How fast should a personalization decision be?
Fast enough that ranking does not delay the page. Set a hard timeout and show a useful default when the decision misses it.
Use the smallest rule that works
Start with a useful default. Change it only when a fresh signal is strong enough to improve the decision. Keep inventory and availability as hard constraints, then measure conversion, profit, and returns against a control.
What to check before launch
Write down the rule, its data inputs, the fallback, and the condition that turns it off. Review small segments carefully. A dramatic result from a few sessions is not evidence.
Bottom line
Fast enough that ranking does not delay the page. Set a hard timeout and show a useful default when the decision misses it. Test the result against a control before expanding it.
Source: FTC, Bringing Dark Patterns to Light. Reviewed Sep 22, 2026.
Frequently asked questions
How fast should a personalization decision be?
Fast enough that the ranking does not delay the page. Set a hard timeout, and fall back to a useful default when the timeout hits. A slow personalization that delays the page is worse than no personalization at all.
What should the fallback be when personalization times out?
The smallest rule that works: a useful default, usually the best sellers or the category default, that renders instantly. The fallback is not a failure mode. It is the plan for when the decision takes too long.
How do I know if my personalization is too slow?
Measure the time from the request to the rendered page, and compare it against a control that skips personalization. If the personalized page is measurably slower, the latency budget is being spent in the wrong place.
What constraints should a personalization rule respect?
Inventory and availability are hard constraints: never recommend a product that is out of stock. Write down the rule, the inputs, the fallback, and the off-condition, and review small segments carefully, because a rule that works for a large segment can misfire on a small one.