Right too late: the 90-day forecast problem
On this page
Every sales forecast converges on the truth. The $126M projection eventually admitted it was $46M. The model that wandered all quarter landed on the correct $26.69M on the last day. Accuracy was never in question; timing was. Most sales forecasting software is graded on where the number ends, and it should be graded on when the number becomes actionable, because a forecast that turns accurate after the quarter is decided is a press release.
The structural cause sits in the buyer, and the third-party record documents it. 6sense's Buyer Experience Report of more than 4,000 buyers puts buying cycles near 10 months, three quarters long, with 94 percent of buying groups picking a favored vendor before first contact. A 90-day forecast samples the visible tail of a ten-month process, so early-quarter estimates built only on CRM stages are guesses wearing decimals. The signals that would make them real live upstream of the pipeline.
This playbook names the convergence curve and the field standard for early-quarter accuracy. Then it walks the operating cadence, day 15 through day 60, that a revenue team can run next quarter, and it closes with the honesty rules that keep executives believing the number.
The convergence curve
Every forecast in your company's history shares one property: by the last day of the quarter, it was right. Bookings equal bookings. The instructive object is the path the number took to get there, and that path is a curve: projection accuracy plotted against the weeks of the quarter. Every curve ends at 100 percent. The entire value of a forecasting system lives in the shape before the end.
The convergence curve (illustrative pattern)
Projection accuracy by week of quarter: the shape to manage, drawn with schematic values
Illustrative pattern only; values are schematic, drawn to show the shape. The field standard, from a RevSure working session on forecast calibration: 85 to 95 percent accuracy in the early days of the quarter.
The chart is a schematic, drawn to show the shape rather than any customer's data. The two curves share a destination and differ in one economic property: the stretch of quarter where the number is simultaneously accurate and actionable. Everything a revenue team can do about a quarter has a use-by date. Reallocating spend toward converting segments needs weeks to land. Standing up coverage against a weak cohort needs longer. Pulling deals forward needs sales cycles that still have slack in them, and resetting the plan needs to happen before finance re-forecasts the company around your miss. A projection that converges in week two intersects all of those windows. A projection that converges in week twelve intersects none of them, and its terminal accuracy is a consolation prize.
So the KPI for any forecasting motion, human or software, is the convergence date: the day of the quarter on which the projection settles within 10 percent of the eventual actual and stays there. Teams track forecast accuracy obsessively and almost never track when they achieved it. The date is the number that decides whether the accuracy was worth anything.
Four metrics operationalize the curve. The convergence date itself, tracked quarter over quarter as a number to manage down. Early-window accuracy, meaning how close the day-15 projection landed to the eventual actual. The revision-explanation rate, the share of projection changes that shipped with a named cause. And trigger lead time, the days between the first below-plan warning and quarter end, which is the size of the window every intervention has to fit through. A forecasting review that tracks these four is reviewing the system. One that tracks final accuracy alone is reviewing the scoreboard after the game.
Three quarters from the field record
The field record shows what late convergence looks like from the inside.
A customer success software company watched its pipeline model land on the correct number, $26.69M, on the last day of the fiscal quarter. Ninety days of estimates, and the one that was right arrived when the quarter was already a fact. Nobody reallocates anything on day 90. The model's final grade was perfect and its operational contribution was zero.
A healthcare benefits company lived the more expensive version. A projection showed roughly $126M for the quarter. Actual pipeline came in around $46M. With seven days left, the projection still showed a 50 percent increase. Sit with that detail for a moment: inside the final week, the system was still promising expansion.
The 90-day problem in the field record
Sources: RevSure working sessions, descriptor level · published founder account
The damage compounds past the quarter, because executives who get surprised at day 85 stop believing day-15 warnings too, including the accurate ones. This is the forecasting chapter of the wider pattern we documented in the attribution trust gap: numbers that fail late teach leaders to distrust the whole system early. The regression end-state appears verbatim in our working sessions. A revenue operations leader at a software testing company, speaking from inside a company that owns forecasting tooling: "we do all of our forecasting right now off a spreadsheet, which is not scalable." The spreadsheet is what surviving your software looks like.
if it were me, I'd almost prefer a conservative number than forecasting over what actually happens.
a revenue operations leader at a compliance software company
That preference is rational risk management. An inflated projection defers action until action is impossible, which is the expensive direction to be wrong. A conservative number that triggers an unnecessary intervention costs some redundant pipeline generation. The costs are asymmetric, and the operators who have been burned all lean the same way.
What produced the overshoot is worth naming, because most sales forecasting software shares the ingredients. Stage-weighted CRM values compound two optimisms: the rep's, encoded in amounts and close dates, and the stage model's, encoded in weights calibrated on better quarters. Neither input registers what the buying group is actually doing, and with buying cycles near 10 months, most of the evidence about a deal's health sits upstream of the CRM stage field. A projection computed from stages alone is a poll of the sales team's mood, taken with decimals.
The standard worth canonizing
Buried in one of our calibration sessions is the sentence we think should hang on the wall of every revenue operations team.
if you're hitting 85 to 95% accuracy in the early days of the quarter, that is exactly what you want
from a working session on forecast calibration
Read it as a trade, because it is one. You give up 5 to 15 points of precision to move the number 70 days earlier, and the exchange rate is spectacular. An 88 percent number at day 15 funds decisions all quarter. A 95 percent number at day 88 funds a retrospective. Deepinder Singh Dhingra, RevSure's founder, has a line about enterprise buying that applies squarely here: "We think buyers want speed. But what they're really paying for is certainty." Certainty has a delivery date, and the executives buying it price it by when it arrives.
The standard carries an honest boundary. 85 to 95 percent at day 15 is a target for calibrated systems running on reconciled data, and a team can only claim it if it also measures it, which means logging every projection and grading it against the landing. A forecast graded only by the people who issued it converges on flattery, and the healthcare benefits quarter is what flattery costs.
The cheapest evidence available is a backtest. Before trusting any system's curve, run it backward: feed it the data as it stood on day 15 of each of the last four quarters and compare its projections to the landings you already know. The exercise costs nothing but compute. It produces a real convergence curve from your own history, and it separates vendors who talk about early accuracy from systems that demonstrate it. It also calibrates your own expectations, because your team's manual forecast can be backtested on exactly the same terms.
Projections, forecasts, and other load-bearing words
The calibrated teams in our record run a language discipline that looks pedantic and works.
A revenue operations leader at a contract lifecycle management company drew the line for us: "that's why we don't call our projections forecasts... projections, for directional nature." A forecast, in that team's usage, is a number leadership commits to and gets judged against. A projection is an instrument reading that improves as evidence accumulates. Collapsing the two is how a directional day-10 estimate ends up on a board slide as a commitment, and how the whole apparatus takes the blame when the estimate moves.
The same team projects volume before value, deal counts before dollar sums, because early-stage amounts are the noisiest field in the CRM. That practice, and what it does to coverage math, gets its own treatment in the coverage delusion.
A CMO at a field service management software company adds a third discipline: plural models. "I'm also comfortable in a world where I have three different versions of the same thing, and I take a weighted average... an ensemble approach." An ensemble does two useful things a single oracle cannot. It shows disagreement, which is information about uncertainty, and it fails gracefully, because separate models rarely share a blind spot.
The shared thread is epistemic honesty as an operating practice: every number arrives labeled with what kind of number it is and how much to trust it today. Executives do not punish uncertainty. They punish discovering it late.
The operating cadence: day 15, 30, 60
The published case for what an early number buys: a team caught a $3M revenue deficit 90 days before the quarter would have surfaced it, with enough runway to reallocate spend and save hundreds of thousands of dollars. The mechanics were unromantic. The warning existed and was believed, and it arrived inside the window where budget still moved. Ninety days buys a correction; nine days buys an explanation. Here is the cadence that manufactures that outcome on purpose, as numbered practices with the failure mode each one replaces.
- Day 15: publish a graded projection with a trigger attached.
Less effective: the first serious number appears at the mid-quarter business review, assembled by hand the night before and presented without an error bar, then defended like a commitment.
Recommended: by day 15, a projection graded against the 85 to 95 percent standard, decomposable to the cohorts and deals behind it, published with an explicit trigger: a projected landing below plan opens the gap play immediately. The gap play itself is ordinary work: reallocate spend toward converting cohorts, expand generation where readiness is high, escalate the at-risk segment to sales leadership, reset expectations with finance while resetting is cheap. What makes it work is calendar position. In the $3M case, the entire miracle was timing.
- Day 30: publish the delta, deal by deal.
Less effective: the projection moves and nobody says why, so every revision spends trust the system cannot afford.
Recommended: a standing day-30 review of what changed since day 15 and which deals or cohorts caused it. Convergence gets checked and the trigger re-armed. Revisions arrive with causes attached. A number that explains its own movement keeps its credibility while it moves.
- Day 60: lock the landing zone, shift to next quarter.
Less effective: the final month spent renegotiating the current quarter's number nightly while next quarter's pipeline quietly starves.
Recommended: by day 60 the projection should sit inside its band, and the review's center of gravity moves to next quarter's coverage and readiness, because with a ten-month average buying cycle, next quarter was being decided months ago. Quarter-end heroics get replaced by the dullest sentence in revenue operations: the landing matched the day-60 number.
The cadence compresses to a page.
| Day | The question | The artifact | The trigger |
|---|---|---|---|
| 15 | Where does the quarter land? | Graded projection, decomposed by cohort and deal | Projected landing below plan opens the gap play |
| 30 | Why did the number move? | Delta review with causes named | Off-track convergence escalates to CRO |
| 60 | Is the landing zone locked? | Band check, next-quarter coverage review | A slip past the band resets the plan with finance |
The cadence fails by hand. A software testing company we work with measured its manual version: about 90 minutes every Monday evening for one dashboard update, and roughly 100 person-hours per quarter for board reporting, both collapsed to minutes once computed on reconciled data. A cadence that costs 90 manual minutes per update runs weekly at best and gets skipped in crunch weeks. The version that survives is computed continuously and merely reviewed by humans. That is the layer RevSure occupies: the Predictive AI Engine computes projections and pipeline readiness from the Full Funnel Data Graph, and the numbers decompose to the deals and signals underneath, so the day-30 question, why did it move, has an answer a human can inspect.
The macro trend is running toward the teams that operate this way. Gong Labs' December 2025 study, built on 7.1 million opportunities across 3,613 companies, found 70 percent of enterprise revenue leaders now trust AI for regular business decisions, and teams that use AI heavily generate 77 percent more revenue per rep. Trust at that level gets built one graded, explained projection at a time, and it gets spent all at once the first time a day-85 surprise arrives.
Label the noise, keep the trust
The cadence has one integrity rule, and teams break it in the first week of the quarter. First-week projections are noisy. The models have a handful of days of in-quarter signal. Deal amounts are placeholders, and cohort behavior is an extrapolation from quarters that may resemble this one. The honest move is a label: directional and low confidence, firming by day 15. A marketing analytics advisor in our sessions put the trust mechanics in civilian terms: "you go on a first date, it sounds great, but it's only after you've known someone a while that you start to have some confidence." Confidence is earned on a schedule, and claiming it early is how systems lose it permanently. A team that presents day-5 noise as precision teaches its executives to discount day-45 precision as noise. Executives do not remember which week the number was wrong. They remember that it was wrong.
Noise has a mirror image that deserves equal suspicion. A projection that never moves is saying nothing about the quarter, and the same advisor has the line for that failure too: "when you send the same value, it's the same as sending nothing." Day-15 and day-60 numbers that agree to the dollar have usually been negotiated rather than computed. The healthy signature is movement with explanations attached, inside a narrowing band. A forecast is an instrument, and instruments that never register anything are broken in the quiet direction.
The concession this playbook owes: standing up a calibrated projection is real implementation work, and the failure mode is real too. An operations leader at a contract lifecycle management company told us, mid-implementation: "we got too focused on configuring the system correctly, but we are not really able to use the system in our day-to-day right now." Calibration needs quarters of your own actuals, reconciled identities, honest stage data, and a logged record of past projections before the day-15 number deserves the confidence this playbook asks you to place in it. Plan the first quarter as the calibration quarter, label its outputs accordingly, and grade the system from quarter two.
What to do next quarter
- Compute last quarter's convergence date. Find the day your projection settled within 10 percent of the actual landing. That date is the baseline this playbook exists to move.
- Adopt the standard. 85 to 95 percent accuracy at day 15, written into the RevOps charter as the grading rule for every projection source, software included.
- Split the vocabulary. Projections are directional instruments; forecasts are commitments. Rename the artifacts, then defend the distinction in every review.
- Label early-quarter confidence explicitly, and let the label expire on a published date.
- Stand up the day 15, 30, and 60 reviews with trigger thresholds written down before the quarter starts.
- Re-grade your sales forecasting software on convergence timing and decomposability. A tool that is right at day 90 and silent about why is a scoreboard. You are buying an early-warning system, so evaluate the warning, and evaluate the date it arrives.
Where this comes from
Built from our work inside enterprise GTM teams: indexed, verbatim working sessions with marketing, RevOps, and revenue leaders, quoted here at descriptor level with permission discipline. The forecast-miss figures are single-company cases from those sessions, never market averages, and the convergence chart is a labeled schematic rather than measured data. The record skews toward B2B software and technology enterprises, and toward teams engaged enough with measurement to work with a revenue platform, which is the population you would expect to converge earliest. Mostly, they did not.
Frequently asked questions
Why are sales forecasts wrong until the end of the quarter?
Because most forecasts are built from rep judgment and stage-weighted CRM values, which are opinions that only harden as deals close. The forecast converges on truth exactly as the quarter runs out of room to act. Field cases include a pipeline model that landed on the correct $26.69M only on the last day of the fiscal quarter.
What is good sales forecast accuracy?
Grade accuracy against timing. A practitioner standard from RevSure's working sessions: 85 to 95 percent accuracy in the early days of the quarter is exactly what you want, because that number arrives while spend and coverage can still move. A 95 percent number on day 88 describes the quarter; it cannot change the quarter.
What is the difference between a projection and a forecast?
In field practice, a forecast is a committed number that leadership is judged against, while a projection is a directional estimate that improves as evidence accumulates. One revenue operations team refuses the word forecast entirely: 'we don't call our projections forecasts... projections, for directional nature.' The label sets expectations, and expectations protect trust.
What should sales forecasting software be measured on?
Convergence timing: the day of the quarter when the projection settles within 10 percent of the eventual actual. Beyond that, ask whether early-quarter numbers carry confidence labels, whether the model projects volume separately from value, and whether any miss can be traced to the specific deals and signals that caused it.