If You Only Contact The Top 10%, How Much Do You Capture?

How it works

Rows are sorted by the score from best to worst and responders are counted as they accumulate. Cumulative gain at depth d is the share of all responders found in the best-scoring d of the population; cumulative lift is that gain divided by d, so random targeting sits at 1.00. The ranked list is cut into equal buckets — twenty at 2,000 rows or more, ten from 100, five below that — and each bucket reports its responders, response rate, lift and running totals. Where a bucket boundary falls inside a run of tied scores the responders in that run are shared out in proportion to how much of it is contacted, which is the expected result of breaking those ties at random and the only order-independent answer. The break-even value-to-cost ratio at each depth is one divided by the response rate achieved to it; when a contact cost and a value per response are supplied, net profit is computed at every depth on a fine grid and the maximizing depth reported.

Use it when you have a scored list and a real outcome and the decision in front of you is how many of them to contact — a mailing depth, a call-list cut-off, a retention-offer budget.

Not for judging a classifier's ranking quality in the abstract, where the ROC analysis and its AUC are the better tool; not for multi-class outcomes; and not for establishing that contacting anyone caused a response, which needs an experiment rather than a ranked list.

What data do you need?

Any scored list with the outcome attached, one row per contactable unit. For example, a mailing file scored by a propensity model:

Minimum 50 rows · Best with 1,000-200,000 rows with at least a few hundred responders

What's in the report?

Standard-library analysis: cumulative gains and lift for campaign targeting decisions. Map the actual yes/no outcome and any ranking score — a model's propensity, a RFM rank, a hand-built priority number — and get the answer to "if I only contact the top N%, how much of the total value do I capture?". Delivers the cumulative gains curve against the random-targeting diagonal, the cumulative lift curve, a decile or ventile table with response rate and lift in each bucket, the top-decile lift as the headline number, the break-even value-to-cost ratio at every depth, and — when a contact cost and a value per response are supplied — the profit-maximizing contact depth.