First, reconcile one closed day order by order
Totals cannot tell causes apart, because a repeated purchase and a missing one cancel out. Pick a recent closed day, so late processing and late refunds have settled, and export two lists: the shop's orders for that day, with order number, value, tax, shipping, discount, currency, status and sales channel; and GA4's purchases for the same day, with transaction identifier, revenue and currency, from an exploration or from the BigQuery export if it was running. The common GA4 mistakes page recommends this order-by-order check because it usually names the cause quickly.
Join the lists on the order number. Every order falls into one of four groups: matched with the same value, matched with a different value, in the shop only, or in GA4 only or more than once, and each group points to different causes below. If GA4 shows no purchases or zero revenue at all, the problem is collection, covered by purchase event not tracking. If every channel's claimed sales add up to more than the shop sold, see channel conversions that exceed sales; Google Ads has its own routine in reconciling conversions with sales, and the analytics troubleshooting collection routes other measurement symptoms.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| The same transaction identifier appears more than once in GA4 | Purchase events per identifier; the confirmation page's reload and return behaviour | Purchase sent again on a reloaded or revisited confirmation page | Guard the purchase push so each order is sent once |
| Matched orders differ by a steady share, or only on some templates | Each matched order's value against its item total, tax, shipping and discount | Value including tax, shipping or discounts differently from the shop's revenue | Agree one value convention and send it from every template |
| Revenue blank or far off for orders in some currencies | The currency on each purchase against the order's currency | Currency missing or different from the order's currency | Send the order's own currency code with every value |
| GA4 above the shop's net figure, gap growing over the month | Refunded and cancelled orders against refund events with the same identifier | Refunds and cancellations recorded in the back office only | Send refunds with the original identifier, or document them as out of scope |
| Orders in the shop only, spread across channels and days | Payment method and return path of the missing orders; the period's consent rate | Orders the browser never reports: declined consent, blocked tags, buyers who stay on the payment provider | Fix the return path; document the consent and blocker gap |
| Orders on one side only, clustered by type, person or channel | Order source, customer and test markers against GA4's filters | Different scope: test and staff orders, or phone, marketplace and in-store orders on one side only | Align scope on both sides and record what each excludes |
| Daily totals disagree but weekly totals agree | Order timestamps near midnight against the property's time zone | Different day boundaries and time zones | Align time zones or compare in the zone of record |
Fix, verify and prevent each cause
A purchase sent again from the confirmation page
A purchase pushed every time the confirmation page loads is sent again when the buyer refreshes, opens the page from a bookmark or an order email, or comes back from a payment provider. The e-commerce tracking tutorial notes that the developer guide documents the transaction identifier as helping GA4 avoid counting a purchase twice, and warns that event counts and destinations such as a Google Ads import can still see the repeat. A second route to the same symptom is a platform integration and a Tag Manager tag both sending the purchase, which DeduplicationRemoving double counts when one conversion is reported by more than one source. on the identifier should not be trusted to undo.
Fix: guard the push on the server or with a flag stored against the order, so the confirmation page sends the purchase once per order, and remove any second integration sending the same orders. The guide to confirmation pages covers counting each conversion once from the page's side.
Verify: reload a test order's confirmation page, reopen it from the email, and confirm the event appears once in DebugView.
Prevent: add the reload and revisit cases to the release test for checkout changes, and write down which single route sends purchases.
A value built differently from the shop's revenue
The developer guide's convention, as the e-commerce tutorial applies it, sends value as the item total and tax and
shipping as separate parameters. A shop that reports revenue including tax and delivery, a template that adds
shipping to value, or a discount applied on one side and not the other produces matched orders whose values differ.
A gap that stays the same share month after month is usually this, and the
metrics layer and definitions guide owns the wider point: revenue is a
definition that must say what it includes.
Fix: agree one conversion valueThe monetary value an advertiser assigns to a conversion, such as the order's revenue or a lead's expected worth. convention, write it into the data layer contract, and send it from every template; compare GA4 with the matching line of the shop's report, not with a total built on a different rule.
Verify: matched orders in a closed day agree to the order line on value, template by template.
Prevent: keep the convention in the metric's written definition and check one order per template whenever checkout code changes.
A missing or wrong currency
The developer guide documents currency alongside value, and the
purchase event not tracking page explains how a missing
code leaves revenue unrecorded. In a shop that sells in more than one currency, a hard-coded code labels some orders in a currency
they were not paid in.
Fix: send the currency code of the order itself with every purchase and refund, read from the same source as the value, as the e-commerce tracking tutorial sets out.
Verify: place a test order in each currency the shop accepts and confirm the purchase carries the matching code.
Prevent: add any new currency to the release test the day it launches.
Refunds that stay in the back office
Refunds and cancellations usually happen in the order system rather than in a browser, so nothing sends them unless
someone builds it. The e-commerce tutorial sets out the event: refund with the original transaction_id, currency
and value, plus the refunded items for a partial refund, sent from the order system or a scheduled job. Without it,
GA4 keeps the original revenue for good and stays above the shop's net figure.
Fix: send refunds with the original identifier, or decide that refunds are out of scope and say so in the definition, so readers compare GA4 with gross rather than net revenue.
Verify: refund a test order and confirm the refund arrives with the purchase's identifier.
Prevent: check any new refund route, such as a partial refund from customer service, for the event before it goes live.
Orders the browser never reports
Three groups of real orders never reach GA4 as observed purchases. For buyers who decline consent, the privacy and consent guide and the Consent ModeA Google tag feature that adjusts what tags collect according to the visitor's consent choices. developer page describe tags that stay silent or send limited signals that may be modelled in aggregate. Tag blockers send nothing. Buyers who pay on a provider's page and close the tab never load the confirmation page, which the guide to embedded widgets and hosted pages covers.
Fix: turn on the payment provider's return to the shop where it is offered and list the provider as an unwanted referral, as that guide describes; for the consent and blocker share, fix nothing, because the gap is the consent choice working.
Verify: missing orders split by payment method no longer cluster on one provider, and the remaining missing share moves with the consent rate.
Prevent: record that share as part of the tolerance below, and annotate every banner or payment change.
Orders that belong to one side only
The shop's list often holds orders the website never saw: phone orders keyed in by staff, marketplace orders, in-store sales on the same system. GA4 may hold test orders, staff orders and orders from a staging site reporting to the production property. The data quality guide lists internal and test traffic among the usual faults and asks for comparison against the source of truthThe system whose figure is treated as authoritative for a given number, such as billing for revenue or the CRM for pipeline, with other tools reconciled to it. on a defined scope.
Fix: filter the shop's list to website orders before comparing, mark test orders so they can be excluded, and keep staging out of the production property.
Verify: orders on one side only fall to the expected kinds, each with a reason.
Prevent: write the scope of each side into the definition and add a new sales channel to it on the day it opens.
Day boundaries in different time zones
An order placed late in the evening can land on one date in the shop and the next in GA4 when the two systems keep different time zones, which the platform versus analytics discrepancies guide explains for every pair of systems. Daily totals then disagree at the edges while weekly totals agree.
Fix: set the property's time zone to the shop's where the business allows, or compare in one zone of record and say which.
Verify: orders near midnight fall on the same date in both lists.
Prevent: record both time zones in the measurement plan and check them whenever either system is set up or moved.
A prevention routine: a monthly reconciliation with a tolerance
The order-by-order check finds the cause once; keeping the gap understood takes a routine, run after each month closes and written down even when nothing has changed. Use the tracking verification checklist for the release tests that stop faults appearing, and this routine for noticing the ones that slip through.
| Part of the gap | Treatment | Where it is recorded |
|---|---|---|
| Repeated purchases | Fault: fix and re-test | Change log, with the date fixed |
| Value or currency differences | Fault: fix the convention or the code | Metric definition and data layer contract |
| Refunds not sent | Fault, or a documented scope choice | Metric definition |
| Consent, blockers, buyers who never return | Expected: document | Tolerance, read against history |
| Orders outside the website's scope | Expected: exclude before comparing | Scope note in the definition |
| Time zone edges | Expected on daily figures | Measurement plan |
Each month, compare the shop's website orders and revenue with GA4's for the closed month, by day and by payment method, and record the ratio of GA4 to the shop beside previous months. That history is the tolerance: the business's own normal range, read with the care the small numbers guide recommends when counts are low. A ratio that leaves the range, or moves on the date of a release, a banner change or a new payment method, starts the order-by-order check. A gap above the shop points to duplicates or refunds, and a gap below to collection or consent.
When an investigation finds a fault, ask why more than once. The five whys takes a finding such as "purchases repeat on reload" back to the reason the release test missed it, which is the part that stops it returning. If the reconciliation runs against exported rows that disagree with the interface, the BigQuery export comparison covers that gap. Keep the value convention with its date, a log of tracking and checkout changes, and the monthly ratio table beside the routine, so "why does GA4 not match the shop?" has an answer on file.
A candle shop finds its missing revenue (illustrative)
Ravvick Candles reconciles a closed day
Ravvick Candles, an invented online candle shop, sees GA4 revenue above its order system in some weeks and below it in others. For one closed Tuesday the shop records 84 website orders worth £3,276 in item value, while GA4 shows 91 purchases worth £3,511. Joining on order number shows 71 orders matched exactly, 9 identifiers sent twice by buyers reopening the confirmation page from their order email, 6 orders whose value includes £4.95 shipping because the gift-set template adds it, and 13 orders with no purchase at all. The developers guard the push with a flag stored against the order, the gift-set template is changed to send the item total, and a daily job starts sending refunds after the team finds 5 that week never sent. The 13 missing orders cluster on no payment method and track the share of visitors declining consent. The next closed month shows GA4 at 88% of the shop's website revenue, inside the previous quarter's range of 86% to 90%, which the team records as its tolerance.
Common mistakes
- Comparing totals instead of orders. Repeats and missing orders cancel out, and the totals hide both.
- Reconciling an open day. Late processing and late refunds still move the figures.
- Relying on the transaction identifier to stop duplicates. Guard the push so each order is sent once.
- Comparing against revenue built on a different rule. Agree what value includes before judging the gap.
- Forgetting refunds. Without refund events, GA4 stays above the shop's net revenue for good.
- Trying to fix the consent gap. Declined consent is a choice to respect and document, not a tag fault.
- Keeping no history. Without the usual ratio on file, nobody can tell a new fault from the normal gap.
