Forecast Accuracy: Bias, MAPE, WAPE, Lead Time and a HubSpot Snapshot Procedure

Sales forecast accuracy is reported as one percentage, which can be correct while saying nothing about how well anyone forecast. A team that submits $1,020,000 and closes $1,000,000 reports 98% accuracy, although each of its six account executives missed by $30,000 or more and the misses cancelled. At week 10 the same team scores 91.9% on a measure where misses cannot cancel, because seven-tenths of the quarter had already closed. Forecasting skill shows only in the error on revenue still open when the forecast was made, and measuring it needs a number no documented HubSpot report returns for a past date: the amount closed at that moment.

This article defines bias, absolute percentage error and its weighted form, and shows why the mean absolute percentage error breaks on small and zero actuals. Its scope, and the absence of any published benchmark, come first. After a sample quarter, the lead-time effect and HubSpot's scores and records, it gives a recording procedure, its costs and the symptoms of a misleading figure.

A forecast error is forecast minus actual, so a positive error means the forecast was high. Bias is the signed error summed across a team's reps; accuracy is the size of the error regardless of sign. Lead time, or forecast horizon, runs from the cutoff, the moment a forecast is fixed for measurement, to the end of the period it predicts. In HubSpot, a forecast submission is a user's estimate of what they will close in the month or quarter, per the forecast tool documentation.

Scope and Dated Boundaries

This article covers point forecasts of closed-won revenue, made by reps for a month or quarter. It excludes demand planning, interval forecasts and the calibration of deal-level win probabilities. How the forecast itself is produced, by stage weighting, forecast categories or a statistical model, belongs to revenue forecasting.

Tiers, settings, retention limits and API behaviour are as documented in October 2026, and the linked pages are authoritative over this text.

The sample below demonstrates arithmetic, not typical values. No source consulted publishes a benchmark for good sales forecast accuracy, and a benchmark quoted without its measure and lead time could not be compared with any team's figure.

Four Measures of Forecast Error

The measures differ in whether opposite errors cancel and how reps are weighted. The error, APE and MAPE follow section 5.8 of Hyndman and Athanasopoulos, Forecasting: Principles and Practice, which writes the error as actual minus forecast and defines neither a bias ratio nor WAPE; reversing the sign here flips only the direction of bias.

MAPE carries the defects the textbook lists for percentage errors: an extreme value near a zero actual, none at zero, and a heavier penalty on forecasts above the actual than below it. An under-forecast misses by 100% at worst while an over-forecast has no ceiling, so a rep scored on MAPE pays less for forecasting low. WAPE, dividing once by the team total, weights each rep's percentage error by that rep's actual, counts a rep who closed nothing in full, and stays defined whenever the team closed anything.

Sample Quarter: Six Reps and Two Submissions

The team below is sample data, built so that every figure can be recomputed; it describes no client. Six reps submitted a forecast at the end of week 1 of a 13-week quarter and updated it at the end of week 10. Nothing had closed by the end of week 1, so each week-1 submission forecast the whole quarter.

At week 1 the errors for reps A to F are −$30,000, +$42,000, −$60,000, −$33,000, +$45,000 and +$56,000. They sum to +$20,000, a bias of +2.0% and the 98% that the team total reports. The absolute errors sum to $266,000, a WAPE of 26.6%, because $143,000 of over-forecasts and $123,000 of under-forecasts net to $20,000 only while their signs are kept.

MAPE cannot be computed for the team, because Rep F's actual is zero, and dropping that row removes the team's largest over-forecast, $56,000. Without Rep F, MAPE is 74.0%, and Rep E supplies 60 of those points: a $45,000 miss on a $15,000 actual is 300%. That weights Rep E's miss at twelve times Rep C's larger miss of $60,000, which is 25% of $240,000. Dropping Rep E as well leaves 17.5%, so one quarter yields MAPEs of undefined, 74.0% and 17.5%, against one WAPE of 26.6%.

Lead Time and the Closed Share of a Submission

The week-10 submissions score a WAPE of 8.1% on $81,000 of absolute error, a gain that does not come from forecasting. A HubSpot submission estimates everything a user will close in the period, so $700,000 of the team's $1,021,000 week-10 call was revenue already booked.

The finding is an identity any team can check on its own figures. Against the whole period, a submission's weighted error equals its weighted error on revenue still open at the cutoff, multiplied by the open share of the period's actual. Each rep's dollar error is the same either way, since subtracting the closed amount from both forecast and actual leaves their difference unchanged; only the denominator moves.

Divided by the $300,000 still open at the cutoff, the same $81,000 is 27.0%, against 26.6% at week 1: on open deals the reps were no more accurate at week 10. The 8.1% is that 27.0% multiplied by the 30% still open. Bias scales alike, from +2.1% of the quarter to +7.0% of open revenue, so whole-period figures from different cutoffs are not comparable. Open-portion WAPE has its own limit: its denominator shrinks toward zero as a period closes, making it unstable late in a period as MAPE is near a zero actual. It is therefore read at team level and at an early, fixed cutoff.

HubSpot Accuracy Scores and Forecast Records

The forecast tool needs Sales Hub or Service Hub Professional or Enterprise, an assigned Sales or Service seat and Forecast permissions, and changing its settings needs Super Admin, per the setup documentation. Tracking the accuracy of forecasts scores a user or team, per pipeline and period, as one minus |Actual − Forecast| ÷ Actual, floored at zero, beside Closed won and a Submission average, the mean of every submission for the period. The AI projections page names that average as the forecast term. The score therefore blends lead times and lets a rep's opposite errors cancel: Rep A's $270,000 and $318,000 average $294,000, scoring 98.0% against $300,000. The documentation does not say what a rep with no closed won is shown. Beta AI projections, whose accuracy table requires a monthly forecast cadence, are scored on the same formula and averaged across the five issued on days 1, 7, 14, 21 and 28 of each month. HubSpot forecasting examines the submission score further.

Submissions are kept: the forecast tool's History tab shows past submissions and their deals up to one year back. A read-only Forecasts API in HubSpot's 2026-09 API version, the current release, returns every submission action, newest first, from /forecast-submissions/2026-09/events, filterable by occurredBefore. The documentation does not state how long those events are retained. Editing the forecast period resets forecastable revenue goals and forecast submissions for all pipelines, according to the setup documentation, which does not say whether past history survives.

The closed amount at the cutoff is the one number no documented report returns as it stood. The forecast tool's Closed Won column sums closed-won deals as they stand now. Deal property history keeps the inputs, stage, amount, close date, owner and pipeline, as up to 20 timestamped revisions per property, API edits included. Replaying five exports deal by deal rebuilds the figure within that limit; the procedure below avoids the work and the limit by capturing it at the cutoff. The Historical snapshots report counts deals at points in time by stage, amount or forecast category. Its documentation does not say whether it can sum Amount by owner at a past date; where it can, it replaces the cutoff file.

Snapshot Procedure for a HubSpot Forecast

1. Settle the forecast period under Settings → Objects → Forecast, per the setup documentation, before the first measured quarter, and leave it: an edit resets goals and submissions in every pipeline.
2. Set the cutoff, for example 12:00 on the seventh day of each quarter, as the due time of a submission schedule: under Sales → Forecast, a Super Admin opens Actions → Set submission schedule and sets its start date, due time, repeat, teams and reminder, per the forecast tool documentation. The submission standing then is the forecast of record.
3. Under Reporting → Reports → Create report → Single-object report → Deals, build three single object reports as tables of Deal name, Deal owner, Amount, Close date and Forecast category, each filtered to the submissions' pipeline unless they cover All pipelines: closed-won deals with a close date this quarter, closed-won deals with a close date last quarter, and open deals with a close date this quarter. Single object reports, on Professional and Enterprise, query the live CRM database; custom report builder reports refresh about every two hours. A custom forecast type replaces Amount and Close date with its own properties. With deal splits on, a Sales Hub Enterprise feature, a rep's forecast counts only the rep's split share, as the Forecasts API documentation describes; that share comes from a custom report on Deal splits, exported by hand at the cutoff. A multi-currency account uses Amount in company currency.
4. Share all three as monthly recurring emails with an XLSX attachment at 14:00 on the seventh. Report data is captured two hours before the send, so the files hold closed won at the cutoff, the frozen actual for the quarter just ended, and the open pipeline behind each call. Recurring emails go only to HubSpot users, and only single-object, event-based and customer journey reports attach as files.
5. Record each rep's submission of record from the History tab, or from the events endpoint with occurredBefore set to the cutoff, keeping the latest event per owner and period whose hs_deal_pipeline_id matches the reports' pipeline, null for All pipelines.
6. Compute, per rep, the error as submission minus actual and the open actual as actual minus closed won at the cutoff, then team bias, WAPE and open-portion WAPE.
7. Verify the first quarter before reporting it: trace one rep's three figures to an event dated before the cutoff, the cutoff file and the frozen file, and confirm that no rep's open actual is negative. A negative value means closed won in the cutoff file later shrank for that rep: a deal was reopened, lost, reduced, reassigned to another owner, moved to another pipeline or redated outside the quarter; the histories of dealstage, amount, hubspot_owner_id, pipeline and closedate on that deal show which.

The Business Case for Accuracy Measurement

Measurement buys a forecast whose error is known before the next one is used: a finance team holding several quarters of week-1 WAPE can plan against a range, and rep-level bias that persists across quarters shows which calls to adjust before they are summed.

Beyond the forecast tool's subscription and seats, it costs a discipline held at a fixed moment: every rep submits by the cutoff, three files arrive each month, and someone owns the spreadsheet. The larger cost is behavioural, because a forecast can be made accurate by making it late. A deal held out of the call until signature cannot miss, and the lead time that made the forecast worth having goes with it.

The case is strongest where each rep closes enough deals that errors reflect judgement rather than one contract's timing, and where the forecast feeds hiring plans or external guidance. It is weakest for a team closing a few large deals a quarter, where one slipped contract moves every measure, and for new reps with small actuals.

Symptoms of a Misleading Accuracy Figure

Accuracy improves every quarter while win rates and deal sizes stay flat. Submissions drifting later produce this with no change in forecasting, because the whole-period figure rises with the closed share; a fixed cutoff and open-portion WAPE separate the two.

The team total lands within 2% while individual reps miss by 25%. The total measures bias, in which over-forecasts and under-forecasts cancel, and WAPE reported beside it shows the size of the misses.

HubSpot's accuracy score for a rep sits near 100% while the rep's first call of the quarter missed widely. The score averages every submission in the period, so late resubmissions that already contain closed revenue pull the average toward the actual; scoring the submission of record removes the effect.

Calls fall below results after accuracy enters performance reviews. Percentage measures charge less for forecasting low, so a persistent negative bias is the visible trace. Bias reported beside WAPE at the fixed early cutoff counts each deal withheld until signature as an under-forecast, and keeping the percentage score out of rep reviews removes the reward for withholding.

In Summary

A single forecast accuracy percentage conflates bias, the signed error in which offsetting misses cancel, with the size of the misses, which WAPE measures in dollars against what closed. MAPE is dominated by small actuals and broken by zeros.

Lead time decides what any of these figures means, because a submission's whole-period error is its open-portion error multiplied by the open share. HubSpot's own score averages every submission and so blends lead times, and no documented report returns closed won as it stood at a past cutoff. Scoring the submission that stood at one fixed cutoff, every quarter, gives the only trend that separates better forecasting from later forecasting.