Personalization

1. What is personalised

   Which parts of the experience change by person, and what the customer gains from each. The inventory that makes the maintenance cost visible.

   Learn

2. The signals

   What the personalisation is based on — stated preferences, observed behaviour, inferred attributes — and how reliable each is.

   Learn

3. The limits

   Where the organisation chooses not to personalise, and why. The boundary that keeps relevance from becoming intrusion.

   Learn

4. Whether it works

   Evidence that the personalisation improves something for the customer as well as for the organisation, tested rather than assumed.

   Learn

1. RFM analysis

   Direct marketing practice, formalised in the 1990s · 1990s

   Sort customers by how recently they bought, how often, and for how much — three fields that predict a great deal.

   It long predates machine learning and remains an unreasonably strong baseline. Recency does most of the work: how long ago someone last bought predicts whether they will buy again better than almost anything else on file, and all three fields are already in every order table.

   Reach for it when

   As the first personalisation anyone builds, and as the baseline any more sophisticated model must beat.

   Where it stops

   It describes what someone did, not what they want. It cannot help with a new customer, and it will keep recommending more of what they already bought.

   Long-standing direct marketing practice; formalised in Jan Roelf Bult & Tom Wansbeek, “Optimal Selection for Direct Mail”, Marketing Science, 1995.

2. Collaborative filtering

   Goldberg, Nichols, Oki & Terry · 1992

   Recommend to someone what people who behaved like them went on to choose, without needing to know anything about the item.

   The insight is that similarity of behaviour carries more information than description of content, and it needs no understanding of what is being recommended at all. Two problems have followed it ever since: it has nothing to say about a new user or a new item, and it narrows what people are shown to a reflection of what they already did.

   Reach for it when

   Where there is enough behavioural data for patterns to be real, and as the standard against which content-based approaches are judged.

   Where it stops

   It amplifies the popular and the already-chosen. Left alone it will recommend a company’s catalogue into a small corner of itself.

   David Goldberg, David Nichols, Brian M. Oki & Douglas Terry, “Using Collaborative Filtering to Weave an Information Tapestry”, Communications of the ACM, 1992.

These are other people’s models, named here so you can go to the source and use them properly. The Omnigoal is not affiliated with their authors and is not endorsed by them; nothing of theirs is reproduced here — no canvas, no diagram, no wording. Each is described in our own words, with the originator credited, because the framework is a place to put thinking, not a replacement for the people who did it. Model names and trademarks belong to their respective owners and are used here only to refer to the work itself.

The line between helpful and unsettling is drawn by the customer, and it is crossed by revealing knowledge they did not know you had.