Should quiz answers expire? Setting different shelf lives for size, style, and budget data
Quiz answers decay at different speeds. Fit and size data stays valid for years. Style preferences shift seasonally. Budget and intent answers can expire in weeks. Most personalization systems treat every quiz answer as permanent, which means the recommendations slowly drift from who the customer is now to who they were when they took the quiz. Setting a shelf life per answer category, and re-asking only what has expired, keeps the profile fresh without re-quizzing the customer from scratch.
The three speeds of preference decay
Think of quiz answers in three buckets. Stable facts change rarely: sizes, fit preferences, skin type, dietary restrictions. These are good for years and should be treated as durable profile data. Seasonal tastes shift on a cycle: style preferences, color affinities, category interests. These deserve a refresh every few months, timed to the natural calendar of the category.
Situational answers expire fastest: budget for this purchase, the occasion being shopped for, urgency. These describe a moment, not a person, and they should influence the current session far more than the long-term profile. The common failure is storing a situational answer as a permanent trait: the shopper who said their budget was tight during a sale in November gets budget messaging forever.
Designing expiry into the data model
The implementation is straightforward: every stored answer gets a category tag and a last-confirmed date, and each category has a shelf life. When an answer passes its shelf life, it stops driving personalization and becomes a candidate for re-confirmation. This is a data-model decision, not a campaign decision, so it belongs in the platform layer where every use of the data inherits it.
Be deliberate about what expiry means. Expired does not mean deleted; it means demoted. The old answer can still inform a default or a fallback, but it should not drive primary recommendations. Deletion is a separate question governed by privacy policy and retention rules. Expiry is about relevance, and relevance has a clock.
Re-asking without annoying people
The fear that stops teams from implementing expiry is quiz fatigue: nobody wants to re-take a twenty-question quiz every season. The answer is to never re-ask the whole quiz. Re-confirm one or two expired answers in context, at the moment they matter.
A shopper browsing the new season collection is the right moment to ask whether their style preferences have changed, with a single tap to confirm or update. A shopper starting a new project is the right moment to re-check budget. Contextual micro-asks get far higher response rates than re-quiz campaigns, and each confirmation resets the clock on that answer. The quiz becomes a living profile instead of a one-time event.
What to do with conflicting answers
Expiry creates a new problem: the customer gives a different answer the second time. Treat the newer answer as current for personalization, but keep the history. Answer changes are themselves a signal: a shopper whose style answers swing wildly is telling you they are exploratory, and the personalization should widen rather than narrow.
For stable facts like size, a changed answer deserves a confirmation step, since it is more likely to be a correction of an old error than a real change. For tastes, accept the change silently and move on. The system should be curious about changes, not suspicious of them.
Measuring whether expiry helps
The metric that matters is recommendation relevance over time. Compare engagement with recommendations driven by fresh answers versus answers past their shelf life; the gap is the cost of never expiring. Most teams find the gap is large and grows with customer tenure, because long-tenure customers have the stalest quiz data.
Also track re-confirmation rates on the micro-asks. High confirmation rates mean your shelf lives are about right; high change rates mean they are too long. The expiry schedule is itself something to optimize, and the customer's own corrections are the training data.
Reviewed
Published Oct 5, 2026.