Sales forecasting models explained: which one fits your data, your stage and your sales cycle.

What are sales forecasting models?

Sales forecasting models are the rules a team uses to turn an open pipeline and a sales history into one expected revenue number for a period. The model decides which deals count, how much of each one counts, and what the total means.

Every model answers the same question in a different way: what predicts the future here? One says the rep knows. One says last quarter knows. One says the stage a deal has reached knows. One says a regression across your whole history knows.

The forecast that comes out is not a target and not a wish. It is an estimate of revenue your team expects to book in a defined window, produced the same way every period so the error can be measured.

That last part is what separates a forecasting model from a guess. A guess cannot be scored. A model produces a number, a method, and a record you can compare with what actually closed.

Sales forecast methods, models and the forecast itself

People use the words interchangeably, and the difference is small but useful. Sales forecast methods and sales forecasting models both name the calculation. The forecast is the output. The forecast call is the meeting where the output gets challenged.

A forecast and a sales quota are not the same thing. Quota is what you promised to deliver. The forecast is what the pipeline currently says you will deliver. When they match exactly every period, someone is reporting the quota, not the pipeline.

Which model fits your team's stage and data

Choose by the data you have, not by the model you admire. A regression needs hundreds of closed deals. A weighted pipeline needs stages with written exit criteria. An intuitive forecast needs nothing but reps and honesty, which is why new teams start there.

Combining two models on purpose

Few teams should run one model alone. The useful pairing puts a pipeline based model against a history based one, because the two fail in opposite directions and the disagreement between them is informative.

• Weighted pipeline plus historical: the pipeline says what is open, history says what your sales team has ever actually delivered.
• Opportunity stage plus sales cycle length: stage says how far a deal got, deal age says whether it is still moving.
• Bottom up plus top down: one shows the revenue that exists today, the other shows what the plan demands.
• Lead driven plus pipeline: lead math covers the future periods your current pipeline cannot reach yet.

Do not average the two outputs into one number and move on. Keep both, and when they diverge, decide which assumption you believe this period and write down why. That written reason is the fastest way for sales teams to get better at forecasting.

When you have no closed history yet

A new business, a new product or a first move into a new market has no closed data to learn from. The honest answer is that the future is genuinely uncertain, so present the forecast as a range with named assumptions rather than one confident number.

Build it from things you can count: meetings booked, proposals sent, average deal size in comparable segments, and how much selling capacity the team has. Write the assumption behind each input beside it, so next period you correct one assumption instead of the whole method.

Treat the first two or three periods as measurement. As soon as real closed won and closed lost deals exist, your own win rates replace the guesses and the team can move to opportunity stage or weighted pipeline math.

The data every forecasting model needs

Every model on this page reads the same small set of fields, and none of the sales forecast methods survives bad ones. A simple model on clean data beats a sophisticated model on guesswork, every quarter, in every business.

• Amount: the revenue expected in the period being forecast, not total contract value spread across several future years.
• Close date: a date tied to a customer event such as a budget approval or a renewal, moved deliberately when it changes.
• Stage: set by written exit criteria, so the same stage means the same thing across reps, segments and teams.
• Owner, segment and market: so win rates and forecast accuracy can be measured where they actually differ.
• Closed won and closed lost history: losses matter as much as wins, because every probability is computed from both.
• Created date and stage timestamps: without them you cannot compute cycle length or spot a deal that stopped moving.

Run one audit before you trust any forecast. Count open deals with a close date already in the past, deals with no activity for a month, and deals with no identified signer. That list is usually the gap between an accurate forecast and a hopeful one.

Intuitive forecasting

Intuitive forecasting asks the people closest to the deal how likely it is to close, then rolls those answers up. It is the oldest method and still the first one most young teams use, because it needs no history at all.

It is not worthless. A rep knows the champion went quiet, that procurement added a step, that the budget owner is leaving. No CRM field holds those facts on the day they matter.

• Use it when: the product is new, the pipeline is small, or you are entering a market with no comparable history.
• Force structure on it: ask for the same three things per deal, the signer, the evidence and the next dated step.
• Watch for: sandbagging near quota time, and optimism right after a good week.
• Retire it when: you have enough closed deals to compute stage win rates from your own data.

Historical forecasting

Historical forecasting takes what you closed in comparable past periods and projects it forward, sometimes with a growth assumption added. Last quarter closed a certain amount, so this quarter should land near it, adjusted for seasonality and headcount.

It is fast, it needs no pipeline hygiene, and it is an excellent sanity check against pipeline math. When your weighted pipeline says the quarter will be double anything you have ever closed, the history is telling you something the pipeline is not.

The weakness is that it assumes the world holds still. A pricing change, a new competitor, a churned enterprise account or a hiring freeze all break the assumption, and the model will not notice until the period ends.

Time series forecasting and seasonality

Time series methods extend historical forecasting by treating your sales data as a sequence over time. Instead of comparing one period with the last, they separate the underlying trend, the seasonal pattern that repeats each year, and the noise around both.

Moving averages smooth month to month swings so a trend becomes visible. Exponential smoothing does the same while weighting recent periods more heavily, which helps when the business is changing. More advanced time series methods model the seasonal cycle explicitly.

These methods need several years of clean revenue data to tell a real season apart from a run of good luck. They also forecast the business, not individual deals, so they sit beside pipeline models rather than replacing them.

Seasonality is worth measuring even if you never build a time series model. If your market buys on budget cycles, a forecast that ignores the calendar will look accurate on average and be wrong every single quarter.

Length of sales cycle forecasting

Length of sales cycle forecasting judges each deal by its age against your average cycle. A deal that has been open for two thirds of the normal cycle is treated as further along than one opened last week, whatever stage the rep has selected.

The appeal is that it uses an objective input. Time in the pipeline is recorded automatically, and no rep can talk it up in a meeting. It also exposes zombie deals, because a deal far past the average cycle is a warning, not a commit.

It works best where the process rarely varies. Segment first: an enterprise cycle and a self serve cycle averaged together produce a number that describes neither. Compute a separate average cycle per segment, and per lead source if sources behave differently.

Opportunity stage forecasting

Opportunity stage forecasting assigns a close probability to each stage of the pipeline, based on the historical win rate of deals that reached that stage. Each open deal is multiplied by its stage probability, and the results are added.

This is the model most CRM reports assume. It only works when stages mean something specific. If a deal can sit in "proposal" because the rep sent a deck, the probability attached to that stage is measuring nothing. Our guide on how to build a sales pipeline covers writing exit criteria that hold up.

Derive the probabilities from your own closed data: of the deals that ever reached this stage, what share ended as closed won? Recompute them at least once or twice a year, and separately per segment when segments win at different rates.

The blind spot is time. Stage forecasting treats a deal that reached negotiation yesterday and one that has been stuck there for five months identically. Pair it with deal age, which is exactly what the sales cycle model contributes.

Weighted pipeline forecasting

A weighted pipeline is the sum of open deals after each amount is multiplied by a close probability. It is the same arithmetic as stage forecasting, presented as a running pipeline number rather than a period estimate.

The mechanics are built into the CRM. HubSpot's documentation states that each deal stage has an associated probability indicating the likelihood of closing deals in that stage, and that stage probability is used to determine the weighted amount shown in board view, calculated by multiplying the total amount in each stage by the stage probability.

At the deal level, HubSpot defines the weighted amount property as the amount multiplied by the deal probability, and the forecast amount property as the deal amount multiplied by the forecast probability, which is a custom percent probability that the deal will close.

A worked example

The stage probabilities below are the defaults HubSpot documents for its Sales Pipeline. The amounts are invented for this page, and the arithmetic is the whole point: five deals worth two hundred thousand dollars do not forecast as two hundred thousand dollars.

HubSpot also documents Closed won at 100% and Closed lost at 0%, and requires a pipeline to include stages for both Won and Lost so that reports and sales analytics process deals correctly.

Two cautions. First, replace the shipped defaults with probabilities computed from your own wins, because a template number is a placeholder, not a measurement. Second, a weighted total is an average outcome, and no single deal closes at 60 percent: it closes or it does not.

Lead driven forecasting

Lead driven forecasting works from the top of the funnel instead of the bottom. Take the number of qualified leads by source, multiply by the historical value each source produces, and you have an estimate for the periods those leads will close in.

It suits high volume inbound motions where individual deals are small and the law of averages holds. It is also the only model that produces a useful forecast for a period further out than your current pipeline covers.

• Inputs you need: lead volume per source, the share that becomes an MQL, the conversion to closed won, and average deal size.
• Segment by source: value per lead differs sharply across lead sources, so one blended average hides the truth.
• Respect the lag: leads created this month close in a later period, so the model forecasts forward, not now.
• Recheck quarterly: source quality drifts as channels, offers and ideal customer profile definitions change.

Multivariable and regression forecasting

Multivariable forecasting fits a statistical model across several inputs at once instead of relying on one. Deal size, lead source, industry, rep, stage, deal age and activity count all get a weight, and the model returns a probability for each open deal.

Regression analysis is the usual engine. Rather than assuming every deal in a stage behaves the same, it learns that a certain segment closes faster, or that deals sourced one way rarely reach signature, and reflects that in the number.

The cost is data and explanation. You need a long, clean history of both wins and losses, someone who can maintain the model, and a way to answer the question every sales leader asks: why does it say that about my deal?

Start smaller if that is out of reach. Splitting stage probabilities by one or two dimensions, such as segment and source, captures much of the benefit and can be explained on a whiteboard.

AI and predictive forecasting

AI forecasting is multivariable forecasting with more inputs and less human tuning. Vendors score deals using CRM fields plus activity signals such as email and meeting patterns, and some also predict the month a deal is most likely to close.

Whether it helps depends entirely on the data underneath. A model trained on a CRM where close dates are guessed and stages are skipped will learn those habits and repeat them back with more confidence than they deserve.

Treat a score as a participant in the forecast call, not the chair. If the model disagrees with the rep, that disagreement is the most useful minute of the meeting, and someone should have to explain it.

Bottom up versus top down

Every model above is bottom up: it builds a number from deals, leads or history that already exist. Top down starts from the other end, with a market size or a company target, and divides it into segments, territories and rep quotas.

Run both and read the gap between them. A top down plan that needs far more than bottom up shows is a pipeline generation problem, not a forecasting problem, and it belongs in your B2B sales strategy review rather than the weekly call.

Forecast categories in the CRM

Most CRMs layer a confidence label over the stage, so a rep can say a late stage deal will slip without dragging it backward through the pipeline. These labels decide which column of the forecast grid a deal lands in.

Microsoft's documentation is blunt about one habit: do not select Won or Lost manually. Close the opportunity through the close dialog so the category updates itself, because setting it by hand can make the deal vanish from the forecast or produce inaccurate rollup values.

The categories are also what makes a rollup possible. In Dynamics 365 Sales the forecast grid shows quota, committed, best case, pipeline, omitted, won and lost, aggregated level by level up the sales hierarchy defined by the manager field on each user record.

What the sales forecast is used for

The sales forecast leaves the sales team quickly. Finance builds the cash plan on it, leadership sets hiring against it, and the board reads it as a statement about business performance. That is why an accurate number beats a flattering one.

• Revenue and cash planning: finance schedules spending against expected revenue per period, not against the whole open pipeline.
• Hiring and capacity: how many reps, delivery people and support staff the next two quarters actually need.
• Quota and territory design: next year's targets are built on this year's forecast accuracy and market read.
• Supply and onboarding: in businesses with goods to ship or customers to onboard, the forecast decides what gets bought and staffed.
• Early warning: a forecast falling short while time is left in the period is the cheapest signal sales teams get.

Each use has its own tolerance for error. Finance can absorb a small miss on a monthly number. A hiring plan built on a forecast that missed by a wide margin costs far more and takes several quarters to undo.

Forecast accuracy and how to measure it

A forecast you never score is a story. Accuracy is measured by comparing the number you submitted with the revenue that actually closed in the same period, and keeping that comparison as a running series.

HubSpot's forecast tool documents the arithmetic plainly. Forecast error percentage is the absolute difference between actual and forecast, divided by actual, multiplied by one hundred. Accuracy is the maximum of zero and one minus that error percentage.

Two rules keep the measurement honest. Record the forecast before the period ends, not after, and keep the record where nobody can quietly edit it. Then judge on the trend, because a single accurate quarter can be two large errors canceling out.

Treat accuracy as a performance measure for the forecasting process, not as a stick for reps. Where forecast error is punished, teams protect themselves by committing late, and the forecasts then arrive too close to period end to be useful to anyone.

Score each method separately for a few periods before you pick a favorite. Run the weighted pipeline number and the historical number side by side, record both, and let the error data decide which one your business should lead with.

Why forecasts miss

Forecast misses are rarely mysterious. They come from a short, repeating list, and almost all of them are data or discipline problems rather than model problems.

• Close dates are wishes. Dates set to the end of the quarter because that is where the rep hopes the deal lands, with no customer event behind them.
• Template probabilities. Shipped defaults left in place for years while your real win rates moved.
• Stages without exit criteria. A deal advances because a demo happened, not because the buyer did something.
• No identified signer. Nobody can name who signs or what the approval path is, so the last two weeks are a surprise.
• Weak qualification. Deals enter the pipeline that never had budget or a problem worth solving, see how to qualify sales leads.
• Nobody scrubs. Dead deals sit open, inflating both the raw and the weighted number, including the obvious tire kickers.
• Averages on small numbers. A team closing a handful of large deals per quarter cannot be forecast with percentages alone.
• Pressure at the top. When a low forecast gets punished, the forecast stops describing the pipeline.

A forecast cadence that holds the number honest

The model matters less than the rhythm around it. A forecast cadence is the fixed set of moments where the number gets built, challenged, submitted and scored, with the same fields inspected every time.

Keep the forecast call separate from pipeline generation and from deal coaching. Mixing them turns an hour of inspection into a status meeting, and the commit list never gets the scrutiny it needs. Your sales cadence covers the outreach rhythm; this one covers the number.

How to run a weekly forecast call

1. Freeze the data the night before

   Reps update amount, close date, stage and category by a set hour. The call works from that snapshot, so nobody edits the pipeline while the meeting runs.

2. Start with last week's number

   Read out what the team committed last week and what actually closed. Naming the miss first sets the tone for honest numbers.

3. Walk the commit list deal by deal

   For every committed deal ask for the evidence, the signer, the open approval step and the next dated action. No evidence means no commit.

4. Pressure test best case

   Ask what would have to be true this period for each best case deal to move to commit. Move the ones with a dated path, park the rest.

5. Name the gap and the plays

   Subtract the commit from the target, say the gap out loud, and agree which specific deals or plays close it. Assign each play an owner.

6. Log the number and score it later

   Record the submitted forecast, then compare it with actual closed won at period end and keep the error rate as a running series.

The questions matter more than the agenda. A manager who asks the same four questions of every committed deal will surface slipping deals weeks earlier than one who asks how each deal feels.

What a forecast actually runs on

This page does not rank vendors. These are the categories any forecasting model depends on, and the first two matter far more than the rest.

Common sales forecasting mistakes

• Leaving shipped stage probabilities in place instead of computing your own win rates.
• Forecasting a small number of large deals with percentages, where the average outcome never happens.
• Changing the model mid period, which makes the accuracy record meaningless.
• Letting the forecast drift upward to match quota because a low number is unwelcome.
• Running the call on whatever the CRM shows at that moment, with no frozen snapshot.
• Skipping the scoring step, so nobody ever learns which model or which rep is reliable.
• Treating an AI score as a verdict instead of a second opinion that must be explained.
• Forecasting the whole business on one blended cycle length when segments behave differently.

Most of these are cheap to fix and none of them require a new tool. They require a defined B2B sales process and a manager willing to ask for evidence in front of the team.

The note a rep writes before the call

The template below was written for this page. It is the short note a rep posts on each committed deal before the forecast call, so the meeting starts from evidence instead of adjectives and takes half the time.

Deal: {{account}}, {{dealName}}
Amount: {{amount}} | Close date: {{closeDate}} | Category: {{category}}

Why it closes this period: {{evidence}}
Who signs: {{signer}}
Step still open: {{approvalStep}}
Next dated action: {{nextAction}} on {{nextDate}}
Biggest risk: {{risk}}
If that risk lands, this deal moves to {{fallbackPeriod}}.

Frequently asked questions