Cold Start Is Not a Data Problem. It Is a First Screen Problem.

A first time visitor gives you a landing page, a referrer, a device, and about eight seconds. Cold start personalization is deciding what those four things are allowed to change, and what they are not.

Cold start gets described as having no data, which is wrong in a useful way. A first visit carries a landing page the visitor chose from a search result or an ad, a referrer that hints at intent, a rough location, a device, and the first two clicks, which are worth more than a month of third party segments ever were. The stores that handle cold start well treat those signals as a narrowing device, not a portrait. Landed from a search for waterproof boots in a rainy region on a phone: show the waterproof range, in stock, with delivery times up front, and stop there. Do not infer a price band, a gender, or a lifestyle from one click and rebuild the store around the guess. Cold start errors compound because the system then measures the visitor's response to its own guess and calls the measurement a preference. The visitor who bounces off a wrongly narrowed homepage is recorded as not interested, when the record should say the store guessed wrong in the first eight seconds.

Borrow popularity, but borrow it locally

The standard cold start answer is bestsellers, and the failure mode is borrowing the whole store's bestsellers for a visitor who landed deep in one category. Popularity should be scoped to the context the visitor arrived in: best in this category, best from this collection, what people who landed on this page bought next. Shopify's own personalization guidance and the recommendation tooling in its Search and Discovery app both lean on that scoping, because a global bestseller list on a niche landing page mostly tells the visitor the store did not notice where they came in.

Session signals should outrank arrival signals within minutes. Two product views in one category outweigh the referrer. A search inside the session outweighs both. Rank your signals by freshness and intent, query beats click, click beats referrer, referrer beats geography, and let each stronger signal retire the weaker ones instead of blending them into a soup where the ad campaign from three weeks ago still shapes the homepage.

Related reading on this site: PersonalizeIQ and Search Shoppers and Browsers Want Opposite Things From Personalization cover the neighbouring problems in more depth.

What not to personalize on a first visit

Prices, discounts, and urgency. A first visit personalized discount trains exactly one behavior: leave and come back as a stranger to see if the offer improves, and it prices your margin against a competitor the visitor may not have open. Keep first visit personalization to assortment, ordering, and information: which products lead, which content explains, which delivery promise shows. Those help a visitor decide. Price games ask a visitor to negotiate, and most would rather just buy the boots.

Also leave the navigation alone. Reordering a store's categories around a guessed profile on visit one breaks the one mental model the visitor is actively building. Personalize the shelves before you move the aisles.

The handoff to known visitor

Cold start ends the moment the visitor gives you something durable: an email, an account, a purchase, or a second session the platform can tie to the first. Design that handoff deliberately. The preferences implied by session one, categories viewed, price points clicked, sizes filtered, should be stored against the identity when it appears, not discarded because they predate the signup. Stores routinely keep the purchase and lose the browsing that explained it, then wonder why their known visitor personalization feels thinner than the anonymous session did.

Frequently asked questions

What is cold start in ecommerce personalization?

It is the problem of personalizing for a visitor the store has no history on. In practice a first visit is not dataless: landing page, referrer, device, location, and the first clicks and searches in the session are usable signals if they are ranked by intent and freshness.

What should a store show a first time visitor?

Assortment scoped to the context they arrived in, such as best sellers within the landing category rather than the whole store, with stock and delivery information prominent. Session behavior like searches and product views should quickly override arrival signals.

What should not be personalized on a first visit?

Prices and discount offers are poor first visit levers because they train visitors to game stranger status, and navigation structure should stay stable while the visitor is still learning the store. Assortment, ordering, and information are safer and more useful.

When does cold start end?

When the visitor provides a durable identifier or a linkable second session: an email signup, an account, or a purchase. The browsing signals from the anonymous session should be attached to that identity rather than discarded.