Every founder running a sales team has the same Monday. Open the forecast. Compare it to last week. Try to work out which deals are real and which are hope with a close date attached. The number at the bottom is presented as a projection. It is closer to a mood.
The data on this is worse than most operators assume. Gartner puts only 7% of companies at 90% or better forecast accuracy. Fewer than 25% of sales organizations clear even 75%. And just 45% of sales leaders express high confidence in their own organization's forecast — which means the majority are presenting a number to their board that they privately do not trust.
The reflex is to blame the reps. Tighten the pipeline reviews. Add a stage-gate. Insist on better hygiene. That reflex has been running for two decades and the numbers have not moved, which is usually a sign the diagnosis is wrong.
The forecast is made of the wrong material
Walk backwards from the number. The forecast is a sum of deals. Each deal carries an amount, a stage, and a close date. Every one of those three inputs is entered by the person whose compensation depends on how it looks.
The amount is an estimate. The stage is self-assessed. The close date is, in most CRMs, the date the rep hopes it lands — frequently the last day of the quarter, because that is the default the system offered. None of these are observations. They are opinions, stored in a database, and then added up as though the addition made them factual.
Rep-submitted forecasts still run roughly plus or minus 30%. That is not a forecast. It is a range wide enough to contain both a good quarter and a layoff.
Underneath that sits a data problem that makes the whole exercise fragile. Roughly 76% of CRM records are less than half complete. If the primary contact is missing you cannot tell whether the rep has access to the buyer. If the industry field is blank you cannot benchmark the deal against anything. The model inherits every gap in the system beneath it.
of sales organizations rate pipeline management and forecasting as a strength — despite spending an average of $1,866 per sales rep per year on CRM and sales automation software.
Gartner
The market got harder while the tooling stayed the same
Forecasting on opinion was survivable when deals closed predictably. They no longer do. Median B2B win rates fell to 19% in 2024, down from 23% in 2022. Sales cycles have stretched roughly 22% longer since 2022, to a median of about 84 days. Longer cycles mean more opportunities for a deal to change shape after the rep last touched the record — and more distance between what the CRM says and what is true.
This is the same structural problem we described in your pipeline isn't broken, your revenue system is, viewed from the forecasting end rather than the conversion end.
Why this hits smaller companies hardest
A $500M company with a RevOps team has people whose entire job is making the number trustworthy. A company between $1M and $20M ARR with three to fifteen reps has none of that. The founder is the RevOps function, doing it at 11pm in a spreadsheet, using judgment built from having personally closed most of the early deals.
That judgment is real and it is genuinely valuable. It is also unscalable, undocumented, and it degrades exactly as the team grows past the point where the founder is in every deal. The forecast gets less accurate at precisely the moment the consequences of being wrong get larger — when you are hiring against it, raising against it, or committing to a board against it.
The cost of being wrong is not abstract. Forecast $10M and hire fifteen reps against it, and land at $7M, and you have added fixed cost with no revenue underneath it. Forecast conservatively and land well above it, and you left capacity on the table you could have used. Both errors are expensive and both come from the same place.
What actually moves accuracy
The organizations that forecast well did not lecture their way there. They changed what the forecast is made of.
- Score the signals, not the stage. Engagement recency, stakeholder count and seniority, stage-progression velocity, close-date integrity, and deal-size movement are all observable without asking anyone how they feel. Deals that have slipped their close date twice behave differently from deals that never have — and the CRM already knows which is which.
- Weight deals by data completeness. An opportunity missing a decision-maker contact and a next step is not worth the same as a fully-instrumented one. Discount it in the roll-up rather than pretending the fields are filled.
- Segment before you average. Inbound SMB deals and outbound enterprise deals have different win rates, cycle lengths, and slippage patterns. Blending them produces an average that describes nothing you actually sell.
- Calibrate reps against their own history. A rep who has overstated commit by 20% for six straight quarters is not being dishonest. They are being consistent, and consistency is correctable — apply the factor rather than the lecture.
- Track accuracy as a metric in its own right. Forecast variance, measured quarter over quarter by rep and by segment, is the only way to know whether any of this is working.
The honest version of the fix
None of this requires a RevOps hire, and that matters, because at this company size a RevOps hire usually is not justifiable. Gartner expects 75% of high-growth B2B companies to be running a formal RevOps model, and companies with the function report roughly 36% higher revenue growth and up to 28% more profitability. Those numbers are real — and they are also mostly measured at companies that could afford to build the function.
For a company with a dozen reps, the practical version is narrower: instrument the signals you already generate, stop treating rep confidence as data, and make the forecast reproducible by someone other than the founder. That is the gap StageFlow was built to close, and the RISQ™ score is that logic made operational — it ranks deals by probability of closing this period based on observed behavior rather than stated confidence.
But the tool is downstream of the decision. The decision is to stop accepting a number you do not believe. If your forecast variance is running above 20%, you do not have a forecasting problem to solve with more discipline. You have a data problem wearing a forecast costume.