Personalisation
Relevance, not a first name in a subject line.
We personalise the things that change a decision: which products are shown, which message leads, and when it arrives. A name in the greeting changes nothing and everybody can tell.
Done badly it is worse than nothing. Recommending the product someone bought last week, or a shade they returned, tells a customer precisely how little is being paid attention.
Rules before models
Most of the available gain comes from a handful of clear rules: last category browsed, replenishment interval, size and shade already owned, market and season. These are explainable and easy to correct.
Algorithmic recommendation is worth adding once there is enough behaviour to learn from, and it is worth resisting before that, because a model trained on thin data is confidently wrong.
Held to a standard, and to consent
Every personalised surface has a fallback for the customer nothing is known about, because the first-time visitor is the one most often shown an empty or absurd module.
Consent and data minimisation are designed in, not bolted on. What is used is what the customer agreed to, and the programme should be comfortable explaining any of it.
Want personalisation a customer would actually notice?
Let's talk.
