Layer one: segmentation that updates on every event
Static segments are stale the moment they are exported. In ReactorCX segments are evaluated dynamically: a member's segment membership updates as activities arrive, so a lapsed member who transacts is no longer lapsed by the time the next offer is evaluated. AI's role is to propose segments worth having, by correlating transaction patterns against configuration and surfacing the ones that under- or over-perform.
Layer two: preference-based routing
Personalization is not only who gets an offer but what the offer becomes. Preference-based routing lets two members earning from the same promotion receive different outcomes: one takes points, another a voucher, a third a donation, according to stated or inferred preferences. Because routing runs inside the same eligibility framework as the promotion, the outcome is auditable and the liability is attributed correctly. AI's role is to recommend routing rules from observed preferences and to project their cost before publication.
Layer three: personalized gamification
Missions, badges, progression chains and reward catalogs can be member-specific: a challenge set relative to a member's own history, a catalog filtered to their interests. Mechanics evaluate on the same transaction event as earn, burn and tiers, so a challenge completion and its reward are one auditable outcome. AI's role is to draft the mechanic, the thresholds and the cool-off periods, and to simulate participation before a human approves.
Where the governance sits
| Layer | AI proposes | Platform governs |
|---|---|---|
| Segmentation | Segments from observed behaviour and anomalies | Real-time evaluation; row-level permissions on data |
| Routing | Outcome rules from preferences; cost projection | Eligibility framework; lineage on every outcome; approval before publish |
| Gamification | Mechanics, thresholds, cool-offs; participation simulation | Same engine as earn and burn; budget caps; audit trail |
Why this matters for finance
Personalization multiplies the number of distinct outcomes a program produces. Without lineage, that is a liability problem: finance sees a total and cannot explain it. With lineage, every personalized outcome still carries its rule, currency, brand and funding party, so the rollforward ties out no matter how many variants ran. See Financial Integrity.