AI Assistant (Leat MCP)
July 20
→
Adaptive decisioning
Let every outcome make the next decision a little better. Feed redemptions, visits, and results back into the models that chose them, inside the boundaries you've already set. Because adaptation never changes what the engine is allowed to do, the program gets smarter without ever getting less predictable.
Every action has a result, and the result is recorded against the decision that produced it: the offer redeemed or ignored, the visit that followed or didn't, the spend that changed. Where a holdout was running, the result is causal rather than coincidental. That record is what the models train on, so they learn from your customers and your mechanics, in your market, and from nothing else.
Uplift for a €5 voucher in a segment where it stopped working drifts down. Propensity to return for customers whose pattern has shifted updates. Affinity for a category that's newly in season rises. The estimates the engine ranks options on are recalibrated on a continuous cadence, so a decision made this week reflects behavior from last week, and the summer program isn't running on what the winter taught it.
Adaptation changes what the engine estimates. It never changes what it's permitted to do. Governance rules, caps, budgets, margin floors, and consent are fixed until a person changes them, and the signals a model may learn from are set under Responsible AI controls. Where the engine spends a small share of decisions to learn about a new mechanic or segment, that share is bounded and its cost visible. The program gets smarter within the lines; it doesn't redraw them.
What the models learned, when, and which decisions it changed is on the record. A shift in estimated uplift for a segment shows as a dated entry with its cause. Model performance is tracked against a fixed baseline under model monitoring, so learning is proven to be improvement and not just change. And a model version can be held or rolled back if a change isn't wanted, because a system that learns should also be one you can stop.
Responsible AI governance
Set the boundaries the system operates within and prove it stayed inside them
Program integrity
Keep stamps, points, and rewards worth what they're meant to be worth
Lapsed customer recovery
Spot fading visit patterns across locations before a customer quietly stops coming back
Margin-safe promotions
Run offers that can't combine into a loss
Program change management
Change a rule once and have every campaign, channel, and decision honor it
