The program was designed to hold customers who might otherwise leave. In practice it was paying two groups it never intended to: abusers who had worked out the mechanics, and loyal customers who were going to stay regardless.
Both leaks look identical in the aggregate. Redemption is up, cost is up, and retention is stable — so the program appears to be working exactly as designed.
Defined the segmentation, quantified each leak, and specified the redesign with the campaign team.
Behavioural segmentation to separate the two leaks from genuine retention response, then redraw eligibility so the rewards followed the behaviour that actually needed rewarding.
Retention was treated as a constraint, not a target. The question was not how much cost could be removed, but how much could be removed before retention moved — a different and much more answerable question.
Rp XX billion per month in cost reduction, with retention held.