Strategy & governance · 9 min read
AI personalisation without crossing the line
AI makes it easier to choose content, offers, timing, and journeys for smaller groups or individuals. That capability does not answer whether a personalisation is welcome, lawful, accurate, or commercially worthwhile. Good programmes define boundaries before chasing maximum variation.
Personalise the task before the person
Start with declared needs, current journey stage, product context, location where relevant, and recent interactions the customer reasonably expects the service to remember. These signals can improve usefulness without building an intimate profile.
Sensitive traits, inferred vulnerability, and surprising cross-context connections carry much greater risk. Excluding them should be a product decision, not merely a model instruction.
Give the customer a coherent explanation
Teams should be able to explain why a recommendation or message appeared in language a customer can understand. If the explanation sounds uncomfortable, the personalisation probably needs redesign.
Offer meaningful controls and make preference changes propagate through connected systems. A preference centre that does not affect activation is interface theatre.
Test value against a simple alternative
Compare AI-driven personalisation with a strong non-personalised or rules-based experience. Measure incremental conversion, customer value, complaints, unsubscribes, and operational cost.
More variants can make learning slower when each receives too little traffic. Use hierarchy and shared learning rather than treating every micro-segment as an independent campaign.
Govern the full decision chain
Document permitted data, feature logic, model or vendor, content sources, exclusions, review owner, and fallback experience. Monitor uneven outcomes and stale assumptions after launch.
Privacy, legal, brand, data, and commercial teams need one operating process. A late approval meeting cannot repair a personalisation system whose data flows and incentives were never designed responsibly.