Designing a next-best-use-case programme for Amazon Pay cross-sell
How a cross-sell problem became a customer-decision problem, leading to a next-best-use-case approach built with the ML team.
Identified cross-selling issues and created the next-best-use-case programme in push notifications, working with the ML team to make cross-sell recommendations more relevant to customer behaviour.
Cross-sell works when the recommendation follows the customer journey rather than the company product catalogue.
The exact incremental uplift is not published here. The case focuses on the strategy, operating model and measurement logic behind the programme rather than reconstructing a result that should come from the underlying experiment or reporting.
What was actually stuck?
Customers often adopted one Amazon Pay use case without naturally discovering adjacent services. The obvious response would have been more cross-sell campaigns. The better question was which next use case was most relevant for each customer and when that recommendation should appear.
What did the evidence suggest?
A product-level view asks, “How do we sell service B to users of service A?” A customer-level view asks, “Given what this customer has already done, what is the next useful job Amazon Pay can help them complete?” That shift changes targeting, creative, timing and measurement.
From diagnosis to intervention.
Defined behavioural signals that could indicate readiness for an adjacent use case.
Partnered with ML stakeholders to translate those signals into next-best-use-case recommendations.
Designed push communication around relevance, not catalogue breadth.
Created eligibility and suppression rules so recommendations did not compete with higher-priority customer journeys.
Structured measurement around adoption and incremental cross-sell rather than message engagement alone.
What should move if the strategy is working?
What changed?
Created the next-best-use-case programme in push notifications with the ML team, connecting customer behaviour, recommendation logic and lifecycle communication.
The programme shifted the cross-sell question from 'which product do we want to promote?' to 'which use case is most useful for this customer next?'
Principles, not playbooks.
Next-best-action is a marketing problem before it is a machine-learning problem.
The model is only valuable if marketing can turn its output into a useful customer decision.