AI Assistant (Leat MCP)
July 20
→
A/B testing
Test two versions before committing to either. Compare a reward value, a message, or a mechanic design on a random split of the same audience, and read the result on revenue and margin. Because significance is calculated as the test runs, you find out when the result is real, not just when someone gets impatient.
Reward value: €5 against €10. Incentive type: a discount against a free item against bonus points. The message and its timing. The design of a mechanic: eight stamps against ten, a tier threshold at one level or another. A policy: thirty-day expiry against ninety. Anything the program can vary can be a variant, and the variants are the questions a program owner actually argues about.
The audience is split at random, so the groups match on everything except the variant. A holdout can run as a third arm, so each version is measured against nothing as well as against each other. Leat shows the sample size needed before the test starts and reports significance as it runs, so a result is called when it's real and not when someone's impatient. The variant a customer receives is consistent across every channel, so the test isn't muddied by a wallet notification that didn't know which arm they were in.
Variants are compared on incremental visits, orders, revenue, and margin after the incentive's cost is subtracted, and by segment where the effect differs. Response rate is reported, but it doesn't decide. A €10 voucher that's redeemed more often than a €5 one and earns less after cost loses, and the test says so. The cheaper version wins more often than anyone expects, which is the most common finding and the most valuable.
When a test concludes, apply the winner to the whole audience with one action. Or, where the result differs by segment, let Decisioning use it: uplift per variant per segment becomes what the engine draws on to pick the right version for each customer, so a test that found €5 works for regulars and €10 for lapsed members ends with both being used, correctly. Every test, its variants, and its result are recorded, so the organization stops re-running the same arguments.
Promotion governance
Hold eligibility, spend, and frequency limits in one place and apply them before any incentive is issued
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
Seasonal and limited-time campaigns
Launch and retire rewards on schedule without manual upkeep
Lapsed customer recovery
Spot fading visit patterns across locations before a customer quietly stops coming back
