.svg)
A click-rate model, not a guess
Every grid position has an expected click rate, built from your own store's click curve by device — the chart literally shows “where shoppers click” by row, indexed against the page-1 average. A product's lift is its actual clicks compared to what its position predicts. That lift is what separates “this product sells itself” from “this product looks good because it's in a good spot.”
The same model renders as a heatmap of the grid itself — “the grid, as shoppers saw it” — so you can see at a glance which tiles are over- or under-performing their position, by device, for page 1 over the last 28 days.
Three findings, each with an action
- Hidden gem (
buried_not_dead) — a product getting more clicks than its position predicts, or selling elsewhere in your store, but sitting too low in the grid for most shoppers to scroll to. The suggested action is to move it up. - Clicked, not bought (
clicked_not_bought) — a product that gets plenty of clicks but rarely converts into an order. This one doesn't suggest reordering the grid — it suggests recovering the shopper, with a one-click path to create a recently-viewed offer for exactly that purpose. - Weak top spot (
overexposed) — a product holding a prime position but earning fewer clicks than that position usually produces. The suggested action is to swap it out.
Every finding shows its evidence alongside the recommendation — views, clicks, store-wide orders, and the lift number itself — so the suggestion is never a black box.
Campaigns, ads and paid traffic
Collection sessions are attributed to a landing source — paid social, other paid ads, search, social, affiliate, email, SMS, or direct — and rolled up by campaign (utm_campaign) and ad (utm_content), ranked by session volume, each with its most-clicked products. A separate view compares what paid-traffic shoppers click most against everyone else's click rate on the same products, across every collection — the product your paid ads are actually driving interest in, independent of what the grid position suggests they should click.
How shoppers arrive, and what they do next
For any single collection, a funnel breaks sessions down by landing source into clicked, added to cart, and ordered — so a collection with strong sessions but a weak funnel from one channel is visible immediately, not buried in an aggregate conversion rate.
Self-serve segments
Every view can be filtered by date range (7, 28, or 90 days), device (mobile, desktop, or both), and traffic channel, so a question like “how does this collection perform for paid social shoppers on mobile in the last week” is a filter change, not a data request.
Turning it on
Collection Intelligence runs on a theme app embed setting — “Track collection page product grids,” inside the same Traffic & Browsing embed used elsewhere in the storefront — so there's no separate script to install. Data appears the day after shoppers start browsing. Findings compare a 28-day window, so a newly-enabled shop sees a running day-count while evidence accumulates before the first findings firm up.
