Most forecast meetings skip all three and argue about individual deals instead. That is why the number moves every week and nobody trusts it by the end of the quarter.
In short
- Default stage probabilities are placeholders. HubSpot's default pipeline ships at 20/40/60/80/90% and multiplies deal amounts by those values to produce weighted amounts (HubSpot).
- Measure conversion by cohort, following deals that entered a stage through to their outcome.
- Gate the pipeline before counting: stale deals, past close dates and placeholder amounts are not pipeline.
- Run three forecasts — stage conversion, run rate, and rep commit — and treat disagreement as a diagnostic.
- Coverage is a sanity check, not a target. The right multiple is the inverse of your real close rate.
- Score the forecast every period for error and bias, or it never improves.
What a forecast is for
A forecast is a decision input: hire or wait, spend or hold, promise the board or manage expectations. So the quality bar is not "was it exactly right" but "was it right enough, early enough, to change a decision correctly."
Microsoft's own guidance frames forecasts the same way: they work best when "reviewed regularly and used as a planning tool, not just a reporting artifact," with committed revenue, best-case opportunities and the gap to target compared so teams can act earlier in the cycle (Microsoft Learn).
That is the job. Everything below is how to make the number worth acting on.
Method 1: Stage conversion, measured properly
The standard weighted pipeline formula is simple: for each stage, multiply open value by the probability of closing from that stage, then add closed-won so far.
The trap is the probability. Both major platforms compute weighted amounts straight from a stage percentage — HubSpot multiplies amount by deal probability, set per stage in the pipeline (HubSpot) — and almost nobody changes the defaults.
Measure your own, by cohort:
- Pick a window that is old enough for deals to have finished, for example deals that entered each stage between 6 and 12 months ago.
- For each stage, count how many of those deals were eventually won, regardless of how long it took.
- Divide. That is your stage-to-close rate.
Do not divide this quarter's wins by this quarter's open pipeline. That mixes cohorts and will understate every rate.
A worked example
Everything below is hypothetical, invented to show the arithmetic.
Quarterly target: $600,000. Already closed: $120,000. Remaining target: $480,000.
Open pipeline:
| Stage | Open value | Cohort close rate | Expected |
|---|---|---|---|
| Discovery | $900,000 | 12% | $108,000 |
| Proposal | $600,000 | 30% | $180,000 |
| Negotiation | $300,000 | 55% | $165,000 |
| Open total | $1,800,000 | $453,000 | |
| Closed won to date | $120,000 | 100% | $120,000 |
| Forecast | $573,000 |
Against a $600,000 target, that is a $27,000 gap — close enough to chase, far enough to act on now.
Now run the same pipeline through typical out-of-the-box probabilities (40%, 60%, 80%):
$900,000 × 0.40 = $360,000 $600,000 × 0.60 = $360,000 $300,000 × 0.80 = $240,000 Plus $120,000 closed = $1,080,000
Nearly double, and comfortably above target. Nobody escalates anything, and the quarter ends 45% short. Same deals, same CRM, different assumption.
The hygiene gate
Before a deal enters the forecast at all, it passes five checks. This is MitHub's hygiene gate, and it runs as a saved view anyone can open.
| Check | Fails when | Treatment |
|---|---|---|
| Alive | No logged activity in 21 days | Excluded until re-contacted |
| Dated | Close date is in the past or missing | Excluded; owner must fix |
| Sized | Amount missing, or a round placeholder never updated | Excluded |
| Staged | In the stage longer than 2× the median for that stage | Demoted one stage |
| Owned | No owner, or owner inactive | Reassigned before counting |
Applying the gate to the example: suppose 30% of Discovery value — $270,000 — has had no activity in three weeks. Discovery drops to $630,000, and $630,000 × 0.12 = $75,600.
Revised forecast: $75,600 + $180,000 + $165,000 + $120,000 = $540,600.
The gate cost $32,400 of forecast on paper and cost nothing in reality, because those deals were never going to close. What it bought is a number the team can act on in week 4 instead of discovering in week 12.
Method 2 and 3: run rate and commit
A single method is a guess with arithmetic. MitHub runs three and reads the spread.
- Stage conversion (above): $540,600 after the gate.
- Run rate: average of the last four comparable periods. Say those were $470,000, $545,000, $560,000 and $600,000 → ($470,000 + $545,000 + $560,000 + $600,000) ÷ 4 = $543,750.
- Commit: what reps and managers commit to, using forecast categories. Dynamics uses pipeline (low confidence), best case (medium) and committed (high), rolled up through the hierarchy with room for manual adjustments when something isn't captured in the system yet (Microsoft Learn). HubSpot's forecast tool uses a similar set — not forecasted, pipeline, best case, commit, closed won (HubSpot).
In the example, stage conversion ($540,600) and run rate ($543,750) agree within 0.6%. Two independent methods landing in the same place is the strongest evidence a forecast can offer.
Reading the disagreement
| Pattern | Likely cause | What to do |
|---|---|---|
| Commit ≫ stage conversion | Happy ears, or deals staged too far forward | Re-test stage entry criteria with evidence |
| Commit ≪ stage conversion | Sandbagging, or rates measured on a better era | Check whether the market or the offer changed |
| Stage conversion ≫ run rate | Pipeline inflated with stale deals | Tighten the hygiene gate |
| Stage conversion ≪ run rate | Pipeline genuinely thin, or new pipeline not yet entered | Look upstream at lead flow |
This is the operate loop applied to forecasting: observe, hypothesize, adjust — every period, in public.
Coverage, properly understood
Coverage = open pipeline ÷ remaining target. In the example: $1,800,000 ÷ $480,000 = 3.75×.
Is that good? Only one thing decides: the inverse of your real blended close rate. If the whole pipeline converts at 25%, you need 4× to hit the number, and 3.75× is slightly short. If it converts at 40%, you need 2.5× and you are fine.
Copying someone else's "3× rule" imports their close rate along with it. Compute yours and state it next to the ratio.
Score the forecast, every period
A forecast that is never scored cannot improve. Track two numbers.
| Period | Forecast | Actual | Error | Absolute % error |
|---|---|---|---|---|
| Q1 | $500,000 | $470,000 | +$30,000 | 6.4% |
| Q2 | $520,000 | $545,000 | −$25,000 | 4.6% |
| Q3 | $610,000 | $560,000 | +$50,000 | 8.9% |
| Q4 | $580,000 | $600,000 | −$20,000 | 3.3% |
Mean absolute percentage error: (6.4 + 4.6 + 8.9 + 3.3) ÷ 4 = 5.8%. Total signed error: +$30,000 − $25,000 + $50,000 − $20,000 = +$35,000, an average bias of +$8,750 per quarter.
Precision and bias are different diseases. A 15% error with no bias means noisy deals. A 5% error that is always positive means the team is systematically optimistic — which is the easier problem, because you can correct a known bias while you fix its cause.
What breaks forecasts
- Stage inflation. Deals pushed forward to look busy. Cured by evidence-based stage entry criteria in your CRM architecture.
- Close-date roulette. Dates dragged forward each month. Track "number of times the close date moved" as a field and watch it predict losses.
- Mixed revenue types. New business, renewals and expansions convert differently. Forecast them separately.
- Small numbers. With twelve deals a quarter, percentages are noise. Forecast in deal counts and average values instead, and say the range out loud.
- Seasonality. A four-quarter average hides a business where Q4 is double Q1. Compare against the same period last year as well.
- One person's spreadsheet. If the forecast lives outside the system and dies when that person is on holiday, it is not a system.
The weekly forecast ritual
Thirty minutes, same agenda, no deal-by-deal theatre:
- Publish the three numbers (stage conversion, run rate, commit) and the gap to target.
- Review the hygiene gate exclusions — what fell out and why.
- Review only the deals that changed stage or date this week.
- Name the actions that close the gap, with owners.
- Log this week's forecast so it can be scored later.
Step 5 is the one everybody skips, and it is the only one that makes next quarter's forecast better than this one.
A forecast is downstream of everything else in the revenue system: the scores that prioritize, the routing that ensures leads are worked, the architecture that records what happened. If the forecast is unreliable, the fault is usually upstream. The Faculty of Revenue Reverse Engineering starts at the other end of that chain — with the money that already arrived — precisely because it is the only number that never needs a model.