Who owns the forecast

The BD lead's half is a promise. The commit is what the team says it will deliver, deal by deal. It carries information no model has: the buyer who said yes on the phone yesterday, the procurement freeze the rep heard about at lunch, whether the biggest office really signs before the quarter ends. That knowledge is real, and so is the incentive to shade it. Only the person who runs the team can coach both.

The data team's half is a measurement. The method takes the pipeline as it stood on the freeze date and asks what deals like these have usually done. Its job is not to be right this quarter; it is to be wrong by a known amount, published every quarter, so that everyone can see how far to trust it.

Neither can sign the other's half. An analyst who signs the number is promising delivery on a team they do not run. A BD lead who signs the method is vouching for an error rate they cannot measure. The ownership map that assigns these decisions between data and BD is taught in Marketing Analytics, the ownership map ↗; use it as the rulebook. Tonight is only the forecast row of that map, made concrete.

Splitting ownership is the mechanism. Whether the sales floor and the board trust the result is a leadership problem: how the BD lead talks about the method's number in the Monday meeting, whether a miss is met with curiosity or blame, how AI forecasting tools fit in. That side is covered in AI Sales, "A forecast people actually believe" ↗. Forecasting models themselves, from moving averages to machine learning, live in Timeseries Forecasting ↗. The three methods Kettle prints are walked through, without maths, in Analyst session 4.

The commit landing. Kettle's reps committed 7.86 million yen this quarter; 5.86 million really closed. The commit missed by 2.00 million, 34.1 percent high. Over 20 replays of the quarter (different 300-deal draws from the same real open pool) the rep commit erred by 48.2 percent on average and ran 33.9 percent high. Its range, commit up to commit plus best case, held the actual in only 4 of 20.

How each rep calls deals. The call ratio is a rep's commit divided by what they really closed, over the 20 replays. Ben's is 4.60: he commits four or more times what lands, because he commits from stage 4. Dev's is 0.00: he commits nothing until a contract is signed, so his commit is always zero and always wrong. Aiko sits at 0.98 and Chiara at 1.45. These are behaviours, and behaviour is the BD lead's to coach, in a one-to-one, with the rep present.

The judgement calls. Whether the biggest deal lands this quarter or slips is a call about one buyer, and the person closest to that buyer owns it. The data team's job is to show what happens to every number if the call is wrong.

The method's number and its range. The history-based method said 5.01 million yen, 0.85 million under what closed (14.5 percent low), with an 80 percent band from 2.03 to 8.00 million that covered the actual. Over 20 replays its error was 38.8 percent, the smallest of the three methods, and its range covered the actual in 17 of 20. Note what that does not say: a 38.8 percent typical error is not accurate, it is the least inaccurate method on the bench.

The method's blind spots, in writing. Over 20 replays history-based ran 18.6 percent high, because the May 2018 cohort in the real rows converted at 10.0 percent against 13 to 14 percent for January to April: the method assumes next quarter behaves like the last. It also reads origin and deal age, not stage, so a real late-stage signal is invisible to it. Data owns saying both out loud.

Never editing the commit. When the commit looks high, data does not correct it. Data publishes the method's number beside it, with the range and the track record, and lets the gap start the conversation.

1 · Stage definitions. The buyer-act stages from Business session 1, each with its record. If a stage changes mid-quarter, every method's number changes with it, so the definitions are frozen with the quarter.

2 · What counts as pipeline. Stage 2 and later. Contacted deals are leads, reported separately. Business session 2's coverage sentence goes here.

3 · The freeze date and snapshot. The pipeline as it stood on one named day, saved and never edited. Without it, nobody can later check what the forecast was based on. Analyst session 1 shows how to rebuild a snapshot from the event log when the CRM has already overwritten history.

4 · The methods printed, with their track record. Every method that is shown, with its bias, error and coverage over past quarters, from Analyst session 5. On Kettle's bench: rep commit bias +33.9 percent, error 48.2 percent, covered 4 of 20; stage-weighted +62.7 percent, 68.2 percent, 10 of 20; history-based +18.6 percent, 38.8 percent, 17 of 20. Stage probabilities come from Kettle's own history: on the same deals, the CRM's default probabilities produce 63.64 million yen against the 5.86 that closed.

5 · The range and its stated coverage. A range without a stated coverage is decoration. "80 percent band; held the actual in 17 of 20 replays" is a claim someone can check next quarter.

6 · Who calls the whale. The BD lead, by name, before the quarter starts, with the method's numbers shown both ways.

7 · The gap trigger. When the commit and the method's number differ by more than an agreed share, the BD lead and the data lead meet before the number goes anywhere. Kettle uses 25 percent of the method's number; that threshold is constructed for teaching. This quarter the gap is 7.86 minus 5.01, or 2.85 million yen, which is 56.9 percent of 5.01 (constructed arithmetic on bench numbers), so the meeting happens. Note that a rule of "commit outside the method's range" would not have fired: 7.86 sits just inside the 8.00 ceiling.

Categories are the BD lead's vocabulary. HubSpot's forecast tool, read on its knowledge base in October 2026, lets each deal sit in a forecast category, including Pipeline (low likelihood of closing), Best case (moderate likelihood) and Commit (high likelihood, committed to the forecast), with Closed won alongside. Categories can be mapped from deal stage automatically. The same page is not consistent about the full default list, so teach those three and check your own CRM. Salesforce's standard categories are reported by third-party guides to be similar. Either way, the category is a rep's judgement recorded in a field: it belongs in the BD lead's half of the contract, not the method's.

The accuracy surveys are contested. Forecasting and revenue-intelligence vendors and analyst firms regularly publish figures on how few sales forecasts land close to the actual. Some leaders use them as an outside anchor: proof that a miss is normal and a reason to invest. Others treat them as marketing: the method and sample are usually not published, and the publisher sells the fix. Both readings are held by serious people. The contract sidesteps the argument by putting your own track record in clause 4, measured on your own quarters, which is the only accuracy figure you can check.