First, place a test order and follow it from push to report
Use a real checkout path with a low-value product or a test payment method, not a simulated page load, because
several causes live in the payment hand-off. Open Tag Manager (GTMTag management system.) preview before you start so
that the confirmation page is captured. When the order completes, check four things in order. Did the browser
land on the confirmation page at all, or stay on the payment provider? Does the Data Layer tab show a push with
event: "purchase" and an ecommerce object carrying transaction_id, value, currency and items? Did the
GA4 event tag fire on that push, once? Does the outgoing request to the property carry the item and value
parameters?
Then open DebugView in the GA4Google's event-based analytics platform. property and confirm the purchase arrives with its parameters. Finally, after processing, look for the order in the e-commerce purchases report and the key events report. The step where the order vanishes decides the row of the table. If purchases do appear but the totals disagree with the shop, that is a reconciliation question for common GA4 mistakes, not this page; and if a single field reads undefined while the event fires, the page on data layer variables that are undefined is narrower and quicker.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Purchases missing mainly for one payment method; the test buyer stays on the provider's page | Whether the provider redirects buyers back, and how many orders lack a confirmation page view | Buyers never returning to the confirmation page after an external payment | Enable the provider's return redirect, or send the purchase from the order system |
| Push visible in the Data Layer tab, but the tag fired earlier or not at all | Event timeline order: the tag's trigger against the purchase push | Purchase pushed before the container loads or after the tag reads it | Trigger the tag on the purchase custom event, and push before any redirect |
| Tag fires, but items are empty or the event is rejected in DebugView | The shape of the ecommerce object and each item against the documented fields | Items array or ecommerce object malformed | Push the documented structure and enable the tag's e-commerce data option |
| Purchases counted with no revenue | value type and currency code on the request | Value or currency missing, so revenue reads zero | Send a numeric value and an ISO currency code on every purchase |
| Visible in DebugView, absent from key event reports or the Ads import | Key event list; data filters and their state in admin | Purchase not marked as a key event, or excluded by a data filter | Mark purchase as a key event; correct the filter or test outside it |
| Purchases short by a share close to the banner's decline rate | Consent state on the confirmation page for test orders that accept and decline | Consent denied for most buyers | Confirm the behaviour matches policy and the consent implementation |
Fix, verify and prevent each cause
Buyers who pay elsewhere and never come back
When checkout hands the buyer to a payment provider, the confirmation page only loads if the provider sends them back and they wait for it. Many close the tab once the bank approves the payment, or the provider shows its own receipt page. The order exists; the page that would have announced it never loaded.
Fix: turn on the
provider's automatic return to the shop where it is offered, and for orders that still never return, send the
purchase from the order system with the same transaction_id, as the purchase step of the
e-commerce tracking tutorial allows for events without a page.
Verify: the share of orders with a matching purchase rises, measured over a closed week against the order list.
Prevent: test every payment method in the release checklist, not only the card form on the shop's own domain. Where the whole booking or checkout runs inside a third-party widget or hosted page, the guide to tracking embedded widgets and hosted pages sets out the options for each situation.
The tag reads the order too early
A purchase tag attached to a page view or DOM ready trigger can fire before the site has pushed the order, and then reads nothing. The opposite also happens: a confirmation page that pushes the order and redirects immediately may unload before the request leaves. The data layer guide recommends pushing when the data is true and letting tags react to the push's own event.
Fix: push the order with
event: "purchase" as soon as the confirmation template has it, trigger the GA4 tag on that custom event, and
avoid redirects between the push and the page's settling.
Verify: in preview the tag fires on the purchase event, after the push, exactly once.
Prevent: keep e-commerce tags on custom event triggers only, and say so in the data layer contract.
An order object in the wrong shape
GA4 reads e-commerce parameters from a documented structure: an ecommerce object holding the order fields and
an items array of objects. The GA4 developer guide
documents that each item needs at least an identifier or a name. A flat object without the ecommerce key, an
items value that is a string, an empty array or items without identifiers all arrive as a purchase with
nothing useful in it, or not at all.
Fix: push the documented structure, clear the previous object with
ecommerce: null first, and enable the GA4 tag's option to send e-commerce data from the data layer, as the
e-commerce tracking tutorial sets out.
Verify: DebugView shows the purchase with its items expanded and each item carrying an identifier.
Prevent: validate the purchase payload against the contract in automated tests on the confirmation template.
Revenue with no value or currency
A purchase can be counted while revenue reads zero. The usual reasons are a value sent as text with a currency symbol in it, a value field left empty on some templates, or no currency code. The developer guide documents currency as required whenever value is sent, so a missing code leaves revenue unrecorded. Average order value (AOVAverage revenue per order.) and every revenue report then read low or blank.
Fix: send value as a number and
currency as a three-letter code on every purchase, following the agreed value convention in the
e-commerce tracking tutorial.
Verify: the request shows a numeric value and a currency, and the purchase revenue metric is populated after processing.
Prevent: include value type and currency in the payload validation, and record the tax and shipping convention in the contract.
Counted, then hidden by settings
Some purchases reach the property and still do not appear where people look. If purchase is not marked as a key event, it is missing from the key events report and cannot be imported into Google Ads; the key events guide explains how marking works and that it applies going forward. Data filters remove traffic as well: an active internal traffic filter hides orders placed from the office, and a developer traffic filter hides events sent in debug mode, which is why test orders can appear in DebugView and nowhere else.
Fix: mark purchase as a key event if it is not, and check each active filter against how the test orders were placed.
Verify: a test order placed outside the filtered traffic appears in the reports after processing.
Prevent: keep a list of the property's key events and filters in the measurement documentation, and test from a network and device the filters do not cover.
Buyers who said no
Under consent mode, a buyer who declines storage either sends nothing, in the basic pattern, or sends signals without identifiers that the property may model, in the advanced pattern. If most buyers decline, purchases in the reports fall short by roughly that share. That is the consent choice working, not a broken tag.
Fix: confirm with a test order in each state that accepted orders send a full purchase and declined orders behave as the policy intends, using the checks in the consent mode guide; if accepted orders are also missing, the update after acceptance is the fault.
Verify: the gap between shop orders and reported purchases is close to the decline share, not larger.
Prevent: annotate the date the banner changed, and read purchase trends alongside the consent rate.
Prevention: a checkout test in every release
Purchases break silently, usually in a release that changed something nearby: a checkout redesign, a new payment method, a template refactor or a consent banner update. The defence is a test order on every release that touches the shop, run through each payment method, with the four checks from the first section: confirmation page reached, push present and well formed, tag fired once on the purchase event, and the event visible in DebugView with value, currency and items. Automate what can be automated: a payload check on the confirmation template and a daily comparison of reported purchases against the shop's order count for the previous day.
Make the comparison a standing alert, with a tolerance the team agrees in advance that reflects the consent rate and orders that never return from payment. When the gap widens beyond it, the newest release is the first suspect. Write the purchase contract down, including the structure, the value convention, the transaction identifier and the trigger, so a developer changing the confirmation template knows what must survive. The analytics troubleshooting hub routes neighbouring symptoms, such as a property that records nothing at all. When the confirmation template is being changed anyway, the guide to confirmation and thank-you pages covers what the page should tell the buyer alongside the purchase it reports.
A homeware shop with half its purchases missing
A homeware shop takes 1,240 orders in March, but GA4 records 610 purchases, and 140 of those show no revenue.
A test order by card on the shop's own page tracks correctly. A test order through a buy-now-pay-later provider
leaves the buyer on the provider's success screen, with no return to the confirmation page; the order list shows
that payment method accounts for about 520 orders. The 140 zero-revenue purchases all come from the mobile
template, where value is pushed as the text "£42.00". The team enables the provider's automatic return, sends
purchases for orders that still do not return from the order system with the same transaction identifier, and
changes the mobile push to a number with currency: "GBP". In April, GA4 records 1,090 purchases against 1,200
orders, a gap close to the banner's decline rate, and every purchase carries revenue.
Common mistakes
Testing with a page reload instead of a real checkout. Triggering the purchase tag on page view. Reading DebugView as proof the reports will show the order. Sending value as text. Fixing a consent gap by bypassing consent.
