When would you stop sizing items and forecast a delivery date from counted history?

## The question underneath the practice Nobody wants story points. Stakeholders want an answer to 'when will this be ready, and how confident are you'. Sizing is one route to that answer: size the outstanding items, divide by a velocity range, quote a span of Sprints. Counting is another route: observe how many items the team finished per Sprint over its recent history, divide the number of remaining items by the low and high of that observed rate, quote a span of Sprints. The second route skips the sizing entirely. That is attractive — sizing a large backlog is expensive and the numbers it produces are used exactly once — but it is only honest when several preconditions hold. ## Preconditions, all of which must hold 1. **Items are broadly similar in size.** Counting works because each item is roughly interchangeable. One item worth ten of its neighbours breaks the arithmetic silently, because the count still says one. 2. **The history is long enough.** Eight to twelve Sprints, not three. A short history produces a range so wide it forecasts nothing, or so narrow it lies. 3. **The team and the Definition of Done are stable** across that history, for the same reason a velocity is invalidated by either changing. 4. **The coming work resembles the history.** A rate measured on one kind of work does not transfer to a body of work unlike it. 5. **The organisation can hear a range.** If every forecast is compressed into a single date on arrival, the method's central honesty is discarded before it reaches anyone. ## A worked forecast A team building a translation-agency portal, releasing on a six-week cadence, finished 9, 7, 11, 6, 8, 9, 5 and 10 items across its last eight Sprints. That is 65 items, a mean of 8.1 per Sprint, and a worst observed Sprint of 5. Thirty-eight items remain in the release scope. - At the mean rate: 38 / 8.1 is about 4.7 Sprints. - At the worst observed rate: 38 / 5 is 7.6 Sprints. The honest statement is 'five to eight more Sprints, and we have never gone slower than the eight-Sprint bound' — which, on a six-week release cadence, is roughly two to three more releases. Notice what the statement does not do: it does not average the two into 6.2 and call that a date, and it does not present the optimistic end as the plan. The seasonal traffic peak sitting inside that window is a reason to lean toward the pessimistic end, and to say so explicitly rather than to fold it into the arithmetic. ## Comparing the two instruments | | Summed story points | Counted history | |---|---|---| | Input required | Every remaining item sized | A count of remaining items | | Cost to produce | Hours of the whole team | Minutes, from existing records | | Handles uneven item sizes | Yes, that is what sizes are for | No; degrades silently | | Handles genuinely novel work | Poorly, but it notices | Not at all | | Surfaces hidden assumptions | Yes, in the conversation | No | | Natural output | A span of Sprints | A span of Sprints with a confidence claim | The row that decides most real cases is 'surfaces hidden assumptions'. The forecast is the visible product of sizing, but the shared understanding is the valuable one, and it is the thing a counting method silently removes. ## What you lose that is not the number - **The moment two people discover they imagined different work.** No count produces this. - **The early warning that an item is too big.** A size that will not fit the ladder is a signal; a count treats that item as one item like any other. - **A common language with the Product Owner** about relative cost, which is what makes trade-off conversations possible before the work starts. - **A forecast for anything new.** History cannot forecast work that has no analogue in it, and the honest answer there remains a time-boxed investigation. ## The conversation with leadership Expect 'we no longer estimate' to be heard as 'we no longer commit to anything'. Frame the change as a change of instrument rather than a withdrawal: the team is answering the same question with a range drawn from what it actually delivered, instead of with a summed guess presented as a date. The strongest move is to run one past release both ways and show which forecast the outcome fell inside. The position worth defending in an interview is rarely the pure version. Keep the sizing conversation, because that is where the value is; drop or coarsen the number, because that is where the cost is; forecast from counted history where the preconditions hold; and keep sizing for novel work, where nothing else can help.