The situation
Two problems ran in parallel. Acquisition spend was allocated across mass media, direct outreach, and online channels using attribution that finance did not entirely believe, which meant every budget conversation reopened the same argument. Separately, cross-sell of credit products into the existing deposit base relied on broadly targeted offers, generating multiple touches per customer and a response rate that made the economics marginal.
The approach
Cross-sell and next-best-action. Rather than modeling who would respond, the work modeled who would respond because of the offer. That distinction matters commercially: propensity models concentrate spend on customers who were going to convert regardless, while uplift models find the population the intervention actually moves. Product recommendation was sequenced so that customers received fewer, better-matched touches.
Marketing mix and channel allocation. Channel spend was modeled to separate genuine contribution from correlation, which produced a defensible basis for reallocation. The exercise also revealed that informed cost per acquisition ran 20% to 40% below the inferred figure the business had been working from, which changed several decisions on its own.
Campaign measurement. A/B testing discipline, message optimization, and product basket analysis contributed a further 3% to 5% incremental customer growth.
The outcome
Cost per acquired customer fell by more than 20%. Acquired card customers generated roughly $2,000 in additional annual spend each. Return on advertising spend improved by around 7%, with the larger benefit arriving through a measurement basis the CFO would accept.
What made it work
Agreeing the measurement approach before producing any numbers. Attribution changes move budgets between people, and a reallocation that arrives as a surprise gets litigated rather than implemented.