Research insights
AI Personalisation: Useful Recommendations and Customer Control
AI personalisation uses a model and available information to vary what a user sees or receives. It may rank products, suggest content or help route a service request. Its value depends on whether that variation helps the customer complete a task, not merely whether it changes the screen.

On this page
A prediction about preference is uncertain. A past click can reflect curiosity, a purchase for someone else or an accidental action. Personalisation should therefore allow correction and avoid treating inferred interests as unquestionable facts about a person.

Define what is being personalised
Specify the element: a recommendation list, message, search order or support response. State the input and intended customer benefit. These uses need different data and evaluation, so a broad promise of an individualised experience is not enough.
For a hypothetical shop, a recommendation might help someone find an accessory compatible with an item they selected. The usefulness depends on verified compatibility and availability, not just similarity to earlier clicks. This is an illustrative task, not a claimed implementation result.
Use examples without assuming the same outcome
Netflix’s explanation of its recommendations system describes helping members find something to watch. It provides a concrete example of recommendation as a discovery task. The provider’s description does not establish a conversion or retention uplift for another business.
Do not turn the names of large platforms and tools into a current buying list or assume every analytics product is itself a personalisation system. Establish requirements before evaluating a particular tool.
Check the information and its purpose
Identify which observations are relevant, how current they are and whether they represent the person or a shared account. Avoid collecting more information simply because it might be technically available. Explain the purpose and review permissions and responsibilities before using customer information.
The 2024 NIST Generative AI Profile discusses privacy and harmful bias among relevant risks for generative systems. Where generated content is used, fluent language still needs appropriate checks. Not every recommender is a generative system, so keep the technology distinction clear.
Preserve choice and a workable fallback
Users should be able to search, browse or correct an irrelevant suggestion rather than be trapped by an inferred preference. A new user or a person with little history still needs a useful experience. Consider what the service shows when data are missing.
Do not infer sensitive needs or suitability from incidental behaviour. Recommendations in consequential areas require appropriate specialist governance. A general marketing use case cannot justify applying the same approach to every sector.
Evaluate usefulness beyond the click
Define success before the trial: finding a suitable item, understanding an option or completing a relevant task. Track rejection, returns, complaints and unwanted repetition as well as engagement. More clicks can reflect confusion rather than better discovery.
Use a suitable comparison and document differences in availability, pricing and promotion. A before-and-after increase does not establish that the model caused it. Inspect important audience groups where appropriate, rather than rely only on an overall average.
Improve the experience deliberately
Review wrong or unhelpful suggestions and decide whether the issue is data, product information or the objective being optimised. Assign responsibility and monitor changing conditions.
For evaluating customer understanding and experience, review market research services and contact INTAGE Vietnam with the task and decision. Confirm research scope separately from software implementation.
Related reading
Latest posts
- Market Research: Building a Market Research Report for Investment Funds
Survey Sample Size: How Many Respondents Are Enough?
Survey Data Collection Methods in Vietnam: How to Choose- Technology Adoption in Enterprises: How to Research Employee Behavior
How to Write a Market Research Brief for an Accurate Quote


