis.

A $5 million weighted forecast built from 50 similar opportunities means something different from a $5 million weighted forecast built from three enterprise contracts.

Length-of-sales-cycle forecasting

Length-of-sales-cycle forecasting uses timing and progression to estimate whether an opportunity is likely to close and when that close may occur.

It can draw on:

• total deal age,
• average sales-cycle length,
• time spent in the current stage,
• the speed of stage progression,
• and comparison with similar historical opportunities.

This approach helps correct a major weakness in pure stage forecasting.

Stage tells the organisation where the opportunity appears to be. It doesn’t show how the opportunity arrived there, how long it has remained there, or whether its progress matches normal buying behaviour.

Average sales-cycle length

The average sales cycle measures the time opportunities normally take to move from a defined starting point to closed-won.

That starting point might be opportunity creation, initial qualification, first sales meeting, or another consistently recorded event.

Suppose comparable mid-market opportunities take an average of 90 days to close.

A deal that has been open for 70 days and has progressed steadily into commercial negotiation may have a credible chance of closing within the next month.

A deal created seven days ago and manually placed into the same stage requires more scrutiny.

The stage may be legitimate; however, the deal may also have skipped normal qualification, stakeholder validation, technical review, or procurement preparation.

Deal age and time in stage

Total deal age shows how long the opportunity has been open.

Time in stage shows how long it has remained at its current point in the process.

Both matter.

A 120-day-old opportunity may be normal in an enterprise motion with a nine-month sales cycle. The same age may signal severe stagnation in an SMB process where deals normally close within 30 days.

Time in stage can reveal more specific friction.

An opportunity may have progressed normally through qualification and discovery before spending 60 days in Proposal, where comparable won deals usually spend 15.

That gap should prompt questions like:

• Is the buyer still engaged?
• Has the decision process changed?
• Is legal or procurement blocking progress?
• Has the seller lost access to the right stakeholders?
• Does the close date still reflect the buyer’s timeline?

Time alone doesn’t answer those questions, yet it tells the organisation where to inspect.

Microsoft’s predictive opportunity scoring, for example, compares how long an open opportunity has remained in a stage with the time previously won opportunities spent there. [19]

Stage velocity

Stage velocity describes how quickly opportunities move through the sales process.

A team can examine:

• the average time from Qualification to Discovery,
• the average time from Discovery to Proposal,
• the time from Proposal to Negotiation,
• and the time from Negotiation to Closed won.

This helps the organisation understand more than the total length of the sales cycle.

Two deals may both close in 90 days while following different paths. One may spend most of its time in discovery. Another may progress quickly before slowing down in legal review.

Those patterns are critical when forecasting current opportunities.

A deal moving through stages at a pace consistent with historical wins may deserve more confidence. A deal moving far more slowly may require a lower probability or a later expected close date.

Length-of-sales-cycle example

Imagine that similar opportunities historically take 90 days to close.

Two $150,000 opportunities are currently in Negotiation.

Opportunity A

• Created 75 days ago
• Progressed through each stage in the normal sequence
• Entered Negotiation ten days ago
• Buyer activity remains consistent
• Commercial review is scheduled
• Close date is 20 days away

Opportunity B

• Created eight days ago
• Entered Negotiation almost immediately
• Has limited qualification evidence
• No economic buyer has been identified
• Close date is also 20 days away

A stage-based forecast may assign both opportunities the same probability.

Instead, the sales-cycle evidence suggests they shouldn’t automatically receive the same treatment.

Opportunity A has followed a progression broadly consistent with previously won deals. Opportunity B may be a genuinely fast-moving opportunity, but its compressed path could also reflect an inflated stage or missing evidence.

Length-of-sales-cycle forecasting gives the organisation a reason to examine the difference.

Expected close timing

Sales-cycle analysis can also help determine which forecast period should contain the opportunity.

Suppose a $250,000 deal has reached Proposal after 45 days.

Historically, comparable deals take:

• 30 days to move from Proposal to Negotiation,
• and another 40 days to complete negotiation, procurement, and legal review.

The opportunity may be healthy, but a close date three weeks away is difficult to defend.

Moving it into the following quarter may improve the current forecast even though the organisation’s confidence in eventually winning the deal hasn’t changed.

This brings us back to the earlier distinction:

• eventual win likelihood,
• and in-period close likelihood.

Sales-cycle forecasting is particularly useful for testing the second.

Stalled opportunities and close-date movement

An opportunity can remain active without progressing.

Signs of a stalled deal include:

• excessive time in the current stage,
• no recent buyer activity,
• repeated rescheduling,
• unclear next steps,
• unchanged qualification gaps,
• and multiple close-date movements.

A long sales cycle doesn’t automatically mean the opportunity is weak.

Large buying groups, legal reviews, procurement rules, and security requirements often extend enterprise deals for legitimate reasons.

The organisation should distinguish explained delay from unexplained inactivity.

A deal held in legal review for a documented contractual issue is different from a deal that hasn’t moved because the buyer has gone quiet.

Where sales-cycle forecasting works best

This method works best when:

• the sales process is reasonably repeatable,
• stage progression is recorded consistently,
• similar deals follow broadly comparable timelines,
• the organisation has enough historical outcomes,
• and time-to-close carries useful information about opportunity health.

It can be particularly effective in SMB and mid-market environments where cycle lengths are relatively stable.

It also works well as a refinement to stage-based forecasting. Stage provides the initial likelihood, while age and velocity test whether the opportunity’s timing supports that assessment.

Limitations in complex sales

Sales-cycle forecasting becomes less dependable when buying journeys vary significantly.

This is common in enterprise sales involving:

• complex procurement,
• extensive legal review,
• security and compliance assessments,
• several decision-makers,
• international contracting,
• customised commercial terms,
• and irregular budgeting processes.

A large strategic deal may remain open for a year and still be healthy.

Another may progress rapidly through the CRM while lacking the buyer commitment required to close.

Average cycle length can provide context, but it shouldn’t become a rigid deadline. In complex sales, timing signals need to be interpreted alongside the commercial circumstances of the individual opportunity.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Additional evidence that strengthens pipeline forecasts

The three methods already use several important opportunity variables.

Opportunity-stage forecasting uses the deal’s current position in the sales process to estimate likelihood. Weighted pipeline combines that probability with the opportunity value. Length-of-sales-cycle forecasting examines deal age, time in stage, stage velocity, and expected close timing.

However, those variables rarely tell the whole story.

A stronger pipeline forecast also considers evidence that shows whether the buyer is genuinely progressing towards a decision. These factors don’t constitute separate forecasting methods. Instead, they help sellers and managers challenge, strengthen, or adjust the output produced by the core methods.

Useful evidence includes:

• activity recency and buyer engagement,
• stakeholder coverage,
• qualification evidence,
• next-step quality,
• close-date changes,
• pricing movement,
• and seller forecasting history.

Activity recency and buyer engagement

Recent activity helps show whether the opportunity remains commercially active.

Such activity may come from buyer responses, scheduled meetings, proposal reviews, document exchanges, new stakeholders’ participation, and progress against agreed actions.

Seller activity alone isn’t enough.

Ten unanswered emails don’t represent stronger buyer engagement than one productive meeting with the decision team.

The forecast should focus on whether the buyer is investing time, involving the right people, and completing the actions required to move towards a decision.

Stakeholder coverage

Complex B2B deals rarely depend on one contact.

The forecast should consider whether the seller has engaged the people involved in areas like business approval, financial approval, technical validation, procurement, legal review, or implementation.

A late-stage opportunity with only one engaged contact may carry more risk than its stage suggests.

The same applies when a seller has a strong relationship with an operational contact but no access to the economic buyer, procurement team, or other stakeholders who can delay or stop the decision.

Stakeholder coverage doesn’t prove that a deal will close, yet it helps show whether the opportunity has developed beyond an isolated conversation with one supportive contact.

Qualification evidence

Qualification evidence tests whether the conditions required for a sale are genuinely present.

Depending on the organisation’s sales methodology or qualification framework, this may include:

• a confirmed business problem,
• measurable impact,
• a recognised decision process,
• access to authority,
• available budget,
• agreed selection criteria,
• and a compelling reason to act within the forecast period.

In this regard, I’d like to stress that the purpose shouldn’t be to award points for filling in CRM fields; it should be to determine whether the recorded opportunity reflects a real buying process.

A seller may have completed every required field while relying on assumptions rather than buyer-confirmed evidence. In that case, the CRM record may look complete without making the forecast more dependable.

Next-step quality

A strong next step is specific, mutual, and connected to progress.

“Follow up next week” is weak.

“Commercial review with the chief financial officer and procurement on September 18” provides more evidence.

The forecast should consider whether:

• the opportunity has a documented next action,
• the buyer has agreed to it,
• the right stakeholders will participate,
• and completing it will move the decision forward.

The absence of a clear next step often indicates that the seller is waiting rather than actively progressing the deal.

A scheduled activity isn’t automatically meaningful either. Another introductory call may add little if the opportunity actually needs commercial approval, technical validation, or a procurement decision.

Close-date changes

Length-of-sales-cycle forecasting tests whether the current close date is realistic. Close-date history adds another useful signal: how often that expectation has already changed.

A moved close date doesn’t automatically make a deal weak since the buyer may have a credible reason for delaying the decision, such as an extended legal review, a change in the approval process, or a budget meeting that moved.

Repeated movement without corresponding buyer progress is far more concerning.

The organisation should track:

• how many times the date has moved,
• how far it has moved,
• whether the opportunity progressed at the same time,
• and whether the buyer supplied a credible reason.

A deal pushed from one quarter-end to the next three times shouldn’t carry the same timing confidence as a deal with a stable, buyer-confirmed decision date.

Pricing movement

Changes to price, discount, or contract value can reveal movement in the buying process.

A justified commercial revision may indicate genuine negotiation. For example, the buyer may have requested a different contract term, product configuration, or implementation scope.

Repeated discounting without stronger buyer commitment may signal that the seller is trying to rescue a weak opportunity.

The organisation should assess:

• why the value changed,
• who requested the change,
• whether the buyer offered anything in return,
• and whether the revision improved the likelihood or timing of the decision.

A lower price doesn’t necessarily make a deal more likely to close. It may simply reduce the value of an opportunity that remains uncertain.

Seller forecasting history

Seller history helps managers interpret manually assigned probabilities and forecast submissions.

Some sellers consistently overestimate the likelihood or timing of their deals. Others forecast conservatively and close more than they submit.

The organisation shouldn’t automatically increase or reduce every forecast according to the seller involved.

However, several historical patterns like close-date accuracy, stage discipline, frequency of late forecast changes, and the difference between submitted forecasts and actual outcomes all provide useful context.

For example, a seller who regularly assigns 80% probabilities to deals that close only 40% of the time may need clearer evidence standards or better calibration.

Another seller may consistently understate likely outcomes to avoid missing a commitment. Their submissions may look accurate, but they may still provide leadership with an incomplete view of likely performance.

Seller history should support better review and coaching rather than become a blanket adjustment applied to every deal.

Using the evidence together

No single signal determines the outcome.

A long-running opportunity may still be healthy. A deal with recent activity may still lack decision authority. A late-stage opportunity may have strong stakeholder engagement but an unrealistic close date.

The value comes from reading the evidence together, which modern sales forecasting tools can increasingly support at scale.

Consider a $400,000 opportunity in Contract review with an assigned probability of 85%.

The stage and probability strengthen the forecast.

However:

• the close date has moved twice,
• no buyer meeting has occurred for 25 days,
• the legal contact hasn’t been identified,
• and the next step remains “awaiting feedback”.

Those signals weaken the in-period forecast, even if the organisation still believes it may eventually win the deal.

The right response may be to reduce the in-period probability, move the expected close date, or place the opportunity into a less confident forecast category.

Clari highlights ageing opportunities, repeated close-date movement, missing next steps, absent buying-group contacts, and incomplete CRM data as observable pipeline warning signals. Gong and Microsoft similarly use progression, activity, time in stage, and historical outcomes to strengthen opportunity assessment. [19][20][21]

Here’s the practical takeaway: the stage tells you where the opportunity currently sits in the CRM, while the wider evidence tells you whether that position is credible, how likely the deal is to close, and whether the expected timing still makes sense.

Bringing those signals together gives you a forecast you can have more confidence in.

4. Statistical and probabilistic forecasting

Statistical and probabilistic forecasting covers several different model families and one important simulation approach.

They all use data to estimate future outcomes, but they shouldn’t be considered as equivalent methods or as steps on a ladder from basic to advanced.

They answer different questions.

For example:

• Time-series forecasting asks what future aggregate sales may look like based on patterns observed over time.
• Regression forecasting asks how sales relate to one or more measurable drivers.
• Multivariable forecasting combines several signals to estimate outcomes such as close probability, expected value, risk, or timing.
• Monte Carlo simulation asks what range of outcomes may occur when several inputs remain uncertain.

Machine learning can support several of these approaches. However, it is more accurate to treat it as a way of implementing and extending statistical, multivariable, and probabilistic models than as a separate universal forecasting method.

There is also another important distinction that cuts across the whole category.

Some models produce a point forecast, such as $4.2 million in expected quarterly bookings. Others express uncertainty through prediction intervals, close probabilities, scenario ranges, or complete distributions of possible outcomes.

The right approach therefore depends on:

• what the organisation is forecasting,
• whether the question concerns aggregate performance or individual opportunities,
• how much relevant data is available,
• how stable the underlying relationships are,
• and how the forecast will be used.

A more sophisticated model isn’t automatically a better model; it all comes down to whether it answers the right business question more accurately and reliably than the alternatives.

Time-series forecasting

Time-series forecasting uses observations recorded over time to estimate future values of the same variable.

For example, an organisation might use:

• monthly bookings from the previous three years,
• quarterly sales from the previous 20 quarters,
• weekly renewal volumes,
• or monthly recurring revenue over time.

The model looks for structure within the sequence itself rather than starting with individual opportunities or a separate set of commercial drivers.

Common forms of structure include: trend, seasonality, recurring patterns, cycles, and autocorrelation.

Trend

A trend is a sustained upward or downward movement over time.

A software company may have increased quarterly bookings gradually as it added sellers, expanded into new markets, and improved brand awareness.

Alternatively, sales may have declined over several periods because of customer churn, competitive pressure, or reduced demand.

A time-series model can estimate that underlying direction and project it forward, provided the conditions producing the trend remain relevant.

Seasonality

Seasonality describes a pattern that repeats at a relatively predictable interval.

A business may consistently generate:

• more bookings in the final quarter,
• fewer opportunities during the summer,
• higher renewal volumes at the start of the year,
• or a recurring increase near customer budget deadlines.

The important point is repetition.

If fourth-quarter sales rise every year, the model can treat that uplift differently from random quarter-to-quarter variation.

Recurring patterns and cycles

Not every repeated movement follows a fixed calendar.

Sales may rise and fall with economic conditions, product replacement cycles, investment cycles, hiring patterns, or broader changes in customer confidence.

These movements can last longer and occur less regularly than seasonal patterns.

A recurring December uplift is seasonal. A multi-year rise and decline linked to an industry investment cycle is cyclical.

The difference is important because a seasonal pattern may repeat on a relatively stable schedule, while a cycle can vary in duration and intensity.

Autocorrelation

Autocorrelation describes the relationship between a value and earlier values in the same series.

In practical terms, recent performance may provide information about what happens next.

If a strong month tends to be followed by another strong month, or if quarterly bookings show a recurring relationship with the equivalent quarter one year earlier, the series contains information that a model may be able to use.

Autocorrelation doesn’t explain why the relationship exists. It identifies whether previous observations help predict later ones.

Forecasting literature commonly uses time-series models to capture combinations of trend, seasonality, and relationships between current and previous observations. [1]

Common time-series models

Time-series forecasting is a broad model family.

Several common approaches are worth understanding at a high level.

Moving averages

A moving average calculates an average from a defined number of recent periods.

For example, a three-month moving average may use bookings from the previous three months to estimate the next month.

This smooths some of the short-term noise in the data and makes the underlying pattern easier to see.

The trade-off is that a simple moving average may react slowly when conditions change. A sudden increase or decline gets absorbed gradually because several older periods remain in the calculation.

Exponential smoothing

Exponential smoothing also uses previous observations but normally gives more weight to recent data and progressively less weight to older data.

More developed versions can account for both trend and seasonality.

This makes exponential smoothing useful when recent performance should influence the forecast more heavily, but the organisation still wants to preserve information from the wider historical series.

Autoregressive integrated moving average models

Autoregressive integrated moving average (ARIMA) models use relationships between present and previous values, changes between periods, and previous forecast errors.

The organisation doesn’t need to understand the full mathematical specification to use an ARIMA-family model effectively.

It does need to understand that the model assumes the historical series contains patterns that remain useful for predicting the future. [1]

Time-series forecasting example

Imagine that a B2B software company has recorded quarterly new-business bookings for the previous five years.

The data shows:

• gradual underlying growth,
• a recurring uplift in the fourth quarter,
• and lower bookings during the third quarter.

A time-series model can estimate the trend, account for the recurring quarterly pattern, and produce a forecast for the next four quarters.

The model might estimate next quarter’s bookings at $2.4 million – the point forecast.

It may also provide a prediction interval, such as:

• lower estimate: $2.0 million,
• point forecast: $2.4 million,
• upper estimate: $2.9 million.

The interval recognises that the model doesn’t know the future with certainty.

What time-series forecasting is best suited to

Time-series models are generally most useful for aggregate questions such as monthly bookings, quarterly sales, renewal volumes, recurring revenue, transaction volumes, or total pipeline creation.

They work best when:

• the organisation has enough consistent historical data,
• the series contains meaningful recurring patterns,
• the definition of the forecast variable has remained stable,
• and past behaviour still provides a reasonable guide to future behaviour.

They are usually less useful for assessing whether one named opportunity will close.

A time-series model may estimate that the business will generate $3 million next quarter. It won’t normally explain whether the Acme contract is likely to sign on March 25.

Structural changes

Time-series forecasting becomes less dependable when the business undergoes a structural change.

Examples include:

• a major pricing change,
• entry into a new market,
• the launch or removal of a product,
• a move from small-business to enterprise sales,
• a substantial increase or reduction in sales capacity,
• a redesigned go-to-market motion,
• or an acquisition.

These changes can break the relationship between the historical series and the business now being forecast.

A company may have five years of monthly sales data, but that history may offer limited guidance if it has replaced a $10,000 transactional product with a $150,000 enterprise solution.

The data is still real, but it isn’t relevant in the same way it had been.

Market shocks

Unexpected external events can create the same problem.

A market shock may involve:

• an economic downturn,
• regulatory change,
• a sudden supply constraint,
• a major competitor entering or leaving the market,
• or an event that changes customer priorities.

A model trained on normal conditions may respond slowly because the shock isn’t represented in the history from which it learnt.

This is one reason statistical forecasts often need contextual review rather than automatic acceptance.

Insufficient history

Time-series models also need enough observations to identify meaningful patterns.

A company with six months of data may be able to calculate a basic average or recent trend. It probably can’t make a dependable claim about annual seasonality.

The required amount of history depends on:

• the forecast frequency,
• the stability of the series,
• the number of recurring patterns,
• and the complexity of the model.

Monthly forecasting usually requires more than a handful of months if the organisation wants to estimate annual seasonality. Quarterly forecasting may need several years before recurring quarter-specific patterns become credible.

Comparing models

The organisation should compare several plausible models rather than assuming that the most complex one will perform best.

For example, it might compare:

• a simple historical baseline,
• a moving average,
• an exponential-smoothing model,
• and an ARIMA-family model.

The comparison should use later periods that were not used to build or tune the models. This provides a more realistic test of how accurately each model predicts genuinely new results.

If a simple seasonal baseline predicts future quarters more accurately than a more elaborate model, the simple model is the better forecasting choice for that situation.

Statistical sophistication only matters when it produces a useful improvement.

Time-series models are strongest for aggregate questions, but they can miss deal-level context, structural breaks, and changes the historical series hasn’t experienced. Forecasting research therefore emphasises model comparison and performance on unseen data rather than relying on model reputation alone. [1]

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Regression forecasting

Regression forecasting estimates the relationship between a forecast outcome and one or more predictors.

The forecast outcome might be:

• quarterly sales,
• new-business bookings,
• renewal revenue,
• average deal value,
• or pipeline generated.

Potential predictors might include:

• pipeline created,
• sales capacity,
• marketing activity,
• pricing,
• discounting,
• seller headcount,
• economic conditions,
• and activity volume.

Time-series forecasting primarily asks what the historical pattern suggests will happen next.

Regression asks a different question: how does the forecast outcome change when relevant business drivers change?

A simple conceptual formula

The logic can be expressed without detailed mathematical notation:

For example:

The model estimates how changes in each predictor are associated with changes in the forecast outcome.

Simple regression

Simple regression uses one predictor.

For example, an organisation may examine the relationship between quarterly pipeline creation and bookings in the following quarter.

It might find that higher pipeline creation is associated with higher future bookings.

The model could then use expected pipeline creation to forecast the likely sales result.

This is easy to understand but may oversimplify the business.

Pipeline creation isn’t the only factor that affects bookings. Conversion quality, seller capacity, pricing, customer demand, and deal timing may also matter.

Multiple regression

Multiple regression includes several predictors at the same time.

A B2B bookings model might use:

• pipeline generated,
• number of fully productive sellers,
• average discount,
• marketing-sourced meetings,
• and an economic-confidence indicator.

This allows the model to estimate the relationship between each predictor and bookings while accounting for the other variables included.

In strict statistical language, multiple regression is already a multivariable model.

For this article, however, it is still useful to discuss regression separately because regression describes a recognised model family, while the later multivariable section covers a broader set of practical B2B models, including deal-level scoring approaches that may use other statistical or machine-learning techniques.

Regression forecasting example

Imagine that an organisation wants to forecast next quarter’s bookings.

Historical analysis suggests that bookings are associated with:

• pipeline created during the previous two quarters,
• the number of fully ramped account executives,
• and the average discount offered.

The model estimates the expected contribution of each factor.

Leadership expects:

• $6 million in relevant pipeline,
• 18 fully productive sellers,
• and an average discount of 12%.

The model combines those inputs and produces a bookings forecast of $2.1 million.

The forecast is based on more than historical sales alone. It reflects the relationships the model has learnt between bookings and the selected predictors.

Predictor availability

A regression forecast can only use a predictor if the organisation knows, or can reasonably estimate, its future value.

This sounds obvious, but it creates an important practical issue.

Suppose the model uses next quarter’s pipeline creation to forecast next quarter’s bookings.

At the time the forecast is produced, next quarter’s final pipeline creation isn’t yet known.

The organisation must therefore:

• forecast pipeline creation separately,
• use a value already available,
• or accept that uncertainty in the predictor will flow into the sales forecast.

The same issue applies to:

• future seller headcount,
• future pricing,
• economic indicators,
• marketing activity,
• and planned discounting.

A regression forecast may look precise while depending on several other forecasts underneath it.

Correlation isn’t causation

Regression identifies relationships in the data, but it doesn’t automatically prove that one variable causes another.

Suppose sales rise during quarters in which the number of customer meetings also rises.

That may mean additional meetings contribute to sales, yet it may also mean:

• stronger demand produces both more meetings and more sales,
• a seasonal pattern drives both,
• sellers schedule more meetings because the pipeline is already healthier,
• or another variable explains the relationship.

A variable can improve prediction without proving a causal mechanism. The organisation should simply avoid interpreting every model relationship as evidence that changing the predictor will necessarily produce the expected result.

Changing relationships

Regression assumes that the relationships learnt from historical data remain sufficiently stable.

That assumption can weaken over time.

For example:

• pipeline may convert differently after a change in qualification,
• additional sellers may contribute less while they are ramping,
• discounting may behave differently after a pricing redesign,
• and marketing meetings may become less predictive when lead quality changes.

A coefficient estimated from two years of historical data may no longer represent the current sales motion.

The organisation should therefore monitor whether the relationship between predictors and outcomes is changing over time.

Omitted variables

A regression model can only use variables that have been included.

If an important factor is missing, the model may misattribute its effect to another predictor or leave more of the outcome unexplained.

For example, a bookings model may include pipeline, headcount, and discounting but omit:

• product availability,
• competitor activity,
• seller experience,
• lead quality,
• or customer budget conditions.

No practical model will include every possible influence.

The question is whether the omitted factors are important enough to distort the relationships on which the forecast depends.

Overfitting

Overfitting occurs when a model intakes historical data indiscriminately, including patterns that were accidental or unlikely to repeat.

A model may explain nearly every movement in the historical dataset and still perform poorly on future periods.

This often happens when:

• the dataset is small,
• the model contains too many predictors,
• variables are selected after repeatedly testing what fits,
• or the model is evaluated on the same data used to build it.

A strong historical fit is therefore not enough.

The organisation needs to test whether the model predicts data it hasn’t previously seen.

Where regression forecasting works best

Regression is most useful when:

• the organisation can identify plausible sales drivers,
• those drivers are measured consistently,
• future predictor values are available or forecastable,
• the relationships are reasonably stable,
• and enough historical observations exist to estimate them.

It can help leadership understand how a forecast may change under different assumptions.

For example:

• What happens if productive seller capacity rises by 10%?
• What happens if average discount increases?
• What happens if pipeline creation falls below plan?

Regression gives the organisation a structured way to connect those assumptions with expected sales.

It remains a model of relationships rather than proof of causation, and its usefulness depends on the quality, stability, and availability of the predictors. [1]

Multivariable forecasting

Multivariable forecasting is best understood as a practical family of models rather than one specific model.

These models combine several types of information to estimate an outcome. In B2B sales, that information may come from the opportunity itself, the seller, buyer behaviour, engagement activity, historical conversion, and the wider commercial environment.

The defining feature is that the model assesses several signals together rather than relying on one stage, one conversion rate, or one historical pattern.

B2B opportunity-modelling research commonly uses this type of approach to estimate deal outcomes and support opportunity prioritisation. [23][24][25]

Why several variables can improve the forecast

The pipeline section showed why two opportunities in the same stage can carry different levels of risk.

A multivariable model formalises that idea.

Consider two $150,000 opportunities in Negotiation.

Opportunity A:

• has been open for 80 days,
• progressed through each stage in the normal sequence,
• includes an active executive sponsor,
• has recent buyer activity,
• has a buyer-confirmed close date,
• and is owned by a seller with strong historical conversion.

Opportunity B:

• was created ten days ago,
• moved rapidly into Negotiation,
• has one engaged contact,
• has no confirmed decision process,
• has already changed close date,
• and shows little recent buyer activity.

A simple stage model may assign both opportunities the same probability.

A multivariable model can treat them differently because it evaluates the surrounding evidence.

It may estimate:

• Opportunity A: 72% probability of closing this quarter,
• Opportunity B: 24% probability of closing this quarter.

The stage remains relevant, but it no longer represents the entire forecast.

Deal-level and aggregate outputs

Many multivariable sales models work at opportunity level.

They estimate an outcome for each deal, such as likelihood of winning, likelihood of closing within the period, expected close date, or risk of slippage.

Those estimates can then be aggregated.

For example:

The organisation can add those expected contributions across the pipeline to produce a model-based forecast.

Other multivariable models operate directly at aggregate level.

A model might estimate quarterly bookings from historical sales, current pipeline, seller capacity, marketing activity, pricing, and economic conditions.

The term therefore describes the use of several variables, not one fixed forecast object.

Data requirements

Multivariable models raise the data requirement.

The organisation needs enough historical examples of:

• won opportunities,
• lost opportunities,
• slipped opportunities,
• close timing,
• stage progression,
• and the other variables used by the model.

The data also needs consistent definitions.

For example:

• Does “executive sponsor identified” mean the same thing across teams?
• Is time in stage measured consistently?
• Are lost opportunities closed promptly?
• Do sellers record buyer activity in the same way?
• Has the sales process changed during the historical period?

A large dataset doesn’t guarantee a reliable model if the underlying fields are inconsistent.

Outcome labels

The model also needs a clearly defined outcome.

“Will the deal close?” is too vague.

The organisation should specify whether the model predicts:

• eventual closed-won,
• closed-won within 30 days,
• closed-won within the current quarter,
• close-date slippage,
• or another defined event.

The distinction mirrors the earlier discussion between eventual win probability and in-period close probability.

A model trained to predict eventual wins shouldn’t be treated as a reliable quarter-end timing model unless it has been designed and tested for that purpose.

Small samples and concentrated sales

B2B environments can create difficult modelling conditions.

A company may have:

• relatively few closed opportunities,
• large variation in deal size,
• different sales processes across segments,
• limited examples of major enterprise wins,
• and frequent changes in product or go-to-market strategy.

A model trained on thousands of similar transactions has more opportunity to learn stable patterns than one trained on 120 highly varied enterprise deals.

This doesn’t make multivariable forecasting impossible in enterprise sales; however, it means the organisation should be more cautious about the precision and generalisability of the output.

Selection bias

Historical opportunity data only includes deals the organisation chose to create and pursue.

That can introduce selection bias.

For example, the model may learn from opportunities that passed the organisation’s qualification process but see little data on potential customers that sellers rejected earlier.

It may also learn from historical resource-allocation decisions. If senior sellers received the strongest opportunities, seller identity and deal quality may become difficult to separate.

The model can still provide value, but the organisation needs to understand what the historical data represents and what it leaves out.

Where multivariable forecasting works best

This approach is most useful when:

• the business has enough historical opportunity data,
• the fields used are recorded consistently,
• the sales process is stable enough for earlier patterns to remain informative,
• the output serves a clearly defined decision,
• and the organisation can monitor performance over time.

It is particularly useful for:

• opportunity scoring,
• pipeline-risk identification,
• seller and manager inspection,
• resource prioritisation,
• and model-based probability estimates.

The model shouldn’t replace commercial judgement automatically; its value may be to highlight where the recorded stage, seller assessment, and broader evidence disagree.

Where machine learning fits

Machine learning supports statistical and multivariable forecasting by allowing models to process more variables and identify more complex patterns.

It can help a model:

• analyse a large number of inputs,
• detect nonlinear relationships,
• identify interactions between variables,
• score opportunities at scale,
• update as additional data becomes available,
• and rank opportunities by predicted likelihood or risk.

For example, the effect of time in stage may depend on the opportunity stage, deal size, customer segment, seller experience, and recent buyer engagement.

A conventional model may require the analyst to specify those relationships in advance.

An AI-powered model may detect useful interactions from the historical data.

This flexibility can improve forecasting in the right environment.

Machine learning as an implementation layer

Many commercial platforms describe their forecasting capabilities as “AI forecasting”, which means the underlying system may use machine learning to:

• estimate opportunity probabilities,
• identify deals at risk,
• predict close timing,
• forecast aggregate sales,
• or produce scenario ranges.

Those are still forms of multivariable, statistical, or probabilistic forecasting.

Machine learning describes how the model learns patterns and generates estimates. It doesn’t define one single forecasting question or output.

For this reason, AI forecasting is better treated as an implementation layer than as a separate universal method. Some of the cited research in this overall section reaches the same conclusion: machine learning extends multivariable forecasting, but it doesn’t form a clean, independent forecasting family.

Data leakage

Data leakage occurs when the model uses information that wouldn’t genuinely have been available at the time the forecast was meant to be produced.

This can make the model appear much more accurate during development than it will be in practice.

Suppose a model predicts whether an opportunity will close this quarter.

The training data includes:

• the final opportunity stage,
• the final discount,
• a close date updated after the contract was signed,
• or activity recorded after the deal outcome was already known.

The model may learn to predict the result from information created by the result.

Its historical performance will look excellent, but it won’t provide the same advantage when scoring a live opportunity before the outcome occurs.

The organisation should therefore make sure every predictor reflects information genuinely available at the forecast point.

Poor data quality

Machine learning doesn’t correct poor data automatically.

If the CRM contains inconsistent stages, missing activity, inflated probabilities, delayed closed-lost updates, unreliable close dates, or different field definitions across teams, the model will learn from those inconsistencies.

It may even make them harder to detect because the output appears mathematically sophisticated.

Data quality remains part of forecast quality.

Overfitting

Flexible models can discover patterns that fit historical data extremely well.

Some of those patterns may be accidental.

For example, a model may learn that opportunities involving a particular product, region, seller, and contract size historically closed at a high rate.

If only a few such opportunities existed, that pattern may not repeat.

The model should therefore be tested on unseen data and compared with simpler alternatives.

A complex model that performs marginally better during development but deteriorates quickly in production may offer less value than a simpler, more stable model.

Model drift

Model drift occurs when the relationships learnt from historical data change.

This may happen after:

• a pricing change,
• a new product launch,
• a revised sales process,
• a change in customer mix,
• a market disruption,
• a major hiring cycle,
• or a change in how CRM fields are recorded.

A model trained on last year’s opportunities may become less accurate as the business changes.

The organisation needs to monitor performance and decide when to recalibrate probabilities, retrain the model, change the inputs, or replace the model.

Updating automatically with new data isn’t enough if the new data reflects inconsistent definitions or a structural break the organisation hasn’t understood.

Explainability

A model may produce a 27% close probability without giving the seller or manager a useful explanation, which, in turn, limits its operational value.

Users need to understand enough about the output to decide what action to take.

Useful explanations might show that the estimate is lower because:

• the deal has remained in stage much longer than comparable wins,
• buyer activity has declined,
• stakeholder coverage is weak,
• the close date has moved repeatedly,
• or the opportunity lacks qualification evidence.

The aim should be to help people understand which evidence influenced the result and how they should inspect the opportunity.

Calibration

A machine-learning model can rank opportunities correctly while producing unreliable probabilities.

For example, it may consistently rank healthier deals above weaker ones but assign probabilities that are too high.

If opportunities scored at 80% close only 50% of the time, the model is poorly calibrated.

The scores may still help prioritise opportunities, however, they shouldn’t be treated as accurate probabilities without adjustment.

Calibration therefore needs separate testing rather than being assumed from general model accuracy. [26]

Fairness

Forecasting models may also reproduce patterns embedded in historical decisions.

Suppose some sellers, regions, or customer segments historically received:

• more leadership attention,
• better marketing support,
• stronger territories,
• or more favourable deal assignments.

A model may learn that those groups are more likely to succeed without distinguishing underlying opportunity quality from differences in organisational support.

That matters when model outputs influence account allocation, coaching attention, resource investment, or seller performance assessment.

The organisation should examine whether the model produces systematically different outcomes across relevant groups and whether those differences have a defensible commercial basis.

Evaluating vendor claims

Vendors may describe high forecast accuracy in customer examples.

Those claims can be useful evidence that a product has worked in a specific environment.

They don’t establish that the same level of performance will transfer to every organisation.

Before accepting an accuracy claim, the buyer should ask:

• What exactly was forecast?
• Over what time horizon?
• Was accuracy measured at opportunity or aggregate level?
• Which error metric was used?
• Was the model tested on unseen periods?
• What baseline did it outperform?
• How much historical data was available?
• How stable was the sales process?
• Were users changing the pipeline in response to the model?
• Were the predicted probabilities calibrated?

Without that context, “95% accurate” can mean several different things.

The label attached to the model matters less than the evidence showing how it performs under realistic conditions.

Monte Carlo simulation

Monte Carlo simulation is a probabilistic approach used to model uncertainty.

It repeatedly samples from uncertain inputs and calculates the outcome produced by each combination.

After running the process hundreds, thousands, or tens of thousands of times, the organisation receives a distribution of possible outcomes rather than one fixed forecast.

The basic process is:

1. Define the uncertain inputs.
2. Assign a probability or distribution to each input.
3. Sample one possible value from each.
4. Calculate the resulting sales outcome.
5. Repeat the calculation many times.
6. Examine the distribution of results.

The National Institute of Standards and Technology describes Monte Carlo analysis through this same logic of defining uncertain inputs, assigning distributions, and repeatedly sampling from them. [27]

Inputs to a B2B simulation

A B2B sales simulation might use:

• opportunity close probabilities,
• possible deal values,
• likely discounts,
• expected close timing,
• renewal probabilities,
• churn assumptions,
• new pipeline expected during the period,
• and dependencies between opportunities.

Consider a $500,000 opportunity.

The organisation may estimate:

• 55% probability of closing,
• 30% probability that it slips into the next quarter,
• a likely final value between $425,000 and $500,000,
• and a small possibility that scope expands.

One simulation may treat the deal as lost. Another may treat it as closed at $450,000. Another may move it into the following quarter.

The same process runs across the full pipeline.

What the simulation produces

Monte Carlo can produce outputs such as:

• median forecast,
• average forecast,
• downside scenario,
• upside scenario,
• probability of hitting target,
• probability of falling below a threshold,
• and a full distribution of quarter outcomes.

For example, the result might show:

• median bookings: $4.6 million,
• 20th-percentile outcome: $3.8 million,
• 80th-percentile outcome: $5.5 million,
• probability of exceeding the $5 million target: 34%.

These figures are illustrative; the value lies in showing that several results remain plausible rather than forcing leadership to act as though one number is certain.

Median forecast

The median is the point at which half of the simulated outcomes fall below and half fall above.

It can provide a useful central estimate, particularly when the distribution is uneven.

The average may be pulled upward by a small number of unusually high simulated outcomes. The median can sometimes represent the centre of the distribution more clearly.

Neither should be treated as a guaranteed outcome.

Downside and upside scenarios

A downside scenario may represent a lower percentile of the simulated results.

An upside scenario may represent a higher percentile.

Unlike manually created best-case and worst-case scenarios, the ranges come from repeated combinations of the assumptions built into the model.

That makes the relationship between the assumptions and the outcomes more systematic.

Probability of hitting target

One of the most useful outputs is the estimated probability of reaching a target.

Instead of saying, “The forecast is $5 million”, the organisation might say, “The median outcome is $4.7 million, and the simulation estimates a 38% probability of reaching the $5 million target.”

This way, leadership gets more useful information for decisions about investment, hiring, spending, pipeline intervention, and risk.

Dependencies between opportunities

Simple simulations may assume that opportunities close independently, but such an assumption isn’t always realistic.

Two opportunities may depend on:

• the same customer budget,
• the same procurement team,
• the same product release,
• the same partner,
• the same regulatory approval,
• or the same limited implementation capacity.

Some deals may be mutually exclusive.

A customer may choose one contract configuration or another, but not both. Other opportunities may rise or fall together because the same external event affects them.

Ignoring those dependencies can understate or overstate uncertainty.

A more mature simulation should model meaningful relationships where the evidence supports them.

The quality of the assumptions

Monte Carlo makes uncertainty visible, but it doesn’t remove the need for sound assumptions.

If the close probabilities are too optimistic, the simulation will produce an optimistic distribution.

If the deal values are inaccurate, the outputs will reflect those errors.

If the timing assumptions ignore normal procurement delays, the simulation may place too much revenue into the current period.

Running 100,000 simulations doesn’t improve the quality of a weak input; it simply gives a more stable representation of the assumptions already supplied.

False precision

A simulation may report that the organisation has a 43.7% probability of hitting target.

That level of decimal precision can imply more certainty than the inputs justify.

In many business settings, saying the probability is approximately 40% to 45% communicates the result more honestly.

The organisation should match the precision of the output to the quality of the evidence behind it.

Where Monte Carlo works best

Monte Carlo is particularly useful when:

• several inputs remain uncertain,
• leadership needs a range rather than one number,
• the business wants to assess target risk,
• outcomes depend on a portfolio of probabilistic events,
• and the organisation can defend the probability and timing assumptions.

It can support:

• quarterly planning,
• board-level scenario discussions,
• enterprise deal portfolios,
• renewal forecasting,
• and target-risk analysis.

It should remain a decision-support layer.

A polished outcome distribution can’t compensate for weak opportunity inspection, poor probability calibration, or unrealistic timing assumptions.

Point forecasts, ranges, validation, and calibration

Several concepts apply across the model families.

An organisation can select an appropriate model and still misuse the forecast if it doesn’t understand:

• what the output represents,
• how uncertainty is expressed,
• how the model was tested,
• and whether the predictions remain reliable over time.

Point forecasts

A point forecast gives one expected value.

Examples include:

• $3.2 million in quarterly bookings,
• 1,400 renewals next month,
• a 64% close probability,
• or a predicted close date of April 24.

Point forecasts are easy to communicate and compare with targets.

They also compress uncertainty into one number.

A forecast of $3.2 million doesn’t show whether $3.1 million and $3.3 million are the only realistic alternatives or whether the actual result could plausibly range from $2 million to $5 million.

Prediction intervals

A prediction interval provides a range around a point forecast.

A model may estimate:

• point forecast: $3.2 million,
• 80% prediction interval: $2.6 million to $3.9 million,
• 95% prediction interval: $2.2 million to $4.4 million.

The wider interval reflects a higher desired level of coverage.

Prediction intervals normally widen as the forecast horizon increases because uncertainty accumulates.

Next month is generally easier to forecast than the same month two years from now.

A useful forecast should communicate both the central estimate and the uncertainty surrounding it where the decision requires that context. [1]

Probability ranges and distributions

Some forecasts provide a complete distribution rather than one interval.

That distribution can answer questions such as:

• What is the most likely range?
• How much downside risk exists?
• What is the probability of exceeding target?
• How asymmetric is the uncertainty?
• Is there a small chance of a very large result?

Deal-level models may also express uncertainty through probabilities.

For example:

• 70% likelihood of eventual win,
• 45% likelihood of closing this quarter,
• 20% likelihood of slipping into the next period.

These estimates are useful only when the probabilities mean what the organisation assumes they mean.

Calibration

Calibration measures whether predicted probabilities correspond with actual outcomes.

The practical test is straightforward: if opportunities given a 70% close probability only close 40% of the time, the model isn’t well calibrated.

The model is too confident.

If 90% of them close, it may be too conservative.

Calibration should be assessed across groups of comparable predictions.

For example:

• opportunities scored from 0% to 10%,
• 10% to 20%,
• 20% to 30%,
• and so forth.

The organisation then compares the predicted probability with the realised close rate.

A well-calibrated model doesn’t mean every 70% opportunity will close; it means that across enough comparable 70% predictions, the realised proportion should be close to 70%. [23]

Ranking and calibration are different

A model may be good at ranking deals but poor at probability estimation.

Suppose it consistently places the strongest opportunities above the weakest ones.

That can help managers prioritise inspection and resources.

However, if opportunities scored at 80% only close 55% of the time, the probabilities shouldn’t be used directly in a weighted forecast without recalibration.

The organisation should therefore ask whether the model is intended to rank opportunities, estimate reliable probabilities, predict timing, or produce an aggregate forecast.

Performance should be evaluated against that intended use.

Out-of-sample validation

A model shouldn’t be judged only on the data used to build it because that shows how well the model explains the past, not how well it forecasts the future.

Out-of-sample validation tests the model on data it hasn’t previously seen.

A simple approach is to:

1. use earlier periods to build the model,
2. reserve later periods for testing,
3. produce forecasts for those later periods,
4. and compare the predictions with what actually happened.

This creates a more realistic estimate of future performance.

Rolling-origin evaluation

Time-series forecasting requires particular care because data has a natural order.

The organisation shouldn’t train a model using future periods and then test whether it can predict earlier ones.

Rolling-origin evaluation preserves the sequence.

For example:

1. Train the model using January to December and forecast January.
2. Add January’s actual result and forecast February.
3. Add February and forecast March.
4. Continue moving the forecast origin forward.

This shows how the model would have performed if it had been used at each point in time.

It also allows the organisation to compare models across several forecast periods rather than relying on one unusually easy or difficult test.

Forecasting research recommends this type of time-ordered validation for estimating performance on future observations. [1]

Forecast error

Forecast error is the difference between the forecast and the actual result.

If the organisation forecasts $5 million and achieves $4.5 million, the error is $500,000.

Different metrics describe error in different ways.

The organisation may examine:

• absolute error,
• percentage error,
• squared error,
• error by forecast horizon,
• and error by team, segment, or revenue type.

No single metric answers every question.

Absolute error is easy to understand in financial terms.

Percentage error makes results more comparable across differently sized groups but can behave poorly when actual values are close to zero.

Metrics that penalise large errors more heavily may be useful when major misses create disproportionate business consequences.

The evaluation measure should match the decision.

Forecast bias

Accuracy measures how close the forecast is to the outcome.

Bias measures whether errors repeatedly lean in the same direction.

A forecast can have moderate average error while still being systematically optimistic.

For example:

• Quarter 1: forecast $5 million, actual $4.6 million
• Quarter 2: forecast $5.4 million, actual $5 million
• Quarter 3: forecast $5.8 million, actual $5.2 million
• Quarter 4: forecast $6.1 million, actual $5.6 million

The size of the error varies, but the forecast is consistently high.

Persistent overforecasting can affect hiring, spending, investment, cash planning, and leadership confidence.

Persistent underforecasting can lead to missed investment opportunities and may indicate sandbagging or overly conservative assumptions.

A dependable evaluation process should therefore track both error and directional bias.

Comparing complex and simple models

As we already mentioned a few times, more complex models don’t consistently outperform simpler ones.

A machine-learning model may outperform a seasonal baseline in one environment.

In another, a moving average or straightforward regression may prove more accurate, stable, and easier to govern.

The value of complexity depends on:

• the question,
• the data,
• the number of observations,
• the stability of the relationships,
• the validation method,
• and the organisation’s ability to understand and act on the output.

Research on forecasting complexity has long warned that additional sophistication can introduce estimation error and unstable relationships. More recent forecasting evidence shows that machine learning can outperform classical approaches in some settings.

The balanced conclusion is that model performance is contextual. Model class alone is a poor selection rule. [15]

Comparing against a baseline

Every advanced model should be compared with a sensible baseline.

Possible baselines include:

• the previous period,
• the same period last year,
• a simple historical average,
• a seasonal baseline,
• run rate,
• or stage-weighted pipeline.

If the advanced model doesn’t outperform the baseline consistently on unseen data, its additional complexity may not be justified.

The comparison should consider more than average error.

The organisation should also assess bias, calibration, stability, explainability, implementation effort, maintenance, and whether the model improves decisions.

Governing the model over time

Validation isn’t a one-off project completed before launch.

The organisation should continue to monitor:

• forecast versus actual,
• forecast bias,
• probability calibration,
• error by horizon,
• performance by segment,
• changes in input quality,
• and signs of model drift.

It should also preserve model and forecast snapshots.

Without snapshots, the business may compare the final forecast issued one day before quarter-end with the final result and conclude that the model was accurate.

That ignores whether the forecast was useful six weeks earlier, when leadership still had time to act.

Forecast performance should be assessed at the horizons where decisions are actually made.

The top revenue system I've implemented: reps reclaim 5-8h admin hours weekly, CRM accuracy up 100%, agents that act on pipeline signals, and it's all AI-native.

See how Airspeed works as a full revenue system based on call intelligence.

TAKE THE INTERACTIVE WALKTHROUGH

Sponsored placement – I only recommend companies I know and trust.

Conclusion

Sales forecasting isn’t one standard calculation, and there is no single method that will work best in every situation.

Each approach starts from a different type of evidence and answers a different question, whether that means drawing on seller judgement, using historical patterns, assessing the current pipeline, or applying statistical and probabilistic models.

The right choice depends on what the organisation is trying to forecast, the nature of the sales motion, the forecast horizon, the quality of the available data, and how the result will be used.

A more sophisticated model isn’t automatically more valuable; it still needs to produce a reliable estimate, remain understandable enough to govern, and help people make better commercial decisions.

This guide has focused on the methods and models used to produce a sales forecast. We haven’t gone into the wider sales forecasting process in depth, including ownership, forecast categories, submission and review cadence, the reconciliation of different forecast views, overrides, and performance evaluation.

We’ll cover those elements separately in a future guide to building a reliable B2B sales forecasting process.

Ultimately, a dependable forecast comes from matching the approach to the question, using credible evidence, making uncertainty visible, and learning from the difference between what was forecast and what actually happened.

References

[1] Hyndman, R. J., and Athanasopoulos, G. (2021). Forecasting: Principles and Practice, 3rd ed. Melbourne: OTexts. Accessed 5 August 2026. https://otexts.com/fpp3/

[2] Institute of Business Forecasting and Planning. (n.d.). “Model.” IBF Glossary. Accessed 5 August 2026. https://ibf.org/knowledge/glossary/model-189

[3] Salesforce. (n.d.). “Customize Pipeline Forecast Categories” and “Salesforce Forecasting Elements.” Salesforce Help. Accessed 5 August 2026. Customize Pipeline Forecast Categories; Salesforce Forecasting Elements.

[4] Salesforce. (n.d.). “Forecast Your Opportunity Pipeline.” Salesforce Help. Accessed 5 August 2026. https://help.salesforce.com/s/articleView?id=sales.forecasts3_use.htm&language=en_US&type=5

[5] Salesforce. (n.d.). “Adjustments in Pipeline Forecasts.” Salesforce Help. Accessed 5 August 2026. https://help.salesforce.com/s/articleView?id=sales.forecasts3_adjustments_overview.htm&language=en_US&type=5

[6] Institute of Business Forecasting and Planning. (n.d.). “Forecasting Process.” IBF Glossary. Accessed 5 August 2026. https://ibf.org/knowledge/glossary/forecasting-process-132

[7] Clemen, R. T. (1989). “Combining Forecasts: A Review and Annotated Bibliography.” International Journal of Forecasting, 5(4), 559–583. https://doi.org/10.1016/0169-2070(89)90012-5

[8] Januschowski, T., Gasthaus, J., Wang, Y., Salinas, D., Flunkert, V., Bohlke-Schneider, M., and Callot, L. (2020). “Criteria for Classifying Forecasting Methods.” International Journal of Forecasting, 36(1), 167–177. https://doi.org/10.1016/j.ijforecast.2019.05.008

[9] Lawrence, M., Goodwin, P., O’Connor, M., and Önkal, D. (2006). “Judgmental Forecasting: A Review of Progress over the Last 25 Years.” International Journal of Forecasting, 22(3), 493–518. https://doi.org/10.1016/j.ijforecast.2006.03.007

[10] Green, K. C., and Armstrong, J. S. (2007). “Structured Analogies for Forecasting.” International Journal of Forecasting, 23(3), 365–376. https://doi.org/10.1016/j.ijforecast.2007.05.005

[11] Kahn, K. B. (2014). “Solving the Problems of New Product Forecasting.” Business Horizons, 57(5), 607–615. https://doi.org/10.1016/j.bushor.2014.05.003

[12] Eroğlu, C., and Croxton, K. L. (2010). “Biases in Judgmental Adjustments of Statistical Forecasts: The Role of Individual Differences.” International Journal of Forecasting, 26(1), 116–133. https://doi.org/10.1016/j.ijforecast.2009.02.005

[13] Fildes, R., and Goodwin, P. (2007). “Against Your Better Judgment? How Organizations Can Improve Their Use of Management Judgment in Forecasting.” Interfaces, 37(6), 570–576. https://doi.org/10.1287/inte.1070.0309

[14] Monash Business School. (n.d.). “Sales Force Composite.” Marketing Dictionary. Accessed 5 August 2026. https://www.monash.edu/business/marketing/marketing-dictionary/s/sales-force-composite

[15] Green, K. C., and Armstrong, J. S. (2015). “Simple versus Complex Forecasting: The Evidence.” Journal of Business Research, 68(8), 1678–1685. https://doi.org/10.1016/j.jbusres.2015.03.026

[16] Thompson, J. (2026). “12 Forecasting Models to Leverage for Better Sales Forecasting.” HubSpot Sales Blog, updated 9 June 2026. https://blog.hubspot.com/sales/accurate-sales-forecasting-model-tips

[17] Baker, K. (2026). “Lead Source: What Is It and Why It’s Valuable to Your Team.” HubSpot Marketing Blog, updated 1 April 2026. https://blog.hubspot.com/marketing/lead-source

[18] Salesforce. (n.d.). “Manage Opportunity Stage to Forecast Category Mappings for Pipeline Forecasts”; “Opportunity Fields”; and “Pipeline Forecasting Best Practices.” Salesforce Help. Accessed 5 August 2026. Stage-to-category mappings; Opportunity Fields; Pipeline Forecasting Best Practices.

[19] Microsoft. (n.d.). “Analyze Revenue Outcome by Using Predictive Forecasting”; “Opportunity Management FAQs”; and “Prioritize Opportunities Through Predictive Scores.” Microsoft Learn. Accessed 5 August 2026. Predictive forecasting; Opportunity management FAQs; Predictive opportunity scores.

[20] Oracle. (n.d.). “Forecasting Preferences.” NetSuite Applications Suite. Accessed 5 August 2026. https://docs.oracle.com/en/cloud/saas/netsuite/ns-online-help/section_N1046772.html

[21] Richter, J. (2026). “Is Revenue Leak Hiding in Your Forecast?” Clari, 12 June 2026. https://www.clari.com/blog/what-is-revenue-leak-and-how-can-you-prevent-it/

[22] Duncan, M. (2024). “Which of These 6 Sales Forecasting Methods Is Right for Your Sales Org?” Gong, updated 4 March 2026; Gong. (2024). “See Forecast Projection,” updated 25 June 2026. Sales forecasting methods; See forecast projection.

[23] Rezazadeh, A. (2020). “A Generalized Flow for B2B Sales Predictive Modeling: An Azure Machine-Learning Approach.” Forecasting, 2(3), 267–283. https://doi.org/10.3390/forecast2030015

[24] Yan, J., Gong, M., Sun, C., Huang, J., and Chu, S. M. (2015). “Sales Pipeline Win Propensity Prediction: A Regression Approach.” In 2015 IFIP/IEEE International Symposium on Integrated Network Management, 854–857. https://doi.org/10.1109/INM.2015.7140393

[25] Shaik, M., Sridhar, S., Sriskandarajah, C., and Mittal, V. (2026). “Opportunity Management for Business-to-Business (B2B) Service Organizations: A Theory-Informed Decision Support Framework.” Production and Operations Management, 35(6), 2457–2475. https://doi.org/10.1177/10591478251407559

[26] scikit-learn developers. (n.d.). “Probability Calibration.” scikit-learn User Guide. Accessed 5 August 2026. https://scikit-learn.org/stable/modules/calibration.html

[27] Thomas, D. (2025). Monte Carlo Tool, version 1.2 [computer software]. National Institute of Standards and Technology. https://www.nist.gov/services-resources/software/monte-carlo-tool