Returning Visitors Who Never Log In: Personalisation That Works on a Cookie and a Guess
On a typical store, the logged in shopper is the exception. The returning visitor the team talks about in planning, the one who browsed running shoes last Tuesday, arrives today as a cookie, a device, and a handful of viewed product pages, if she arrives on the same device at all. Personalisation built for identified customers does nothing for her. Personalisation built honestly for anonymous returners can still do useful work, provided the store is clear eyed about how thin the evidence is and how quickly a wrong guess costs more than no guess at all.
What the last session can actually tell you
A prior session gives you categories viewed, price band touched, a size or variant selected, and whether the visitor reached checkout. That is enough to continue a task: resurface the category, keep the size preselected, show the delivery promise that was checked last time. It is not enough to infer intent, budget, or identity. The stores that get this right treat the signal as a shortcut back to where the visitor left off. The stores that get it wrong treat it as a profile, greeting a shared family tablet as if it were one person with one taste, and recommending gifts bought for someone else for the next six weeks.
The gift purchase problem never goes away
One out of pattern purchase poisons anonymous recommendations longer than it should, because the system cannot tell a gift from a conversion. Weight recent browsing over a single past order, decay signals over weeks rather than months, and give the visitor a visible way to dismiss a rail that has latched onto the wrong interest. A simple not interested control, used by a small share of visitors, returns cleaner data than any model tweak, because it replaces an inference with a statement. Stores resist adding it because dismissal rates look like failure in a dashboard. They are the opposite.
Continuity beats cleverness
The highest returning anonymous gains are unglamorous. Remember the store or language selection. Keep the recently viewed rail accurate and in the same place. Do not restart a size guide or a fit quiz the visitor already completed. Hold the sort order and filters from the last category visit. Each of these saves the visitor redoing work, which is what returning actually means to her. Recommendation models get the budget because they are demonstrable in a sales deck. Continuity features get skipped because they look like housekeeping, and then the store wonders why its personalisation programme tests well in segments and disappoints in blended revenue per session.
Measure it split, or do not claim it
Anonymous personalisation must be measured on returning anonymous sessions specifically, split by device, with new visitors held out. Blended figures hide the two common outcomes: a small real gain for the group the feature serves, and a drag on everyone else from slower pages and displaced merchandising. Watch revenue per session and time to first product click, not just rail click through. A rail can win clicks and lose sales by pulling visitors out of a category they were about to buy from. If the team cannot produce the split, the programme is running on faith and vendor screenshots.
Keep reading
Related guides on this site: Zero Party Data That Customers Will Actually Give You Personalizing Category Pages for Visitors You Know Nothing About Practical answers about personalization.
Frequently asked questions
Can you personalise for visitors who are not logged in?
Yes, within limits. Prior session behaviour stored in a cookie, categories viewed, variants selected, recently viewed items, supports continuity features and light recommendations, but it should be treated as a weak signal.
Why do recommendations keep showing a gift I bought once?
Anonymous systems often weight a purchase heavily and decay it slowly, and they cannot tell a gift from a personal preference. Recent browsing should outweigh a single old order, and a dismiss control helps correct it.
Does personalisation slow a store down?
It can. Client side recommendation calls and flicker prevention scripts add weight and delay. Measure page performance for personalised sessions separately so the cost is visible next to the gain.