Recruitment funnels and sports matches share a shape. Both are sequences of stages, and both get summarised by one headline rate that compresses everything. A tennis dataset shows when that compression is harmless and when it destroys the finding.
The corpus holds 116,382 completed best-of-three singles matches from January 2023 to September 2026. For every match it records whether the player who lost the opening set recovered to win. The headline rate is 16.52%.
The cut that tells you almost nothing
Break those matches down by competition tier and six rates appear.
The men's top tour returns 18.21% across 10,945 matches. The men's Challenger tier gives 17.32% over 30,406 matches, and the women's Challenger tier 17.25% over 4,731. The women's top tour returns 17.04% over 13,543 matches. The women's ITF circuit gives 15.97% over 29,331 and the men's 15.18% over 27,415.
Three percentage points separate top from bottom. The confidence intervals run from 0.42 to 1.08 points, so some of those differences are real. None of them is large enough to change a decision.
Break the same matches down by court surface and the result is flatter still. Grass 17.2%, clay 16.79%, hard 16.47%, across 113,288 matches. Under a percentage point of spread.
The cut that carries everything
Now break them down by how the opening set actually finished.
Lost 7-6, the recovery rate is 21.92% across 14,265 matches. Lost 6-4, 19.41% over 25,546. Lost 7-5, 18.89%. Lost 6-3, 17.39%. Lost 6-2, 13.3%. Lost 6-1, 11.06%. Lost 6-0, 7.18% across 5,833 matches.
That is a three-fold range, and the rates fall in order as the margin widens, with no exception across seven categories.
What this says about funnel reporting
The same dataset, cut three ways, produces one split worth acting on and two that are noise. The difference is not the sample size. The surface cut has 113,288 matches behind it and still finds nothing.
The difference is whether the dimension touches the mechanism. Margin of defeat measures how far apart the two sides were. Surface and tier are context. Only the first one is causally close to the outcome.
Applied to a hiring funnel, the implication is uncomfortable. A conversion rate split by source, by region or by month will often be flat, and flatness gets read as stability. It usually means the dimension is not where the variation lives. The dimension that matters is frequently something closer to the decision itself, and often harder to record.
Three questions worth asking of any breakdown
Does it order itself? Categories with a natural order should produce rates in that order. Seven scorelines here do. A breakdown that jumps around has a sample problem or a definition problem.
Are the counts published next to the rates? A rate of 17.25% on 4,731 cases and 17.32% on 30,406 cases deserve different confidence. Only the counts show that.
Is there a cut that found nothing? A method that reports differences everywhere is suspect. This one reports a flat surface result and says so, which makes the scoreline result easier to believe.
All of the above comes from the Live Tennis API, a JSON service for developers. It covers live scores, fixtures, players and match events, with point-by-point history behind them. The figures are published at blog.livetennisapi.com/studies.json under CC BY 4.0, free to reuse with attribution.
Retirements and walkovers are excluded, and the counting is done from the point-by-point tape instead of estimated.
