Topic 9 of 20 · Part III : Personalisation and Responsible AI

Personalisation

Good personalisation does not change objective facts. It changes prioritisation, explanation and presentation.

The control pattern

Data asset

User-context data

Knobs

Relevance, Recency, Consent, Retention, Risk profile and horizon, Portfolio exposure

Outcome

More useful responses for the individual user

Measurement

See how to measure this below

What it is

Personalisation changes which facts and insights are emphasised for a particular user. The data asset is user-context data; the knobs determine which context is relevant, how recent it must be, how long it is retained, and whether the user consented to its use.

Good personalisation does not change objective facts. It changes prioritisation, explanation and presentation. The same earnings report may matter differently to a long-term shareholder, a short-term options trader, or someone considering a first position.

Store the minimum context needed, apply expiration rules, and keep user-provided facts separate from model inferences.

The knobs in detail

Each row is one adjustable property of the data asset, and what moving it tends to do.

KnobWhat you adjustLikely effect
RelevanceWhich context applies to this questionPrevents irrelevant tailoring
RecencyHow current the context must beStops stale profiles driving answers
ConsentWhat the user agreed toDetermines what may be used at all
RetentionHow long context is keptLimits accumulation of sensitive detail
Risk profile and horizonStated tolerance and time frameChanges emphasis, not facts
Portfolio exposureHoldings and concentrationSurfaces correlation risk across positions

Applied: Stocks Assistant

For Stocks Assistant, useful context includes holdings, position sizes, cost basis when supplied, investment horizon, risk tolerance, income versus growth objective, sector concentration and notification preferences. A portfolio owner with heavy semiconductor exposure may need a correlation and concentration warning even if each holding looks attractive in isolation. A dividend-focused user may care more about cash-flow coverage and capital allocation than near-term price momentum. The assistant should show which profile settings affected the result and let the user change or disable them.

How to measure it

Evidence that the knob produced the intended behaviour, rather than shifting the problem elsewhere.

  • Usefulness and consistency of tailored results
  • Override rate : how often users correct the tailoring
  • Whether users understand why a result was personalised
  • Share of features operating on minimised context

Common mistakes

Silently inferring sensitive financial circumstances, or converting weak behavioural signals into permanent traits.

Implying that suitability has been established when important financial information is absent.