A sales forecast is a commitment about how much revenue will close in a period, built from the state of individual deals. Most forecasts miss not because the math is wrong but because the inputs are opinions: reps calling their own deals with their own optimism, adjusted by a manager's gut, adjusted again by a VP's haircut. This playbook replaces the chain of opinions with a chain of evidence.
Why Forecasts Miss
Forecast misses trace back to three sources, in roughly this order. First, deal state fiction: the CRM says stage four, but the economic buyer has never been on a call and the paper process has not started. Second, unexamined optimism: reps carry deals at commit because they need them, not because the evidence supports them. Third, sandbagging in the other direction: reps hide upside to beat a low number, which is just as damaging to capacity planning as missing.
The scale of the problem is measurable. In pilot data from prospects' legacy CRM environments, a small sample so far, roughly 32 percent of committed deals slipped.
Notice what all three failure sources have in common: the forecast is only as honest as the deal data underneath it, and the deal data is only as honest as what actually happened in the deal. Fixing the forecast starts two layers down.
Step 1: Define Categories by Evidence, Not Confidence
Forecast categories fail when they are defined by feeling: commit means "I'm sure," best case means "probably." Redefine each category as a checklist of verifiable facts. A deal qualifies for commit only when the evidence exists in the record, not when the rep feels good.
Commit. Economic buyer engaged and confirmed, decision criteria and process documented, paper process started or scheduled with dates, champion has verbally confirmed the decision. Every item verifiable in the deal record.
Best case. Qualification complete through MEDDPICC or SPICED, decision process known, but at least one commit criterion is still open, and the open item is named.
Pipeline. Qualified opportunity with defined pain and engaged stakeholders, but the decision process is not yet mapped.
Omitted. Everything else, including deals with a close date this quarter and no economic buyer contact. A close date is a field, not evidence.
The category definitions do the cultural work. When commit requires evidence, the argument in the forecast call shifts from "do you believe the rep" to "is the evidence there," which is an argument someone can actually win.
Step 2: Choose Your Forecasting Method Deliberately
There are three basic methods, and most teams should run two of them against each other.
Stage weighted forecasting
Multiply each deal by a probability tied to its stage. Simple, and honest only if stages reflect buying progress rather than rep activity, and if the probabilities come from your own historical conversion rates rather than defaults someone set years ago. Recalibrate the weights quarterly from actuals.
Category roll up forecasting
Sum commit, add a historically derived fraction of best case, and track the result weekly. This is the method the evidence based categories above are built for, and for most B2B software teams between 10 and 200 reps it is the workhorse.
AI driven forecasting
Model based forecasts score each deal from signals humans weigh badly: engagement recency, stakeholder breadth, stage velocity versus similar won deals, qualification completeness. The catch is that a model is only as good as the structure it reads. AI forecasting on top of manually maintained CRM data learns your team's data entry habits, not your buyers' behavior. AI forecasting on top of automatically captured, consistently structured deal data is a different instrument entirely, which is why forecast intelligence is native to Dreamhub rather than a separate tool bolted onto a CRM it cannot trust. Measured across our deployments, 71 percent of deals Dreamer flagged as at risk at least two weeks before the end of the quota period went on to slip or close lost.
The practical setup: run category roll up as the human forecast and an AI forecast beside it. When the two diverge on a deal, that divergence is your inspection list.
Step 3: Run an Inspection Cadence, Not a Reporting Cadence
A forecast call where reps read numbers aloud is a reporting cadence. An inspection cadence examines the deals where the evidence and the call disagree.
Weekly, 30 minutes per team. Review only the deltas. New commits, deals that left commit, and deals where the AI score and the rep call diverge. Never review the whole pipeline in this meeting.
Per deal, three questions. What evidence changed since last week, what is the next verifiable event with a date, and what would have to be true for this to slip? If the answers live in the record already, the meeting takes minutes.
Monthly. Compare forecast snapshots to actuals by rep and by category. Reps who systematically sandbag or overcommit are visible within two cycles, and the pattern, not the miss, is the coaching conversation.
Step 4: Measure the Forecast Itself
Treat forecast accuracy as a first class metric with its own targets. Track week over week: commit accuracy (what percent of week one commit actually closed), slip rate (deals that moved out of the quarter after being committed), and pull through from best case. Healthy teams land commit accuracy above 90 percent by mid quarter.
If your week one commit routinely closes at 60 percent, you do not have a forecasting problem. You have a qualification problem wearing a forecasting costume.
FAQ
What is the most accurate sales forecasting method?
For B2B software teams, evidence based category roll up cross checked against an AI deal score. Each method catches the other's blind spots: humans miss patterns, models miss context.
How do you stop reps from sandbagging?
Define categories by verifiable evidence so a deal that meets commit criteria must be committed, and track per rep forecast accuracy in both directions. Sandbagging survives on ambiguity.
What forecast accuracy should we target?
Within plus or minus 5 percent of the final number by week one of the last month of the quarter, and commit accuracy above 90 percent. Early stage teams with small deal counts should expect wider variance and focus on slip rate instead.
Do we need a separate forecasting tool?
Only if your CRM cannot hold evidence based categories, snapshot history, and deal level intelligence natively. A separate forecasting layer reading a manually maintained CRM inherits every fiction in it; fixing the data layer removes the need for the tool.