10 Best Sales Forecasting Methods (And When Pros Use Each)

Most sales forecasts are wrong, and the people who make them know it. Only 45% of sales leaders and sellers have high confidence in their own forecast. In a 2022 vendor survey of 928 sales professionals, 52% of leaders said their company regularly misses its forecast by more than 10%.

The fix is rarely a smarter spreadsheet; more often it is choosing the right sales forecasting method for the data you actually have and then checking it against a second method. The ten methods below cover the ones working sales leaders use: historical run-rate, time series, opportunity stage, length of sales cycle, lead-driven, regression, rep commit with forecast categories, top-down market sizing, bottom-up capacity, and AI or predictive models. Each one is run on the same made-up company and the same quarter, which shows how far apart honest methods can land.

The Startup Behind Every Example

To keep the math comparable, every worked example uses one fictional company: a 14-person B2B software startup that sells booking software to dental clinics on annual contracts. The numbers are invented for teaching and do not describe any real business.

The goal is to forecast new bookings for Q1 2027, meaning the first-year contract value of new customers signed from January to March. The table lists the startup’s last eight quarters, and the third column is used later by the regression method.

Quarter	New bookings ($k)	Pipeline created in the prior quarter ($k)
Q1 2025	410	1,450
Q2 2025	455	1,580
Q3 2025	430	1,560
Q4 2025	560	1,700
Q1 2026	470	1,690
Q2 2026	520	1,800
Q3 2026	495	1,770
Q4 2026	640	1,940

Two patterns matter, and a good forecast has to respect both: bookings grew about 15% from 2025 to 2026, and Q4 is always the biggest quarter while Q1 is the smallest. If your own business sells subscriptions, keep in mind that new bookings are only half of the revenue story, because you also lose some existing revenue each quarter to customer and revenue churn.

Forecasting From Past Sales

The first two methods ignore the deals in your CRM and look only at what you sold before, on the assumption that the pattern holds. That makes them quick and hard to game, but blind to anything new.

1. Historical Run-Rate Forecasting

A run-rate forecast takes a past period and grows it by a recent growth rate. The most common version uses the same quarter last year, so that seasonality is built in.

The startup booked $1,855k in 2025 and $2,125k in 2026, so the year-over-year growth was 2,125 ÷ 1,855 = 1.146, or 14.6%. Q1 2026 brought in $470k. The forecast for Q1 2027 is 470 × 1.146 = $538k.

All you need is a clean sales history by month or quarter. It works best for stable businesses with steady demand, such as a retailer with years of data or a subscription company in a mature market. It fails at turning points, because a new competitor, a price change, a lost channel partner or a recession breaks the link between last year and this year while the run-rate keeps forecasting the old world.

2. Time-Series Forecasting With Moving Averages and Seasonality

Time-series methods also use only past sales, but they split the history into a trend, a seasonal pattern and noise. The simplest tools are moving averages and seasonal indexes, while more advanced ones, such as exponential smoothing or ARIMA models, follow the same idea with more statistics.

The plain version takes four steps, using the table above.

• Find the seasonal index. Divide each quarter by its year’s average quarter, then average the two years. Q1 comes out at 0.884, Q2 at 0.980, Q3 at 0.929 and Q4 at 1.206. In other words, Q1 normally runs about 12% below an average quarter.
• Remove the season. Divide each quarter’s bookings by its index. The adjusted figures sit near $464k a quarter in 2025 and near $531k in 2026.
• Fit a trend. A straight line through those eight adjusted points rises by about $12.9k a quarter and reaches $555k for Q1 2027.
• Put the season back. Multiply 555 × 0.884 to get $491k.

A plain four-quarter moving average shows the method’s weak spot. The last four adjusted quarters average $531k, and 531 × 0.884 gives just $470k, because a moving average describes the middle of its window rather than today. In a growing company it forecasts low, and in a shrinking one it forecasts high.

You need at least two full years of data by month or quarter, and preferably three or more, so the seasonal pattern is real. Time series suits companies with many small transactions, such as ecommerce stores, consumer apps or distributors, where no single deal moves the total. It struggles with lumpy B2B sales, where three large contracts can make or break a quarter.

Pipeline-Based Sales Forecasting Methods

The next three methods look forward, forecasting from deals or leads that already exist, so they react faster to good and bad news. The price is that they are only as good as the data your reps put into the CRM. Only 47% of sales teams say their organization has high-quality data.

3. Opportunity Stage Forecasting (Weighted Pipeline)

This is the default method in most CRMs. Each stage of your sales process gets a win probability, and the forecast is the sum of every open deal’s value multiplied by its stage probability.

What matters most is where the probabilities come from. Many teams accept the CRM’s default percentages, which were never based on their own sales. Pull your own history instead: the share of all deals that reached the Proposal stage and were later won is your Proposal probability.

The startup’s open pipeline for deals expected to close in Q1 2027 looks like this.

Stage	Open deals	Pipeline value ($k)	Historical win rate	Weighted value ($k)
Discovery	38	1,140	8%	91.2
Demo	21	651	22%	143.2
Proposal	11	374	45%	168.3
Negotiation	6	216	72%	155.5
Total	76	2,381		558.2

The weighted forecast is $558k, the highest of the nine numeric forecasts, and that points to the method’s main flaw: a win rate tells you whether a deal will close, not when. Some of those 38 Discovery deals will close in May or July, but the method counts their full weighted value in Q1. Stale deals that reps never close out add to the problem.

You need a CRM with consistent stages, clear exit criteria for each stage and enough closed deals to set reliable win rates. It fits B2B teams with a defined sales process and deal sizes that are fairly similar. It breaks when reps move deals forward too early or keep dead deals open, because the forecast inflates without any change in real demand.

4. Length of Sales Cycle Forecasting

This method adds time to the pipeline by asking how old each deal is compared with your typical sales cycle, and it gives younger deals a lower chance of closing this quarter.

A simple version divides a deal’s age by the average cycle length. If your average won deal takes 84 days and a deal is 42 days old, it is halfway through and gets a 50% weight. The more useful version, which needs a little more data work, uses your history to find what share of pipeline at each age closed within the next quarter.

Deal age	Pipeline value ($k)	Share that closed within a quarter, historically	Weighted value ($k)
0 to 30 days	980	12%	117.6
31 to 60 days	760	27%	205.2
61 to 120 days	520	36%	187.2
Over 120 days	121	8%	9.7
Total	2,381		519.7

The same $2,381k pipeline now forecasts $520k, about $39k less than the stage method. The difference comes from the young deals and from the $121k of old deals, which have mostly gone quiet even if they still sit in a late stage.

You need the created date and close date for every past deal, which most CRMs store automatically. This method fits B2B companies with a fairly steady cycle. It fails when the cycle varies a lot by deal type. A $5k deal and a $150k enterprise deal do not age at the same speed, so split them before you average.

5. Lead-Driven Forecasting

Lead-driven forecasting starts one step earlier, with the leads coming in. For each lead source, multiply the number of leads by the historical lead-to-customer conversion rate and the average deal size from that source.

Timing matters here too: the startup’s sales cycle is about 84 days, so most of Q1’s deals will come from leads created in Q4 2026.

Lead source	Q4 2026 leads	Lead-to-win rate	Expected wins	Average deal ($k)	Forecast ($k)
Inbound demo requests	260	7%	18.2	11	200.2
Outbound meetings booked	340	4%	13.6	13	176.8
Partner referrals	45	18%	8.1	16	129.6
Total	645		39.9		506.6

The lead-driven forecast is $507k, the closest to the hypothetical actual of $512k, though one quarter of invented data proves nothing.

You need lead source tracking from the first touch to a closed deal, which is harder than it sounds once a lead touches three channels. The method suits marketing-led and inbound businesses, and it doubles as a planning tool: if you want $600k next quarter, you can work backward to the leads you need, and from there to the cost of getting them with your blended and paid customer acquisition cost. It fails when conversion rates change quietly, since a new sales hire, a pricing change or a shift in lead quality can move them long before anyone updates the model. Better call handling moves the rate too, which is one reason teams invest in tested sales call scripts.

Forecasting From the Drivers of Sales

6. Regression and Multivariable Analysis

Regression finds the link between sales and the things that drive them, such as new pipeline, marketing spend, number of reps, website traffic or price. Simple regression uses one driver, while multivariable regression uses several at once and estimates how much each one adds.

For the startup we used two drivers: the pipeline created in the prior quarter and a flag for Q4, the busy year-end quarter. Fitted on the eight quarters in the first table, the model is:

New bookings = −35.4 + (0.304 × prior-quarter pipeline created) + (82.5 if the quarter is Q4)

In plain words, every $1,000 of new pipeline created this quarter adds about $304 of bookings next quarter, and Q4 adds about $82k on top. In Q4 2026 the team created $1,870k of new pipeline, and Q1 is not a Q4 quarter. The forecast is −35.4 + 0.304 × 1,870 = $533k.

The model fits the past almost perfectly, with an R-squared of 0.99, and you should treat that as a warning rather than a win. With only eight data points and two drivers, a model can fit the past by luck and still miss the future. Regression needs more history than any method so far, ideally several years of monthly data, and drivers that make business sense on their own. It suits larger companies with good records and a sales or finance analyst who can test the model. It fails when two drivers move together (such as rep count and pipeline), because the model cannot tell them apart, and when the business changes in ways the past data never saw.

Judgment, Market Math and Sales Capacity

7. Rep Commit With Forecast Categories

This is the intuitive method that most sales teams still run every week: each rep places every deal in a forecast category, and managers roll the categories up into a forecast call. In Salesforce, the standard categories are Pipeline, Best Case, Commit, Closed and Omitted, and many teams use the same words in other CRMs.

• Commit: the rep is confident the deal will close this period and is willing to be held to it.
• Best Case: the deal could close this period if things go well.
• Pipeline: a real opportunity, but too early to call.
• Closed: already won.
• Omitted: left out of the forecast.

In week two of Q1, the startup’s reps call $60k Closed, $340k Commit, $230k Best Case and $600k Pipeline. A raw call of Closed plus Commit is $400k. A better call adjusts each category by how it has converted in past quarters. If Commit has historically closed at 88%, Best Case at 35% and Pipeline at 6%, the adjusted forecast is 60 + (340 × 0.88) + (230 × 0.35) + (600 × 0.06) = $476k.

That is still the lowest result in our set, which is typical early in a quarter. Reps tend to protect themselves by committing only what is nearly signed, a habit called sandbagging, while others do the opposite and commit deals they merely hope for. Only 24% of sales leaders say they trust their reps’ forecast commitments.

The method needs only a CRM and a weekly forecast meeting. It fits almost every B2B team and captures things no model can see, such as a champion who just left the customer, but it fails when categories mean different things to different reps. Write down what each one requires (for example, Commit means a verbal yes and a known signing date) and track each rep’s commit accuracy over time.

8. Top-Down Market Sizing

A top-down forecast starts with the size of the market and estimates what share of it you can win. Founders use it in pitch decks, and investors know to treat it with care.

The startup sells to an estimated 18,000 dental clinics in the regions it covers. If it wins 1.3% of them a year, that is 234 new customers. At an average first-year contract of $10.5k, that equals $2,457k of new bookings for the year. Q1 has made up 22.1% of annual bookings in both past years, so the Q1 2027 forecast is 2,457 × 0.221 = $543k.

All you need is a count of potential buyers, an average deal size and an honest share assumption. It is useful before you have any sales history, and as a sanity check for any other forecast: if the bottom-up math implies you will win 10% of your whole market this year, something is off. Its failure mode is plain optimism, because the share number is a guess and small changes swing the result. Moving from 1.3% to 1.5% adds about $84k to this quarter’s forecast. Founders who use it should show the market math next to a break-even calculation, so the plan states how many customers the company needs, rather than how many it hopes for.

9. Bottom-Up Capacity Planning

Capacity forecasting asks how much your sales team can sell: multiply the number of quota-carrying reps by their quota and by their real average attainment, and adjust new reps for the time they need to ramp.

The startup has 8 fully ramped account executives with a quarterly quota of $85k each. Over the past year they hit an average of 72% of quota. That gives 8 × 85 × 0.72 = $489.6k. Two reps hired in Q4 are still ramping and are expected to produce 40% of a full quota at the same 72% attainment, adding 2 × 85 × 0.40 × 0.72 = $49.0k. The total is $539k.

You need headcount plans, quotas, attainment history and a ramp schedule. This method is the backbone of annual planning and hiring, because it ties revenue goals to the number of people you must hire and when. It also shows how long each new hire takes to pay back, which links directly to your CAC payback period. It fails when there is not enough demand to keep the reps busy, since capacity is a ceiling and ten reps with an empty pipeline still sell nothing.

Machine Learning Enters the Forecast

10. AI and Predictive Forecasting

AI forecasting tools use machine learning to score each open deal, and sometimes the whole quarter, from patterns in your past data. Instead of one fixed win rate per stage, a model looks at many signals: how recently the buyer replied, how many people at the customer are involved, whether a pricing page was opened, how long the deal has sat in its stage, and how similar deals ended.

A small example shows the difference, using four deals from the startup’s pipeline that an AI model has scored.

Deal	Value ($k)	Stage	Stage probability	Model score
A	24	Proposal	45%	18%
B	18	Proposal	45%	71%
C	15	Negotiation	72%	81%
D	32	Demo	22%	9%

The stage method values these four deals at $36.7k and the model values them at $32.1k, but the reason matters more than the total. Deal A has had no reply for three weeks and only one contact, so the model marks it down, while Deal B has several stakeholders and legal review already under way, so the model marks it up. A manager can act on that before the quarter ends.

Adoption is already wide. In 2026, 87% of sales organizations used some form of AI, and 51% of sales leaders using AI said disconnected systems slow their AI work. That is the main failure mode: a model trained on messy, incomplete CRM data repeats the mess with more confidence. You need a large set of closed deals with activity data, and the method suits mid-size and larger teams that already run a clean CRM. Keep a human review in the loop, because a model cannot know about a merger announced yesterday.

Data Needed, Best Fit and Typical Failure for Each Method

Seen together, the methods sort into three groups: those that read the past, those that read the pipeline and those built on plans or judgment. The table sums up what each needs and how it tends to go wrong.

Method	Data you need	Best fit	Typical failure
Historical run-rate	1 to 2 years of sales	Stable, mature businesses	Misses turning points
Time series	2 to 3+ years by month	Many small transactions, clear seasons	Lags growth; poor with lumpy deals
Opportunity stage	CRM stages and win rates	B2B with a defined process	Ignores timing; inflated by stale deals
Length of sales cycle	Created and close dates	B2B with steady cycles	Mixed deal sizes distort ages
Lead-driven	Lead source to close tracking	Inbound and marketing-led teams	Conversion rates drift unnoticed
Regression	Several years plus driver data	Larger firms with an analyst	Overfits small data sets
Rep commit	Weekly rep calls in the CRM	Almost every B2B team	Sandbagging or optimism
Top-down market sizing	Market count, share, deal size	Pre-revenue and new markets	Share assumption is a guess
Bottom-up capacity	Headcount, quota, attainment, ramp	Annual plans and hiring	Assumes demand exists
AI and predictive	Large deal and activity history	Mid-size and larger teams	Garbage in, confident garbage out

The chart shows all nine numeric forecasts for the same quarter. The spread from lowest to highest is $82k, about 16% of the hypothetical actual, and that is with every input invented to be tidy.

One pattern in the chart is worth copying. The simple average of all nine forecasts is $523k, about 2% above the made-up actual, which is closer than seven of the nine methods on their own. Forecasters call this triangulation. Most experienced sales leaders run a pipeline method and a history method side by side, then dig into the gap between them. If the weighted pipeline says $558k and the run-rate says $538k, the question to ask is which deals explain the extra $20k.

How to Measure Forecast Accuracy With MAPE and Bias

A forecast you never score cannot improve. In one vendor study, only 43% of sales leaders forecast within 10% of the result. To know where your team stands, track two numbers every period.

MAPE (mean absolute percentage error) measures how big your misses are. For each period, take the forecast minus the actual, ignore the sign, divide by the actual and average the results. Bias measures the direction of the miss. Add up the forecast minus the actual across all periods, keeping the sign, and divide by total actual sales. A positive bias means you overforecast, and a negative one means you underforecast.

The next table scores two teams with almost the same MAPE over six quarters. All figures are in $k and invented.

Quarter	Actual	Team A forecast	Team A error	Team B forecast	Team B error
1	455	480	+5.5%	430	−5.5%
2	470	505	+7.4%	505	+7.4%
3	445	470	+5.6%	420	−5.6%
4	575	600	+4.3%	610	+6.1%
5	470	500	+6.4%	445	−5.3%
6	520	560	+7.7%	560	+7.7%
MAPE			6.2%		6.3%
Bias			+6.1%		+1.2%

On MAPE alone the two teams look the same, and only the bias shows the real difference. Team A overforecast in every quarter, by $180k in total over $2,935k of actual sales. That is a process problem with a clear fix: find which stage or which reps are too optimistic and adjust. Team B’s misses go both ways and mostly cancel out, which is ordinary noise. If you want to check the arithmetic on your own numbers, our percentage calculator handles the percentage error for each period.

Two cautions keep these numbers honest. MAPE breaks down when actual sales are tiny or zero, because you divide by the actual, so use it for teams or regions with real volume. And score the forecast you made at a fixed point, such as the start of the quarter, not the one you updated in the last week, when most deals are already decided.

Picking a Forecasting Method by Company Stage

The right method depends more on how much data you have than on how advanced you want to look, so a team of three founders does not need a regression model and a 200-person sales floor should not run on gut calls.

• Pre-revenue or pre-seed: You have no sales history, so use top-down market sizing and cross-check it with bottom-up capacity. Ask how many customers your current team can actually win and onboard each month. Put both in your financial model and your investor pitch deck, and show the assumptions openly.
• First customers, roughly the first year of sales: Use lead-driven math and simple rep commit. Your conversion rates will be noisy, so update them monthly. A simple spreadsheet is enough at this stage.
• Growing team with a repeatable process: Add opportunity stage forecasting with win rates from your own deals, and adjust for sales cycle length. Run weekly forecast calls with written category rules. Start tracking MAPE and bias by rep.
• Scaling company with several years of data: Add time series or regression for the longer view and capacity planning for hiring. Consider AI deal scoring once your CRM data is clean. Compare at least two methods every quarter and keep a record of which one wins.

Forecasting With No History, Commit vs Best Case and Accuracy

What Is the Most Accurate Sales Forecasting Method?

No single method wins for every company, although methods that use the open pipeline adjusted for deal age tend to work well for B2B teams with clean CRM data, while time series suits businesses with many small sales. The most reliable approach is to run two methods built on different data, such as a pipeline method and a history method, and investigate the gap.

What Are the Main Types of Sales Forecasting Methods?

They fall into two broad groups. Qualitative methods rely on judgment, such as rep commit calls and expert opinion, while quantitative methods rely on data and include run-rate, time series, regression, pipeline methods and AI models. Most teams combine both, with numbers setting the baseline and judgment adjusting it.

What Is a Good Sales Forecast Accuracy?

There is no single industry standard. Only about 43% of sales leaders land within 10% of the result, so a team that does that quarter after quarter is ahead of most. Track your own MAPE and bias over at least four quarters, then work to shrink both. A steady bias in one direction is easier to fix than random misses.

How Do You Forecast Sales for a New Business With No History?

Combine a top-down estimate (market size times a realistic share) with a bottom-up estimate based on how many customers your team can reach and convert each month. Where the two disagree, trust the bottom-up number. Replace both with lead-driven and pipeline forecasts once you have a few months of real sales.

What Is the Difference Between Commit and Best Case?

A Commit deal is one the rep is confident will close in the period and is willing to be held to. A Best Case deal could close if things go well but carries real risk, such as a missing signature or an unapproved budget. Teams usually forecast Closed plus Commit, with some share of Best Case added based on past conversion.

How Often Should a Sales Forecast Be Updated?

Many B2B sales teams review the current quarter’s forecast every week. Forecasting takes sales leaders an average of 4.9 hours a week. Annual plans are usually revisited each quarter. Score accuracy against the forecast made at a fixed point, such as the first week of the quarter, so late updates do not hide early mistakes.

Gartner, Gong and Salesforce Research on Forecast Confidence

• Gartner: Less Than 50% of Sales Leaders and Sellers Have High Confidence in Their Organization’s Forecasting Accuracy (February 2020)
• Gong: Reality of Forecasting Report press release (July 2022)
• Salesforce: State of Sales Report announcement (2026)
• Salesforce Ben: Forecast Categories in Salesforce
• Outreach: Sales Forecasting Methods