Your sales forecast is a decision, not a number

Every forecasting conversation I've been in eventually turns into an argument about accuracy. Are we going to hit the number? Is the model too optimistic? Why did last quarter's forecast miss by nine percent? Fair questions. But after enough years staring at forecasts, at the scale of tens of millions of records in one job, and a fast-moving sales pipeline in another, I've come to think accuracy is where teams spend their attention and rarely where the value is.

Here's the uncomfortable version: a forecast that is perfectly accurate and changes no one's behavior is worthless. And a forecast that's a little off but causes the right person to act a week earlier can be worth a great deal. The number is not the deliverable. The decision it enables is. Once I started treating the forecast as a decision aid rather than a scoreboard, most of what I'd been optimizing turned out to be the wrong thing.

The scoreboard trap

It's easy to slip into treating the forecast as a grade. Leadership wants a number, the number gets compared to actuals, and the whole apparatus starts optimizing for "were we right." That instinct quietly warps everything. You reward confident point estimates over honest ranges. You punish the analyst who flagged a risk that didn't materialize more than the one who stayed quiet and got lucky. You spend modeling effort shaving a percentage point off historical error instead of asking whether anyone did anything different because of the forecast.

The question I now ask first is not "how accurate is this" but "what decision does this change?" If the honest answer is "none", if leadership was going to run the quarter the same way regardless, then the forecast is expensive theater, no matter how tight the error bars. The forecasts I'm proudest of weren't the most precise. They were the ones that made someone reallocate a rep's time, pull forward a conversation, or stop pouring effort into a deal that was quietly already lost.

Calibration beats precision

If the goal is a better decision, then how a forecast expresses uncertainty matters more than the single number it lands on. A confident point estimate, "we'll close $4.2M", invites exactly one response: believe it or don't. A calibrated forecast, "most likely around $4M, but there's a real chance it's $3.2M if these three deals slip, and here's why", invites a decision. It tells you what to watch and what to do if the world moves.

Calibration is the property that when I say I'm 70% confident, I'm right about 70% of the time. It sounds like a modeling nicety. It's actually the thing that determines whether people trust the forecast enough to act on it. A team learns very quickly whether your numbers bluff. The first time a "sure thing" collapses, every future forecast gets mentally discounted, and a forecast that gets discounted is a forecast that doesn't drive decisions. I'd rather be honestly uncertain and believed than falsely precise and ignored. Precision you can't back up isn't rigor; it's a liability dressed as confidence.

The signal is spoken before it's logged

Most forecasting models run on what the CRM knows: stage, amount, close date, age. But anyone who's sat close to a sales floor knows the CRM is a lagging record of reality. A deal is at risk long before someone downgrades its stage: the champion goes quiet, a competitor shows up, the language on the last call turns hedgy, a new stakeholder appears asking hard questions. The risk was spoken weeks before it was ever logged.

I spent a good while building a workflow that treated call transcripts as a signal source rather than an archive, pulling the leading indicators of a slipping deal out of the conversation itself and surfacing them while there was still time to act. The lesson generalized: the best forecast doesn't just extrapolate the fields in the system, it incorporates the signal that hasn't reached the system yet. If your forecast can only see what's already in the CRM, it's telling you about a reality that's already a few weeks old.

Decision latency is the metric no one measures

Here's the number I wish more forecasting teams tracked: the time between a signal existing and a decision being made on it. Call it decision latency. Everyone obsesses over data freshness and model accuracy and no one measures this, yet it's often what actually determines whether the analytics changed the outcome.

A forecast delivered the morning of the close call is mostly history. Accurate, maybe, but too late to act on. A rough read delivered three weeks earlier, when there was still room to intervene on the deals that mattered, is worth far more even if it's less precise. When I think about improving a forecasting function, I now spend at least as much energy on when the insight lands and whether it prompts an action as on the error rate. Shrinking the gap between signal and decision usually beats shrinking the gap between forecast and actual.

What a good forecast actually does

Put it together and a genuinely useful forecast does four things a number on a slide can't. It tells you not just what's likely but how confident to be and what would change the picture. It draws on the signal that hasn't hit the system yet, not just the tidy fields that have. It arrives early enough to act on. And it points attention: this handful of deals is where the quarter will actually be decided, look here. That last part matters more than people expect: the value of forecasting is often less about the aggregate number and more about reducing what the team has to pay attention to down to the few things that move it.

None of this makes accuracy irrelevant. A forecast that's wildly off loses trust fast, and trust is the whole game. But accuracy is the floor, not the ceiling. The ceiling is influence: did the forecast make the team's decisions sharper, faster, and more honest than they'd have been on gut alone.

So when someone asks me how good a forecast is, I've stopped answering with a percentage. I ask what changed because of it. If the answer is "we saw the risk in time and did something," that forecast did its job, even if it missed the final number by a hair. And if the answer is "nothing, but it was very accurate," then we built a beautiful scoreboard and forgot to play the game.