How should a 2027 RevOps team score rep forecast accuracy?

The two scoring models a 2027 RevOps team actually chooses between

Almost every forecast accuracy program collapses into one of two designs, and the choice determines what behavior you get for the next eight quarters.

Model A — the commit ratio. One number per rep per quarter: actual closed revenue divided by the revenue that rep committed at a fixed point in the quarter, expressed as a percentage. A rep who committed $1.0M and closed $940K scores 94%. It is trivially explainable, computes off two CRM fields, and every AE understands it in one sentence. RevOps can stand it up in a reporting layer in an afternoon.

Its weakness is that it is a *signed* ratio, and signed errors cancel. A rep who closes 120% of commit in Q1 and 80% in Q2 averages to exactly 100% over the two quarters while having been off by a fifth in both directions. On a trailing-four-quarter average, a chronically volatile rep can look identical to a metronome. If you run Model A, you must never average the raw ratios — you average the *absolute deviation from 100* and report the signed mean as a separate bias figure.

Model B — the error score. Instead of a ratio, you compute absolute percentage error per commit, then aggregate. The forecasting literature calls the simple version MAPE (mean absolute percentage error); the version RevOps teams want is weighted absolute percentage error, where each deal's error is weighted by its dollar size before aggregation. Model B answers "how wrong, in dollars that matter" rather than "how wrong, on average across deals of wildly different consequence." It also decomposes cleanly: total error splits into bias (the persistent directional lean) and dispersion (the noise around that lean), and those two failures need completely different coaching.

Model B's cost is comprehension. "Your weighted absolute percentage error was 11.4 with a positive bias of 6.2" lands worse in a one-on-one than "you closed 94% of what you committed." Teams that adopt Model B without a translation layer get compliance, not behavior change.

The hybrid most mature teams land on. Report Model A's ratio as the *conversational* number in the rep-facing view, and compute Model B underneath it for the RevOps and manager view. The rep sees "94% of commit, second consecutive quarter under-delivering." The manager sees the weighted error, the bias sign, and the trailing dispersion. One vocabulary for coaching, one instrument for diagnosis. This costs you a second calculation and a second dashboard tab, and it is worth both.

A third variant worth knowing about, though it is heavier than most teams need: probabilistic scoring. If your reps forecast *probabilities* per deal rather than a single committed dollar figure — "70% likely to close this quarter" — you can score them with a proper scoring rule such as the Brier score, which rewards calibration rather than confidence. A rep whose 70% deals close about 70% of the time is well calibrated even if any individual call was wrong. This is the technically correct way to score judgment under uncertainty, and it is also the way most likely to be abandoned in month three because the sales floor does not think in probabilities. Consider it only if your CRM already captures per-deal rep confidence as a first-class field and your managers have the appetite for a calibration curve.

How to decide between them without a six-month debate

The decision is not about statistical elegance. It is about four practical constraints: pipeline shape, headcount per manager, data hygiene, and what you intend to do with the number.

Pipeline shape decides more than anything else. If your average rep closes 25 or more deals a quarter with a tight size distribution — velocity, SMB, transactional mid-market — the simple commit ratio is genuinely sufficient. Errors average out across enough deals that the ratio is statistically meaningful on its own. If your reps close three to eight deals a quarter with a 20:1 spread between the smallest and largest, the unweighted ratio is nearly useless: one enterprise slip dominates the quarter and the score measures luck. That is where weighting stops being sophistication and becomes a requirement.

Deals per rep per quarter is your sample size. Below roughly six closed opportunities in a quarter, a single-quarter accuracy figure carries almost no signal, and you must extend to trailing-four-quarter before you draw any conclusion or open a coaching conversation. Above twenty, single-quarter scores become usable as an early warning.

Data hygiene gates everything. Both models depend on a commit that was actually locked and stored — not overwritten in place. If your CRM lets a rep edit the commit field retroactively with no history, you have no denominator and no program. Snapshotting the commit into an immutable record at the lock moment is the first engineering task, ahead of any scoring math.

What you'll do with it sets the required precision. If the score is purely a coaching prompt, the ratio is fine. If it will inform manager MBO payout, hiring plans, or a number the CFO repeats to a board, you need the weighted, bias-decomposed version, because someone will eventually challenge it and "we averaged the ratios" is not a defensible answer.

One decision the flowchart does not make for you: what to do about reps with fewer than four quarters of tenure. New hires have no trailing window and their early commits reflect ramp, not judgment. The standard handling is to compute their score but suppress it from any comparison, band, or intervention trigger until quarter four, while still walking through the arithmetic with them monthly so the habit forms during ramp rather than after.

The concrete numbers behind each model

Nothing below is a benchmark from a study — these are the default configuration values that make each model behave sensibly, and the arithmetic that shows why.

Commit lock timing. Lock the commit at the end of week one of the quarter. Later locks let the rep observe early-quarter reality before committing, which improves the score without improving the forecast — you are measuring hindsight. Some teams lock at day one; the practical problem is that quota, territory, and comp plan changes often are not finalized until the first week is over. End of week one is the compromise that most operating calendars support. If you want an in-quarter view as well, snapshot again at the end of month one and month two and score those separately; do not overwrite the week-one number.

Bands for the commit ratio. A workable starting set for a new-business team: 95–105% green, 85–95% or 105–115% yellow, 75–85% or 115–125% orange, outside that red. Renewal or customer-success-owned books should run tighter — 92–108% green — because renewal outcomes are structurally more predictable and a wide band there hides real problems. Enterprise books with three to five deals a quarter should run wider, or should skip single-quarter banding entirely and band only on the trailing-four-quarter figure. Tune the bands once against a full year of historical data before you publish them; if 70% of your reps land red on day one, your bands are wrong, not your team.

Dollar weighting. The weight for each committed deal is its committed dollar value divided by the rep's total committed dollars that quarter. Worked example: a rep commits $1.0M across five deals — one at $600K and four at $100K each. The $600K deal closes at $480K, a 20% miss; the four small deals all close at commit. Unweighted, the rep is accurate on four of five deals and the naive per-deal average is 96%. Weighted, the rep delivered $880K against a $1.0M commit — 88%, with a weighted absolute error of 12 points. The weighted figure is the one that matches what the CFO experienced.

Stage weighting. If you score in-quarter commits and not just the week-one lock, weight late-stage commits more heavily than early-stage ones — a common configuration is 1× for deals in discovery or validation at commit time, 2× for proposal, 3× for negotiation or verbal-agreed. The logic: a rep who mis-forecasts a deal sitting in legal review has made a much worse call than one who mis-forecasts a deal that was still in discovery when they committed it. Skip stage weighting entirely if your stage definitions are not enforced, because you will just be weighting by how aggressively each rep advances stages.

Time decay across the quarter. If you snapshot weekly, weight each snapshot's error by how late in the quarter it was taken — a linear ramp from roughly 0.6× in week one to 1.0× at the midpoint to 1.5× in the final weeks is a reasonable default. This rewards reps who surface bad news early and correct, and penalizes the rep who holds a number until week eleven and then drops it. Report the decay-weighted figure as a single trailing number, but keep the weekly granularity available, because *when* a rep's forecast breaks is the most actionable coaching signal you have.

Bias versus dispersion. Compute both. Bias is the mean signed deviation over the trailing four quarters; dispersion is the standard deviation of those signed deviations. A rep at −12% bias with 3-point dispersion is consistently optimistic by a predictable amount — that is the easy case, because you can correct for it in the roll-up while you coach it out. A rep at 0% bias with 22-point dispersion is a coin flip, and their roll-up contribution is worthless even though their average looks perfect. Different problems, different interventions, and the single ratio cannot distinguish them.

Outlier handling. Define an outlier threshold as a percentage of the rep's quota — deals above roughly 25–30% of a rep's quarterly number qualify. Score the non-outlier base separately from outlier performance and report both. Reps carrying enterprise books get structurally punished by any single-figure accuracy metric, and they will disengage from the program within two quarters if you do not carve this out.

Comp exposure. Keep rep variable compensation entirely free of the accuracy score. The moment accuracy pays, sandbagging pays. Attach accuracy to management instead: a slice of the frontline sales manager's MBO or bonus component tied to team-level accuracy, and a similar or slightly larger slice for the CRO tied to company-level accuracy, are the standard placements. The manager is the correct accountable party because the manager owns the roll-up, the deal inspection, and the coaching.

Implementation details and sequencing

Do not launch this as a scored program. Launch it as measurement, run it silently for a quarter, then turn on bands, then turn on consequences. Teams that ship all three at once get one quarter of hostility and a permanently poisoned metric.

Weeks 1–4: instrument, do not score. The engineering work is small and specific. Add an immutable commit snapshot record — rep, quarter, snapshot date, committed amount, and the deal-level composition behind it. Do not use an editable currency field on the user or forecast object; use a child record written at lock time and never updated. Confirm the actual-closed side is unambiguous: decide now whether you score on closed-won ARR, net new ACV, or booked TCV, and whether mid-quarter territory transfers follow the rep or the territory. Write that definition down in one paragraph and get the CRO to approve it in writing, because every dispute in month six will be a definitional dispute.

Weeks 5–8: backfill and calibrate. Rebuild the last four quarters from whatever commit history exists — even partial history tells you the shape of your distribution. Compute what every rep's score *would have been*. This is where you set bands: pick thresholds that put roughly the top quartile in green and the bottom quartile in orange or red, so the bands are informative rather than uniformly flattering or uniformly damning. Show this backfill to two or three senior managers privately and let them tell you which reps the numbers get wrong; every miss is usually a definitional edge case you have not handled yet.

Weeks 9–12: quiet first quarter. Score live, share with managers only, and have each manager walk their reps through their own number in a one-on-one — no bands published, no comparisons. Frame it explicitly as "this is how we will measure, here is yours, nothing hangs on it this quarter." Collect every objection and fix the legitimate ones before the next quarter starts.

Quarter two: bands and cadence on. Publish team-level accuracy in the monthly forecast committee. Keep individual scores private between rep and manager permanently — a public individual leaderboard converts a diagnostic into a game within one quarter, and reps optimize the metric instead of the forecast. Establish the coaching cadence: monthly review for anyone outside green on the trailing window, quarterly for everyone else.

Quarter three: consequences and MBO. Only now attach manager MBO. By this point the numbers have survived a full cycle of challenge, the definitions are stable, and managers have had two quarters of practice reading them.

The coaching split. Chronic over-committers — reps closing well under commit quarter after quarter — need deal-level retrospectives, not pep talks. Pull their last two quarters of committed deals that slipped and find the shared trait: committed before economic buyer engagement, committed on a verbal from a champion with no procurement path, committed on a renewal-plus-expansion that was really two decisions. The fix is a named entry criterion for what a rep is allowed to commit, not an exhortation to be realistic.

Chronic under-committers are rarer and treated more gently, but they are a real accuracy failure: they deny the company visibility into upside, which distorts hiring and capacity planning as badly as a miss does. The conversation is diagnostic — is this fear of being held to a number, is it a comp plan with an accelerator that rewards holding deals, is it a manager who publicly punishes misses? Sandbagging is almost always a rational response to something structural, and you fix the structure.

The systemic read. Once a quarter, pivot accuracy by stage rather than by rep. If most of the team is red on deals that were in negotiation at commit time, you do not have a rep problem — you have a stage definition that means nothing, or a win-rate assumption that no longer holds. This is the single highest-value output of the whole program, and it is invisible if you only ever look at the individual leaderboard. RevOps owns making this pivot exist and bringing it to the forecast committee; nobody else in the org is positioned to see it.

What to avoid outright. Do not tie rep variable pay to the score. Do not publish individual rankings. Do not punish an honest downgrade — if a rep's reward for surfacing a slipped deal in week four is a hostile meeting, every subsequent slip surfaces in week twelve. Do not compute a rep's score across a territory change without splitting the quarters. And do not let the score become the forecast: the accuracy program measures judgment quality, it does not replace deal inspection.

Related questions

When exactly should the commit be locked?

End of week one of the quarter, snapshotted to an immutable record. Later locks let reps observe early-quarter results before committing, which inflates the score without improving the forecast. Snapshot again monthly for in-quarter trend, but never overwrite the week-one value.

Should rep compensation ever depend on the accuracy score?

No. Paying reps for accuracy makes sandbagging profitable and corrupts the metric immediately. Attach accuracy to sales manager and CRO MBO instead — they own the roll-up and the coaching, and their incentive to inflate is much weaker.

How do you score reps with only three or four deals a quarter?

Extend to a trailing-four-quarter window before drawing any conclusion, weight errors by deal dollars, and carve out deals above roughly 25–30% of quota as separately tracked outliers. Single-quarter scores at that sample size measure luck, not judgment.

Can AI forecasting tools replace rep-level scoring?

No — they complement it. Model-generated close probabilities give you a second opinion to compare against rep judgment, and the gap between them is itself a coaching signal. The judgment about what to do with a chronically optimistic rep stays human.

Should renewal teams use the same bands as new business?

Use the same method with tighter bands. Renewal outcomes are structurally more predictable, so a band wide enough for new business hides genuine renewal risk. Something around 92–108% green for renewals against 95–105% for new business is a reasonable starting split.

FAQ

What is the minimum data we need before we can score anything?

Three things: a committed dollar amount stored immutably at a fixed lock date, an unambiguous definition of actual closed revenue for the same period and rep, and a rule for what happens when accounts move between reps mid-quarter. If the commit field is editable in place with no history, you have no denominator, and instrumenting that snapshot is the entire first phase of the project. Everything else — weighting, decay, bias decomposition — is arithmetic on top of those three inputs and can be added later without re-platforming.

Should we use a ratio or an absolute error metric?

Report the ratio to reps and compute absolute error underneath for RevOps and managers. The ratio is the number people can hold in their head and discuss in a one-on-one. Absolute error is the number that does not let a plus-20% quarter cancel a minus-20% quarter into a fictitious perfect average. Running both costs one extra calculation and gives you a coaching vocabulary and a diagnostic instrument that do not fight each other.

How do we keep the score from being gamed?

Remove the payoff. Rep variable compensation stays entirely out of it, individual rankings are never published, and the commit is locked early enough that reps cannot wait for signal. Then watch the bias figure over the trailing window: persistent positive bias — closing above commit quarter after quarter — is the fingerprint of sandbagging, and it shows up clearly in the decomposition even when the headline ratio looks healthy.

What accuracy level should a team be aiming for?

Set the target from your own backfilled distribution rather than an external number. Compute what the last four quarters actually produced, set the green band so roughly the top quartile of reps sits inside it, and move the band as the team improves. An imported target is either trivially easy or demoralizingly impossible depending on your deal shape, and either outcome kills engagement in the first quarter.

Who owns this metric — RevOps or sales leadership?

RevOps owns the definition, the instrumentation, the calculation, and the systemic pivot that shows accuracy by stage rather than by rep. Sales management owns the coaching conversation and the outcome. That split matters: if RevOps runs the coaching, the program reads as surveillance from an adjacent function, and if sales owns the calculation, the definitions drift toward whatever produces a comfortable number.

How long before the program actually changes behavior?

Plan on three quarters. The first is silent measurement, the second adds bands and a coaching cadence, and the third attaches manager MBO. Behavior change tends to show up in the second and third quarters, once reps have seen their own trailing number twice and understood that an honest downgrade in week four costs them nothing while a surprise in week twelve costs them a difficult conversation.

Sources

• Hyndman, R. & Athanasopoulos, G. *Forecasting: Principles and Practice* — evaluating point forecast accuracy. https://otexts.com/fpp3/accuracy.html
• Wikipedia. *Mean absolute percentage error.* https://en.wikipedia.org/wiki/Mean_absolute_percentage_error
• Wikipedia. *Forecast bias.* https://en.wikipedia.org/wiki/Forecast_bias
• Wikipedia. *Brier score* — proper scoring rules for probabilistic forecasts. https://en.wikipedia.org/wiki/Brier_score
• Wikipedia. *Calibration (statistics).* https://en.wikipedia.org/wiki/Calibration_(statistics)
• Salesforce. *Sales forecasting overview.* https://www.salesforce.com/sales/analytics/sales-forecasting/
• HubSpot. *Sales Hub product documentation.* https://www.hubspot.com/products/sales
• Clari. *Revenue forecasting resources.* https://www.clari.com/
• Gartner. *Sales practice research and insights.* https://www.gartner.com/en/sales
• McKinsey & Company. *Growth, Marketing & Sales insights.* https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights

Related on PULSE

• [How do you coach a rep to improve forecast accuracy?](/knowledge/q13947)
• [Are 2027's forecast accuracy rates actually improving with AI, or are we just getting better at bias confirmation?](/knowledge/q16307)
• [Should I Hire a Fractional CRO If My Forecast Accuracy Is Below 50 Percent?](/knowledge/q15887)
• [How do you report broken lead routing when no dedicated RevOps hire yet and leadership only reviews forecast accuracy monthly on Dynamics 365?](/knowledge/q10208)
• [How do you report call recordings not tied to opps when no dedicated RevOps hire yet and leadership only reviews forecast accuracy monthly on Dynamics 365?](/knowledge/q10418)