Sprint estimation goes wrong in the same direction almost every time: work takes longer than planned. The usual response is to ask people to estimate better, which does not work, because the estimate is not the problem. The problem is that a single number hides its own uncertainty. Planning with p50 and p90 makes that uncertainty explicit. The p50 is the median outcome, the one you hit half the time. The p90 is the one you hit nine times in ten. A team that commits at p50 is flipping a coin every sprint.
1. Where the Two Numbers Come From
Kavanah takes each person's completed tasks from the last 90 days that have both an estimate and logged hours, and computes the ratio of actual to estimate for each. The median of those ratios is that person's p50 multiplier; the 90th percentile is their p90 multiplier. A developer who usually takes 1.2 times their estimate, and sometimes 1.8 times, gets those two numbers. Someone with fewer than five completed samples inherits the team's distribution until they have enough history of their own.
2. Why the Stored Estimate Is Never Changed
It is tempting to write the uplifted number back onto the task. Do not. Calibration divides actual hours by the stored estimate, so if you store the uplifted value, the next ratio measures an already-corrected number, drifts toward 1.0, and the correction quietly unwinds itself. Kavanah applies the uplift only in the planner's forecast and leaves the estimate as the person wrote it.
3. Floors That Keep the Forecast Safe
A team that consistently over-estimates would produce a multiplier below one, which would shrink commitments below the estimates themselves. That is the opposite of what the mechanism exists for, so the p50 multiplier is floored at 1 and the p90 is never allowed below the p50. The forecast can only add caution, never remove it.
4. How to Use the Two Lists
The planner packs the unstarted backlog into the team's capacity twice and reports what fits at each level, per person, with the deferred work listed. Commit the p90 list. Put the difference between the two lists on the board as stretch. When the sprint goes well, the stretch work gets done and nobody is surprised; when it goes badly, the commitment still holds.
5. What This Fixes Downstream
DORA's 2024 research linked unstable priorities to lower productivity and higher burnout. Sprints that spill over are a quiet source of instability: every miss re-cuts the next sprint and moves the roadmap. Committing at p90 makes the sprint a promise the team keeps, which makes the quarter predictable, which is what everyone above the team actually wanted from the estimates in the first place.
Ask the agent for a sprint recommendation and Kavanah returns both lists; committing one is a person's decision in the Sprint Planner. For the rest of the planning, code and AI requirements a tech company should check, see our project management software for tech companies page.
