Built for revenue leaders · powered by Anthropic Claude
All posts
MethodologyJanuary 29, 20267 min read

Stop measuring forecast accuracy. Start measuring forecast precision.

Every sales leader claims to measure forecast accuracy. Almost nobody measures forecast precision. The distinction matters because precision is what tells you whether the system is improving — and accuracy alone can hide a forecast that's accidentally right.

The definitions you actually need

Borrowing from the statistical sense (lightly):

  • Accuracy = how close, on average, your forecast was to actual. "We forecasted $1M, we closed $950K, 95% accurate."
  • Precision = how tight your forecast confidence interval is. "We forecasted $1M ± $50K, we closed $950K — inside our band." vs. "We forecasted $1M ± $400K, we closed $950K — inside the band, but the band was useless."

Most CRMs (HubSpot, Pipedrive, Salesforce default) emit a single point-estimate number for the forecast. Reps either commit a deal or don't. The committed pipeline value rolls up. The variance is implicit — the manager's gut handles it.

Why this matters: the accidentally-correct forecast

Take two sales teams at the end of Q1:

  • Team A: Forecasted $2.0M. Closed $1.95M. Accuracy: 97.5%.
  • Team B: Forecasted $2.0M. Closed $1.95M. Accuracy: 97.5%.

Same accuracy. But underneath:

  • Team A's forecast had a confidence interval of $1.85M-$2.15M (10% spread). They were precise AND accurate. Their system works.
  • Team B's forecast had a confidence interval of $1.2M-$2.8M (40% spread). They were accidentally accurate. The single-point number hit close by chance; the underlying confidence was terrible.

If you only measure accuracy, you don't know which team you have. And next quarter, Team B might forecast $2.0M and close $1.4M. Same system, different luck. The accuracy number gives you no warning.

Where the imprecision comes from

Three places, in order of weight:

1. Reps committing on instinct

A rep marks a deal "commit" because they have a good feeling about the buyer. Their feeling is roughly 60-70% accurate over 6 months — better than chance, worse than data. The variance in their instinct is what propagates into the team forecast.

2. Stage probability defaults

Most CRMs default to "Negotiation = 75% probability." That number was set by the CRM admin two years ago based on intuition. Some deals in Negotiation are 90%; some are 30%. The fixed-probability default smooths over real variance and produces a forecast that looks tight on paper.

3. Activity recency that nobody factors in

A deal in Proposal stage with the last activity 14 days ago is not the same as a deal in Proposal stage with the last activity 2 days ago. Most CRM forecasts treat them identically. The variance from this single oversight can be 20-30% of the forecast band.

What an evidence-backed deal-health score actually does

Rather than asking the rep "commit or not," an evidence-backed deal-health system reads real signal — engagement recency, multi-thread coverage, time-in-stage, surfaced objections, meeting sentiment — and emits a probability + confidence interval per deal.

When you roll these up, the forecast comes with:

  • A point estimate ($1.95M expected close)
  • A confidence band ($1.85M low, $2.05M high)
  • Per-deal contributions to variance ("the wide band is mostly because deals X, Y, Z each have ±$100K of uncertainty due to single-threading")

The actionable piece is the third one. Now your manager knows exactly which deals to coach to tighten the band. "Multi-thread X, Y, and Z this week and we go from ±$200K to ±$80K" is a coachable instruction.

How to start measuring precision today

Even if your CRM doesn't emit confidence intervals, you can build a manual precision metric:

  1. Snapshot the forecast point estimate at the start of each quarter.
  2. Snapshot the actual close at the end of each quarter.
  3. Compute the absolute % error per quarter.
  4. Track the variance of that % error across 4+ quarters.

If your % error is 5% one quarter, 25% the next, 8% the next, 18% the next — your AVERAGE accuracy is fine. Your PRECISION is terrible. You're going to over- or under-shoot quarterly targets unpredictably.

Once you have the precision number, the next question is: what's driving the variance? That's where the deal-health math comes in.

The bottom line

If your sales forecast accuracy is 90% but it's been 60-95% over the last four quarters, you don't have a forecast. You have a noisy guess. The fix isn't more training or more reps — it's a system that emits precision, not just accuracy. That system has to read real signal from deal activity, not rep self-reporting. And it has to be transparent enough that your reps trust the math.

Try the idea

See it run on a real workspace.

The free plan never expires — 2 seats, unlimited deals, the full AI update modal. 60-second setup, no credit card.