The famous retail-prediction story in my notes is usually told as a demonstration of marketing intelligence. I take a different question from it: should a business act on every inference it can make?
Separate a need from a private inference
Someone asking about delivery times is giving you information relevant to a purchase. Guessing a major personal event from their shopping behaviour is a different activity. Accuracy alone does not make an inference welcome.
Before adding a data field or segment, name the customer benefit. “We can target them more precisely” describes a benefit to the advertiser. “We can show the compatible replacement part they requested” describes a benefit the customer can understand.
Design a smaller research question
Start with information the customer intentionally provides about the job they are trying to do. Ask about obstacles, desired outcomes, timing and selection criteria. Avoid turning a general marketing survey into a collection of intimate personal details.
For example, a property-service enquiry can ask which service the person needs and when they need it. It does not need to infer a relationship problem to offer a helpful next step. The same campaign can be relevant through the service context rather than through a hidden personal profile.
Ask how the message would feel
Read the proposed message as the recipient. Would they understand why they received it? Would the wording expose an assumption they never knowingly shared? Could the same useful offer be made with less data?
This is a design review, not a substitute for the privacy obligations that apply to the business. It is a way to challenge the premise before building the targeting system. Sometimes the best customer insight is recognising the boundary between relevance and intrusion.