Loyalty glossary · 7. Data and analytics (31)
Next Best Action
Next Best Action is the use of predictive models to select the most relevant offer, message, or intervention for a member at a given touchpoint, based on their transaction history, engagement signals, and projected value to the programme.
Next Best Action is sold as personalisation, but it is really response optimisation. The engine scores every candidate offer against a member's predicted reaction, then serves the one with the highest score. That is a targeting tool, not a relationship tool. The difference matters because optimising for immediate response is not the same as optimising for lifetime value.
The data problem is that these models learn from past redemption. A member who has never redeemed produces no outcome for the algorithm to fit, so the algorithm treats them as noise. The result is a loyalty scheme that speaks only to its most active members and quietly ignores everyone else. That is not a bug, it is the logic of the training set.
The recommendation logic moves the levers of accrual and activity-based qualification, but it does so incidentally. Serving bonus point offers on stays repeatedly accelerates some members up the tier ladder while leaving others behind. The logic does not know the qualification rules, it only knows which offer gets the click. Tier progression becomes a side effect of targeting, not a designed outcome.
Run the numbers. A programme issues 500 million points in a year. A recommendation system that lifts redemption by 8 percent moves an extra 40 million points out the door. At a cost of 0.9 cents per point, that is 360,000 dollars in added redemptions. Raise the lift to 12 percent and the outflow becomes 60 million points, adding another 180,000 dollars in cost. The finance team rarely sees that number before the system goes live.
Actuarial models in loyalty forecasting assume a stable pool of members with predictable aggregate behaviour. The targeting approach breaks that assumption by systematically changing individual choices. The finance team books an actuarial liability for future redemption, while the marketing team runs a system that quietly makes that liability wrong. The two functions rarely sit in the same meeting.
The honest version of Next Best Action would optimise for member lifetime value, not for the next click. That requires a model that can value a member who does nothing today, which almost no loyalty programme has built. Until then, the technology remains a tool for harvesting existing demand, not creating it.