The dress was almost right. Show them the next four.

The fashion problem

Visitors browse five products and leave on the third.

A visitor lands on a single PDP from an Instagram ad or a Google search. The cut isn't quite right, or the colour photographs differently than expected, or the size on offer isn't their usual. They reach for the back button — and at that moment, the storefront has done one of two things. Either nothing happens and the visit ends, or a popup appears asking for an email in exchange for 10% off, which addresses none of the actual problem.

The actual problem is that the store has four other dresses, three other jackets, or two other coats in the same vibe that the visitor never reached. The catalog is large; the discovery surface isn't. That's the gap a recovery page fills, and it matters more for fashion than for almost any other Shopify category because fashion buyers are unusually willing to consider adjacent options when they see them.

How it works for fashion

Native to your theme. Trained on what actually sells.

Why this matters for fashion

Three things fashion has that other categories don't.

• Catalog depth that the visitor never sees on a single visit. A 40-product visit is a small fraction of a 600-SKU seasonal catalog. The recovery page is the cheapest place to show the rest of what the storefront actually sells, ranked by what similar visitors bought.
• A buyer who's willing to consider adjacent products. Fashion shoppers compare more than buyers in almost any other category — five to ten PDPs per session is normal. A visitor leaving the third PDP isn't done shopping; they're done with that PDP.
• Brand experience that doesn't survive popup chrome. A discount-spinner overlay reads as cheap on a £200 dress in a way it doesn't on a £20 phone case. Native theme rendering means the recovery page looks like the store, not like an ad inside the store.

• Does Before You Go work for fashion stores with size and colour variants?

  Yes. The recommendation engine works on parent products, not individual SKU variants, so a visitor leaving a black size-M dress is matched against alternative dresses by what other shoppers bought, not against other variants of the same product. The recovery page renders product cards in the same shape your theme renders collection cards, so size and colour swatches inherit the storefront's existing display logic.

• Won't a recovery page feel intrusive for a brand at a premium price point?

  The recovery page renders inside your store's own theme — same header, footer, type, URL — so it reads as a category page that quietly assembled itself rather than as a third-party popup interrupting the brand experience. There is no overlay, no dismiss button, no email capture form. For fashion brands at premium price points, that distinction matters: visitors don't see a vendor product, they see your storefront.

• What about visitors who came from Instagram or Pinterest ads — do they get the recovery page?

  Off by default for paid traffic. The middleware chain inspects the referrer and click identifiers (gclid, fbclid, msclkid, ttclid, twclid, pinclid) and skips paid traffic unless you explicitly enable it. The reasoning is straightforward: you already paid for that click. The recovery page earns incremental visits from organic search, direct, and referral traffic where the back-navigation is more likely to be discovery friction. Per-source toggles are available if you want to override the default.

• Does it work on mobile, where most fashion traffic actually lives?

  Yes — and the recovery page inherits your theme's mobile layout, so the product cards render exactly the way your collection pages do on a phone. Detection is based on the actual back-button navigation, not a cursor-leave gesture (which doesn't exist on mobile), so the trigger works identically across phones, tablets, and desktops.

• How quickly does the recommendation engine learn a new catalog?

  Same-day for content-similarity recommendations (built from product titles, descriptions, and metadata via a transformer model) and within a couple of weeks for the behavioural signals (co-views, co-clicks, co-purchases) once enough storefront traffic has been observed. Stores with sparse data fall back to popularity and content-similarity ranking; stores with rich behavioural data get the blended affinity score.