Filed under

attribution

Order slip with a punched hole, tied to a price tag and bead chain. The ad-card cord stops short. A thinner cord ends at a shorter coin stack.

Ecommerce data and measurement

Pick an influencer platform by order ID and payout

A code, a link and a partnership ad record different events. The buying check is whether a platform can name the Shopify order, show which evidence won, and match that credit to the payout.

Muted paper and ink carry one blue line. A receipt lies under two offset invoices, a sealed tin holds one slip, and a cable runs from a server block to a shopping bag.

Ecommerce data and measurement

How to judge an affiliate platform from the order record

Judge a platform on one paid order: who can take a commission, and whether the row matches the shop. Awin's de-duplication page, listed as published on 9 March 2026, automatically credits one Awin partner only when you are not running partner activity on another affiliate network. impact.com requires im_ref in a first-party cookie or on the server, and the cookie's lifetime must match the referral window in the contracts. A successful call is queued for processing.

Two checkout rails: agent-completed orders lack session IDs, while in-app merchant checkout carries client and session stamps into GA4.

Ecommerce data and measurement

How to tag agentic checkout orders so they show a source in GA4

There is no single agentic checkout to tag. In-chat completion has no merchant page for gtag to fire. A Shopify ChatGPT path still hits your store checkout in an in-app browser. The GA4 job follows that split, and any source you write is a merchant convention.

Balance scale with mixed ad-channel tiles crumbling on one pan and a single campaign weight on the other, set over a map of geo holdout regions.

Ecommerce data and measurement

How to judge campaigns when attribution fragments

Google's PMax channel report, GA4's AI Assistant channel and Klaviyo delivery look like scorecards. They are mostly diagnostics. Here is what each can decide, what still needs an experiment, and what the public evidence cannot support yet.