Why do attribution models disagree for a fashion store?

Usually because a fashion purchase mixes steps that each rule weighs differently. If buyers watch a creator, browse a few times and buy from a sale email, last click credits the email. Meta may credit the ad view, and data-driven splits the visits between. Steps nobody records get no credit from any model.

If you sell fashion

If you sell clothes, shoes or accessories, a purchase often starts as a picture. Think of a look on a creator's feed, an outfit on the street, an ad between stories. Then come the checks: the size chart, the reviews, the returns page, perhaps a wishlist. Then the nudge: a restock alert, a sale email, a branded search on payday.

Each rule picks a different hero from that story. Last click crowns the closer, often the email or the branded search. First click rewards the opener, if the opener was a click at all. Data-driven spreads credit over the visits GA4 recorded. Meta can count the sale after a view of its own ad. Four rules, four winners, one pair of boots.

Store A is one store, and its export does not say what it sells, so read it for the mechanics, not for fashion.

One store's anonymised GA4 export, 1 January 2024 to 21 August 2026. It holds shares of revenue only: no ad spend, no order counts.

The three views agree on every row printed here. On the Channels sheet, Direct holds 57.7% under every rule, and Paid Social and Organic Video hold 0.0% under every rule.

That second agreement is the one to worry about if you sell fashion. A channel no journey recorded gets zero from first click, last click and touched alike. The export cannot say whether Store A ran social ads or posted videos at all. But if your persuasion happens in social video, every model can agree on zero and be wrong together.

On the Journeys sheet, 1-touch journeys hold 79.5% of revenue at 0.5 days to buy. Journeys with 2 to 3 touches hold 12.2% at 12.5 days, also on the Journeys sheet. The models can only disagree inside journeys with more than one touch.

If you sell fashion, your multi-touch share may be bigger than this one store's. Fit, price and sale timing bring people back before they pay. Check your own path lengths before you expect the models to agree.

What changes for a fashion store?

Sales and drops shorten journeys. If you run drops or sales, many buyers may come straight from an email or text and pay in one visit. One-touch journeys leave the models nothing to argue about, so they agree that week. In quiet weeks, journeys may stretch and the models drift apart. Compare models over the same weeks, or the calendar will pose as a model problem.

Windows catch some journeys and drop others. If a buyer clicks an autumn ad, browses for weeks and buys in the sale, GA4 can still see that click. Its default lookback for purchases is 90 days. Meta usually counts 7 days after a click. So the same sale can show a Meta touch in GA4's paths and no Meta purchase in Ads Manager. That is the default Meta attribution setting at work.

Creator posts count only when clicked. An unpaid creator post that nobody clicks leaves GA4 nothing to credit. Meta's view window covers your ads, not someone else's posts. If creators drive your discovery, every model undercounts them in the same direction.

What to do this week

1. Compare models in a quiet month and a sale month. In GA4, open Advertising > Attribution > Attribution models, pick purchase and set Data-driven beside Paid and organic last click. Read the % Change column for a quiet month, then for your last sale month. Pass: the gaps look alike. Fail: the models agree in the sale and split in the quiet month, so judge prospecting channels on quiet months.
2. Split the comparison by device. In the same report, click Add filter and create an Include filter on Device category, first for mobile, then for desktop. Pass: the two models give similar shares on both. Fail: they differ much more on one device, so read that device's paths before you move budget.
3. Check how long your multi-touch buyers take. In Advertising > Attribution > Attribution paths, set Path length to greater than 1 and read Days to key event in the top row. Pass: it sits inside a week, so Meta's usual click window and GA4 see the same journeys. Fail: it runs longer, so expect Meta to miss sales that GA4's paths still show with a Meta touch.

Check the homework. Your GA4 Attribution paths export already holds the evidence. Causality Engine reads that one file and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes, for €99 once (excluding VAT), refundable within 30 days. Check the homework

Sources, 1 October 2026: Get started with attribution (Google); Select attribution settings (Google); About attribution models and attribution settings (Meta); Ad Account Insights reference (Meta); Key event attribution models report (Google); Key events attribution paths report (Google); Collect campaign data with custom URLs (Google)

Related answers

• All Attribution models questions
• Why do attribution models give different answers?
• How to trace why attribution models disagree, step by step
• Attribution window, defined