Forecasts That Hold: Building a Number You Can Defend
A forecast is a claim about the future backed by evidence. We break down the process, the evidence, and the review rhythm that keep the number honest.
Forecast accuracy problems are rarely about the model. They are about the inputs, the incentives, and the conversation. If the rep believes a commit is a promise they will be punished for missing, they will sandbag. If they believe it is a wish, they will inflate. Both produce a number the board cannot use.
Define the categories precisely
Commit, best case, and pipeline mean different things in different teams, and often different things to two managers in the same team. Write the definitions down in terms of evidence, not confidence.
- Commit: the customer has agreed to buy, the paperwork path is known, and the remaining steps are scheduled.
- Best case: a realistic path to close this period exists and at least one material step is unconfirmed.
- Pipeline: qualified and active, without a credible path to close this period.
Once the categories are evidence based, a forecast review becomes a conversation about facts rather than a negotiation about feelings.
Ask for the evidence, every time
The single highest leverage change most teams can make is to ask one consistent question of every commit deal: what has the customer done that makes this a commit. Not what they said in a friendly call. What they did. Sent a security questionnaire. Introduced procurement. Agreed a signature date in writing.
Confidence is not evidence. Ask what the customer did, not how the call felt.
Separate the roll up from the judgement
There are two useful numbers and teams often conflate them. The bottom up roll up is what the deals say. The judgement number is what leadership believes after applying history: typical slippage rates, seasonal patterns, and known risks. Report both, and track the gap between them over time. A widening gap is an early signal that deal level inputs are drifting.
Use history rather than instinct
Your own conversion history by stage, by segment, and by source is more predictive than a manager's gut, and it is sitting in your CRM. If commit deals in your mid-market segment historically close at eighty percent, that ratio belongs in the judgement number.
Review deals, not spreadsheets
A forecast call that reads the roll up out loud creates no new information. Spend the time on the deals that would change the outcome: the largest commits, anything that moved category this week, and anything with no customer activity in ten days. This is where an AI system helps a manager most, by flagging the deals whose evidence does not match their category before the meeting starts, so the hour is spent on judgement rather than discovery.
- Start with slipped and downgraded deals, and ask what changed.
- Test every commit against the evidence definition.
- Look for silence: deals with no customer contact recently are usually already lost.
- Close with the actions that would protect the number, with owners and dates.
Make accuracy safe to report
If the culture punishes an early warning more than a late miss, the forecast will always be late. Leaders who thank a rep for flagging a slipping deal in week three get earlier information for the rest of the year. Leaders who react badly get surprises in week twelve. This is a cultural lever, and it moves accuracy more than any tooling change.
Track your own accuracy
Measure forecast accuracy by period, by manager, and by segment, and review it as a standing item. Once accuracy is a visible metric with a named owner, the incentive to inflate drops sharply, because the inflation shows up in someone's number a quarter later.
