Short answer
Sales forecasting methods fall into two groups. Time-series methods (run rate, moving average, seasonal year-over-year and exponential smoothing) project past revenue forward and suit repeat business. Pipeline methods (weighted stage probabilities, forecast categories and sales cycle length) forecast open deals and suit B2B new business. Most teams use one of each, test them against recent actual results, and trust the method with the smallest error.
- Split the forecast: time-series methods for existing customers, pipeline methods for new business.
- Backtest before you choose: forecast quarters you already know and compare the errors.
- In our example, last quarter's run rate missed by 10.9% on average; a seasonal method missed by 2.2%.
- Use your own stage probabilities and cut the odds on deals that have aged past your usual cycle.
- Report a base forecast and a best case, and explain the gap between methods.
On this page
This guide is for sales leaders, founders, finance and revenue operations people who need a forecast they can defend. It explains eight methods, runs one company's real-looking numbers through each of them, measures how wrong each would have been, and shows how to combine them. If you want a ready-made spreadsheet, the sales forecast template has one; this page is about choosing and testing the method.
The eight methods at a glance
| Method | How it works | Data you need | Best for | Weak when |
|---|---|---|---|---|
| Run rate | Next period equals the last period, sometimes plus a growth rate | One recent period | A quick check | Sales are seasonal or lumpy |
| Moving average | The average of the last few periods | A year or more of history | Steady repeat business | There's a trend or a seasonal peak |
| Seasonal year-over-year | The same period last year, scaled by recent growth | Two years or more of history | Seasonal businesses | The business has changed shape |
| Exponential smoothing | A weighted average where recent periods count more | A year or more of history | Repeat business that drifts | Single smoothing ignores trend and season |
| Weighted pipeline | Each open deal's value times its stage probability | Tracked opportunities with stages | B2B new business | Stages are used loosely |
| Forecast categories | Reps place deals in commit, best case or pipeline | A deal review habit | A check on the arithmetic | Reps sandbag or over-promise |
| Sales cycle length | Probability falls as a deal ages past your usual cycle | Created and closed dates | Long or variable cycles | Few closed deals to learn from |
| Regression | Sales modeled from drivers such as meetings, pipeline or price | Many periods of clean data | Larger teams with stable data | Drivers change or data is thin |
Single exponential smoothing has a fixed smoothing constant (often written α) between 0 and 1; the closer to 1, the more the latest period counts. Data with a trend and a seasonal pattern needs more: the NIST handbook describes the Holt-Winters method, which adds trend and seasonal equations. Excel's FORECAST.ETS function uses an exponential smoothing algorithm that detects seasonality automatically.
How to forecast with more than one method
- Split revenue into existing customers and new business. Repeat orders behave like a time series; new deals behave like a pipeline. Forecast them separately.
- Forecast existing revenue with two or three time-series methods. At least one should respect seasonality if your sales have a peak.
- Forecast new business from the pipeline. Weight each deal by your own stage probability, not the CRM default, then compare with the reps' commit and best case.
- Adjust for deal age. Cut the probability on deals that have sat longer than your usual cycle.
- Backtest. Use each method to forecast the last three or four periods you already know, as if you were standing at the start of each one, and record the error.
- Choose and reconcile. Use the method with the lowest error for each part, report a base forecast and a best case, and explain any gap of more than about 10% between methods.
- Record forecast against actual every period and repeat the backtest each quarter.
The error for one period is the absolute percentage error: the difference between actual and forecast, ignoring the sign, divided by actual. Average it across periods to get the mean absolute percentage error (MAPE). Lower is better.
Worked example: a wholesale bakery forecasts its fourth quarter
Illustrative example. The bakery, its revenue and its deals are fictional.
Copper Kettle Bakery Supply (copperkettle.example) sells bread and pastries to cafes, hotels and grocery stores. It's early October 2026, and the owner wants a forecast for the fourth quarter. Revenue from existing accounts, in thousands of dollars:
| Year | Q1 | Q2 | Q3 | Q4 |
|---|---|---|---|---|
| 2024 | 380 | 420 | 430 | 480 |
| 2025 | 410 | 455 | 470 | 520 |
| 2026 | 440 | 478 | 512 | ? |
The pattern is clear: a slow first quarter, a holiday peak in the fourth, and growth of a few percent a year.
Step 1: backtest four time-series methods. Each method forecasts the first three quarters of 2026 using only the data available at the time:
- Run rate: the previous quarter.
- Moving average: the average of the previous four quarters.
- Seasonal year-over-year: the same quarter last year, times the growth in the latest quarter (latest quarter ÷ the same quarter a year earlier).
- Exponential smoothing: smoothing constant 0.3, starting from the first quarter of 2024.
| Method | Q1 2026 forecast (error) | Q2 2026 | Q3 2026 | MAPE |
|---|---|---|---|---|
| Run rate | 520 (18.2%) | 440 (7.9%) | 478 (6.6%) | 10.9% |
| Moving average | 463.8 (5.4%) | 471.2 (1.4%) | 477.0 (6.8%) | 4.5% |
| Seasonal year-over-year | 444.2 (0.9%) | 488.3 (2.2%) | 493.8 (3.6%) | 2.2% |
| Exponential smoothing | 466.1 (5.9%) | 458.3 (4.1%) | 464.2 (9.3%) | 6.5% |
To check one: the seasonal forecast for Q1 2026 is Q1 2025's 410 × (Q4 2025's 520 ÷ Q4 2024's 480) = 410 × 1.083 = 444.2. Actual was 440, so the error is 4.2 ÷ 440 = 0.9%.
Step 2: forecast existing revenue for Q4.
| Method | Q4 2026 forecast |
|---|---|
| Run rate | 512.0 |
| Moving average | 487.5 |
| Seasonal year-over-year: 520 × (512 ÷ 470) | 566.5 |
| Exponential smoothing | 478.5 |
Three methods ignore the holiday peak and land between 478 and 512. The seasonal method, which had the lowest error, says $566,500.
Step 3: forecast new accounts from the pipeline. Copper Kettle's own history sets the stage probabilities: deals that reached discovery won 15% of the time, tasting done 35%, proposal reviewed 60% and negotiation 80%. Values are the revenue each deal would bring in the fourth quarter if won.
| Deal | Stage | Q4 revenue if won | Probability | Weighted | Rep's category | Age (days) |
|---|---|---|---|---|---|---|
| Hotel group | Proposal reviewed | $24,000 | 60% | $14,400 | Commit | 40 |
| Grocery chain, 10 stores | Tasting done | $36,000 | 35% | $12,600 | Best case | 75 |
| University dining | Proposal reviewed | $18,000 | 60% | $10,800 | Commit | 55 |
| Cafe group, 4 sites | Discovery | $6,000 | 15% | $900 | Pipeline | 20 |
| Caterer | Tasting done | $8,000 | 35% | $2,800 | Best case | 30 |
| Hospital cafeteria | Discovery | $12,000 | 15% | $1,800 | Pipeline | 90 |
| Restaurant group | Negotiation | $15,000 | 80% | $12,000 | Commit | 35 |
| Coffee chain, 12 sites | Tasting done | $20,000 | 35% | $7,000 | Best case | 120 |
| Total | $139,000 | $62,300 |
- Weighted pipeline: $62,300.
- Forecast categories: commit is $57,000 (the hotel group, university and restaurant group); commit plus best case is $121,000.
- Sales cycle length: Copper Kettle's won deals took a median of 70 days, and deals older than 105 days (one and a half times the median) won only 10% of the time. The coffee chain, at 120 days, drops from 35% to 10%, taking $5,000 off the weighted total: $57,300.
The age-adjusted pipeline ($57,300) and the reps' commit ($57,000) agree, which is the best sign a pipeline forecast can give.
Step 4: reconcile.
| Part | Method used | Q4 forecast | Other methods |
|---|---|---|---|
| Existing accounts | Seasonal year-over-year (lowest backtest error) | $566,500 | $478,500 to $512,000 |
| New accounts | Age-adjusted weighted pipeline, checked against commit | $57,300 | $57,000 to $121,000 |
| Total | $623,800 | Best case $687,500 |
Had Copper Kettle used last quarter's run rate for existing accounts, it would have forecast $569,300 in total and been caught short on staff and flour for the holiday peak.
How to choose a method
| Your situation | Start with | Check against |
|---|---|---|
| Many repeat orders, two or more years of history | Seasonal year-over-year or exponential smoothing with seasonality | Moving average |
| Few, large B2B deals with a tracked pipeline | Weighted pipeline with your own stage probabilities | Forecast categories |
| Recurring revenue plus new business | Split: time series for existing, pipeline for new | Both, reconciled |
| Long or variable sales cycles | Weighted pipeline adjusted for deal age | Forecast categories |
| New product or no history | Judgment from reps or experts, plus capacity (reps × expected deals) | Early actual results, monthly |
| Large team, stable data over many periods | Regression on pipeline and activity drivers | Weighted pipeline |
For building your own stage probabilities from closed deals, see the method in the sales forecast template. Stage definitions matter as much as the arithmetic: a weighted pipeline is only as good as the stages and exit criteria behind it.
Self-check
- Existing and new revenue are forecast separately.
- At least one time-series method handles seasonality, if you have a peak.
- Stage probabilities come from your own closed deals, recalculated each quarter.
- Deals older than your usual cycle are discounted.
- You've backtested each method on periods you already know and recorded the error.
- You report a base forecast and a best case, with the reason for any gap.
Common mistakes
- Using CRM default probabilities. They're placeholders. Your own win rates by stage will differ, sometimes a lot.
- One method for everything. Pipeline methods miss repeat orders; time-series methods miss a big new deal.
- Ignoring seasonality. A run rate or plain average will under-forecast every peak and over-forecast every trough.
- Never checking accuracy. Without forecast against actual, nobody knows which method to trust.
- Keeping stale deals. Old opportunities with unchanged probabilities inflate every pipeline method.
Next step: put your chosen method into a working spreadsheet with the sales forecast template.
Sources
- Single Moving Average (NIST/SEMATECH e-Handbook of Statistical Methods, 6.4.2.1), National Institute of Standards and Technology, accessed 26 October 2026
- Single Exponential Smoothing (NIST/SEMATECH e-Handbook of Statistical Methods, 6.4.3.1), National Institute of Standards and Technology, accessed 26 October 2026
- Triple Exponential Smoothing (NIST/SEMATECH e-Handbook of Statistical Methods, 6.4.3.5), National Institute of Standards and Technology, accessed 26 October 2026
- FORECAST.ETS function, Microsoft Support, accessed 26 October 2026
- Set up the forecast tool, HubSpot Knowledge Base, accessed 26 October 2026
Written by the Bebop team at Blackpearl Group. Last updated .
Questions people ask
What are the main sales forecasting methods?
The common ones are run rate, moving average, seasonal year-over-year, exponential smoothing, weighted pipeline (opportunity stage), forecast categories, sales cycle length and regression. The first four project past revenue forward and suit repeat business. The next three forecast open deals and suit B2B new business. Regression models sales from drivers such as pipeline and activity and needs plenty of clean data.
Which sales forecasting method is most accurate?
It depends on your data, so test rather than assume. Use each method to forecast three or four recent periods you already know, then compare the average error. In our illustrative bakery example, a seasonal year-over-year method missed by 2.2% on average and last quarter's run rate by 10.9%. For B2B new business, an age-adjusted weighted pipeline that agrees with the reps' commit is usually the strongest signal.
How do you calculate a weighted pipeline forecast?
Multiply each open deal's value by the probability of winning from its current stage, then add the results. Use probabilities from your own history: for each stage, the share of deals that reached it and went on to win over the last year. For example, a $24,000 deal at a stage that wins 60% of the time contributes $14,400. Discount deals that have aged past your usual cycle.
What is the difference between qualitative and quantitative forecasting?
Quantitative methods calculate the forecast from data: past revenue (run rate, moving averages, smoothing) or open deals (weighted pipeline, cycle length, regression). Qualitative methods rely on judgment: reps' forecast categories, manager calls or an expert panel. Most B2B teams use both, checking the reps' commit against the arithmetic and investigating when the two disagree.
How do you measure forecast accuracy?
Compare each forecast with the actual result. The absolute percentage error is the difference between actual and forecast, ignoring the sign, divided by actual. Average it over several periods to get the mean absolute percentage error (MAPE). Also check bias: if you're consistently over or under, the method or the inputs, such as stage probabilities, need fixing.