Every quarter ends with two numbers on the screen: the one you called and the one that closed. Most revenue leaders track the second obsessively and the first barely at all, which is how a team can miss three forecasts in a row without anyone being able to say whether it is getting better or worse.
Forecast accuracy is the discipline of measuring that gap the same way every time. It is unglamorous, and it is the only way to tell whether your forecast has a math problem, a process problem, a people problem or a data problem. Each needs a different treatment, and most teams reach for the wrong one first.
What is sales forecast accuracy?
Sales forecast accuracy is the percentage gap between the revenue you forecast for a period and the revenue that closed in it, measured from a fixed point in time. A team that called $1.0M in week two and closed $0.92M was 91.3% accurate on that call. The fixed point matters: a forecast that only converges in the final week told leadership nothing they could plan around.
Four choices define the measurement. Hold all four constant, or the trend line means nothing.
- The snapshot. Pick the call your plans depend on, usually the first-month call for a quarterly plan, and save it. HubSpot, Salesforce and Pipedrive all recalculate the live forecast as deals move, so a call you did not save is a call you cannot grade.
- The level. Measure the company, each team and each rep. Errors cancel on the way up: two reps who miss by 20% in opposite directions produce a perfect company number and two useless forecasts.
- The revenue type. Keep new business separate from renewals. Renewals are predictable, and blending them in flatters the accuracy of the part that is hard.
- The window. One quarter is an anecdote. Look at four or more periods before you call anything a trend.
The sales forecast accuracy formula, step by step
The standard formula is forecast accuracy = 1 − |actual − forecast| ÷ actual. It ignores direction, so pair it with forecast bias = (forecast − actual) ÷ actual, which keeps the sign. Accuracy tells you how far off you were. Bias tells you which way you lean, and the lean is what points to the cause.
Divide by actual, not by forecast, and say so on the dashboard. Both conventions exist, and two people using different ones will argue about the same quarter for an hour. Here is one team’s year, worked through.
| Quarter | Week-2 call | Closed | Accuracy | Bias |
|---|---|---|---|---|
| Q1 | $1,000,000 | $920,000 | 91.3% | +8.7% |
| Q2 | $900,000 | $950,000 | 94.7% | −5.3% |
| Q3 | $1,100,000 | $980,000 | 87.8% | +12.2% |
| Q4 | $1,050,000 | $1,000,000 | 95.0% | +5.0% |
| Average | 92.2% | +5.2% |
Read the two right-hand columns together. Average accuracy of 92% sounds respectable. Three of the four quarters were over-forecast, though, and an average bias of +5% says this team calls high as a habit. That is a different problem from random noise, with a different cause and a different conversation.
What is a good forecast accuracy?
We treat landing within 5% of actual, at the call you plan from, as strong for a B2B sales team, and within 10% as workable. Most teams fall short of both. Xactly’s 2024 Sales Forecasting Benchmark Report, a survey of 405 North American sales and finance leaders fielded in March 2024, found only 20% forecast within 5% of actuals, and 43% miss by 10% or more.
| Typical gap between forecast and actual | Share of respondents |
|---|---|
| Less than 5% | 20% |
| 5% to 9% | 38% |
| 10% to 19% | 36% |
| 20% to 25% | 6% |
| More than 25% | 1% |
Source: Xactly, 2024 Sales Forecasting Benchmark Report, n=405 sales and finance leaders, combined. Figures as published; they sum to 101% through rounding.
The same report found 4 in 5 sales and finance leaders missed a quarterly forecast in the past year, and more than half missed two or more times. The split by function is the more telling number. About two-thirds of finance leaders (66%) said they are usually off by less than 9%, while 52% of sales leaders said their forecasts are off by 10% or more. Same survey, two functions, two very different relationships with the number. We take that up in why sales reps are bad at forecasting.
Confidence trails accuracy. Gartner’s State of Sales Operations survey, published in February 2020, found only 45% of sales leaders and sellers had high confidence in their organization’s forecasting accuracy. It is six years old. We have not found a newer survey of comparable standing.
You will see a claim that only 7% of sales organizations reach 90% forecast accuracy, credited to Gartner. We could not trace it to a published Gartner document, so we do not use it. The same goes for any “2026 benchmark” without a named survey behind it. Ask for the survey.
Accuracy and bias are different problems
Accuracy measures the size of the miss; bias measures its direction. A team can post respectable average accuracy while missing high every quarter, and that pattern points to a specific cause. Chronic over-forecasting usually means stale pipeline and stages that advance on seller activity. Chronic under-forecasting usually means sandbagging, or deals that never enter the CRM until they are nearly done.
| Pattern | What it usually means | Where to look in the CRM |
|---|---|---|
| Over-forecast, steady | Pipeline carries deals that are no longer real | Open deals with past close dates and no recent buyer activity |
| Under-forecast, steady | Reps hold deals back or create them late | Deals created and closed in the same period; late-quarter stage jumps |
| Swings both ways | No shared definition of commit | Category changes per rep, week over week |
| Accurate total, inaccurate reps | Errors cancelling at the top | Rep-level accuracy and bias, side by side |
The vocabulary in that table, sandbagging, happy ears, commit and best case, gets a precise definition in our forecasting glossary. Use the definitions in the forecast call. A team that argues about what “commit” means will never agree on whether it was right.
Why is my sales forecast wrong?
Most sales forecasts miss for four reasons, roughly in the order I come across them: pipeline that is no longer real, stage definitions that describe seller activity instead of buyer commitment, rep judgment bent by optimism or caution, and CRM data plumbing that makes reports disagree. Forecasting software sits on top of all four and inherits every one of them.
1. The pipeline is not real
Open deals with close dates that passed weeks ago. Deals with no amount. Deals nobody has touched since the last quarterly push. Each one inflates coverage and the weighted number at the same time. Xactly’s respondents named poor CRM or pipeline data hygiene among the reasons forecasts miss, and 66% named reporting systems that cannot access historical CRM or performance data as the most common roadblock to an accurate forecast.
2. Stages measure your process, not the buyer’s
“Demo completed” is something your rep did. “Buyer confirmed budget and a decision date” is something the buyer did. Stages built on seller activity advance whether or not the buyer moved, and the probabilities attached to them are rarely your own. HubSpot’s default pipeline assigns 20% to 90% by stage for every new portal, and Pipedrive sets stage probability to 100% by default until someone changes it. The HubSpot version of this problem gets its own paper: why your HubSpot forecast is always wrong.
3. Reps call what they feel
A rep forecasting a deal sees the champion’s enthusiasm, not the base rate for deals at that stage. Optimism produces happy ears; caution about quota produces sandbagging. Then there is the incentive layer, where some reps enter what the room wants to hear, which we cover in why good reps enter bad data.
4. The plumbing disagrees with itself
Duplicate contacts split a buyer’s activity across two records. Deals without an associated contact drop out of contact-based reports. Two dashboards filter on different date properties and show different quarters. In the same February 2020 Gartner survey, only 47% of respondents believed their organization had high-quality data. See why HubSpot reports don’t match and why HubSpot duplicates keep coming back.
The order to work in
Work from measurement to data to definitions to behavior. Start tracking accuracy and bias at a fixed snapshot, so you have a baseline. Clean the open pipeline next, because no method survives fake deals. Then rewrite stage exit criteria around buyer actions and calibrate probabilities to your own win rates. Coach rep judgment last, once the inputs deserve trust.
The sequence matches what separates strong teams in the largest recent survey. Salesforce’s State of Sales, 7th edition, a survey of 4,050 sales professionals across 22 countries published in February 2026, found high-performing sales organizations prioritize data hygiene at 79%, against 54% of underperformers. That is a correlation, not proof. It matches what I’ve seen running revenue teams: the forecast tends to get better after the data does, rarely before.
- Forecast accuracy checklist
- Save the call you plan from every period, by rep and in total.
- Report accuracy and signed bias side by side, four or more periods back.
- Close out or re-date every open deal whose close date has already passed.
- Require an amount and an associated buyer contact on every deal past your first qualified stage.
- Rewrite each stage exit as something the buyer did, not something the rep did.
- Replace default stage probabilities with your own stage-to-close win rates.
- Define commit and best case in writing, with the evidence each requires.
- Review rep-level bias monthly and coach the direction, not only the size.
If you run HubSpot, start with our free HubSpot audit. For the full read on HubSpot, Salesforce or Pipedrive, the AeolusGTM CRM Diagnostic runs a read-only scan of your records and pipeline and shows which open deals are missing the numbers a forecast needs, which have quietly gone cold and whose close dates keep sliding, each traced to the records behind it.
Every forecast is a claim about the future made with data about the past. You can keep arguing about the claim every quarter, or you can make the data worth arguing from. Only one of those gets easier with practice.