The ecommerce tracking checklist
You are spending on ads. This is how to check whether the numbers you are optimising are real. Fourteen checks, in the order I run them on a client account.
Run this once a quarter, and any time you change a theme, a checkout, a tag manager container, a consent banner or an ad platform. Most accounts fail two or three checks, and the failures are almost always the ones nobody looked at since launch.
Time needed: about an hour, with a test order and two browser tabs. Access needed: GA4, Shopify orders, Meta Events Manager, GTM preview.
The audit at a glance
| # | Check | Healthy target |
|---|---|---|
| 1 | Reconcile GA4 purchases against Shopify orders | 5 to 10 percent lower in GA4 is normal, above 20 percent is a configuration error, and GA4 at roughly double Shopify means duplicates |
| 2 | Purchase carries value, currency and items | One event per order, with a non-empty items array |
| 3 | No duplicate purchase events | GA4 purchases are not about twice Shopify orders |
| 4 | transaction_id is present and consistent | The same order id in GA4 and Shopify, unique per purchase |
| 5 | The tag survives a checkout template change | purchase fires once, on the thank you step |
| 6 | One event naming style across the property | One style, usually snake_case, with no duplicates differing only by case |
| 7 | Forms and calls are measured as leads | generate_lead or a named lead event, with a source parameter |
| 8 | Pixel and Conversions API are deduplicated | Two rows for one purchase in Test Events, one Browser and one Server, counted once |
| 9 | Event Match Quality on Purchase | 8 or above, with 6 as the floor |
| 10 | Server events actually arrive | Server events visible in Events Manager, not only browser events |
| 11 | Consent Mode adjusts rather than blocks | Tags still send a cookieless ping when consent is denied |
| 12 | Denied traffic is modelled, not missing | A gap you can explain, plus modelled conversions after a few days |
| 13 | Safari cookie lifetime is accounted for | Purchase attribution on Apple devices does not collapse |
| 14 | Renewals and POS orders are excluded on purpose | The reconciliation compares like with like |
Start here
-
Reconcile GA4 purchases against Shopify orders
Healthy target: 5 to 10 percent lower in GA4 is normal, above 20 percent is a configuration error, and GA4 at roughly double Shopify means duplicates
How to see it: Shopify → Orders, Online Store channel only, last 30 complete days, same time zone as GA4. GA4 → the purchase event count for the same dates. Divide the difference by the Shopify number.
Events
-
Purchase carries value, currency and items
Healthy target: One event per order, with a non-empty items array
How to see it: GA4 → Admin → Events → purchase → parameters.
-
No duplicate purchase events
Healthy target: GA4 purchases are not about twice Shopify orders
How to see it: If GA4 is roughly double, a native GA4 integration and a GTM tag are both firing. Check the purchase page for two configuration tags.
-
transaction_id is present and consistent
Healthy target: The same order id in GA4 and Shopify, unique per purchase
How to see it: GA4 → Events → purchase → transaction_id. Compare three orders by hand.
-
The tag survives a checkout template change
Healthy target: purchase fires once, on the thank you step
How to see it: Place a test order with GTM Preview and GA4 DebugView open. If it fires everywhere except checkout, the checkout template changed.
-
One event naming style across the property
Healthy target: One style, usually snake_case, with no duplicates differing only by case
How to see it: GA4 → Admin → Events, sorted by count. Look for addToCart next to add_to_cart.
-
Forms and calls are measured as leads
Healthy target: generate_lead or a named lead event, with a source parameter
How to see it: Submit a test form, dial the tracked number, then check Realtime.
Meta
-
Pixel and Conversions API are deduplicated
Healthy target: Two rows for one purchase in Test Events, one Browser and one Server, counted once
How to see it: Meta Events Manager → Test Events, on a real purchase. Both rows must share the same event_id.
-
Event Match Quality on Purchase
Healthy target: 8 or above, with 6 as the floor
How to see it: Events Manager → your dataset → Overview → Event Match Quality, per event.
-
Server events actually arrive
Healthy target: Server events visible in Events Manager, not only browser events
How to see it: Events Manager → Overview, check the Server column for the last seven days.
Consent
-
Consent Mode adjusts rather than blocks
Healthy target: Tags still send a cookieless ping when consent is denied
How to see it: GTM Preview, then GA4 DebugView, with consent denied.
-
Denied traffic is modelled, not missing
Healthy target: A gap you can explain, plus modelled conversions after a few days
How to see it: GA4 → Admin → Data settings → Consent settings, then compare total users against Shopify sessions.
Attribution
-
Safari cookie lifetime is accounted for
Healthy target: Purchase attribution on Apple devices does not collapse
How to see it: Compare conversion share on Safari against its traffic share. A large gap means attribution is being lost, not that Safari users do not buy.
-
Renewals and POS orders are excluded on purpose
Healthy target: The reconciliation compares like with like
How to see it: Shopify orders filtered to the Online Store channel; subscription renewals counted separately, since they never pass through a browser session.
Want me to run this on your account?
A read-only audit takes three days and answers one question: is the data you are optimising real? Nothing is changed without your approval.
Or get the next teardown by email
I write one short teardown most weeks: a real account, what was broken, what the fix was. No course, no funnel.
Sources
Where a number appears in this checklist, it comes from one of these. No benchmark is invented for this page.