7 min read
In brief
- One real order, traced end to end, answers the question. A dashboard showing an active pixel answers nothing — every failure mode in this guide looks green there.
- Three numbers must agree before you scale: the order exists in your store, the platform counted exactly one purchase for it, and the value equals what the customer paid.
- The order of checks matters. Value and deduplication first, then consent, then attribution — fixing them in the wrong order produces a test that passes and a month that lies.
The short answer
Place one real order you control, then trace it through both paths: the browser event and the server event. Confirm three things — the order exists, the platform recorded exactly one purchase for it, and the value matches what the customer paid.
That test takes about ten minutes and it catches the failures that cost money. Everything else in this guide is detail about what to check inside those three numbers.
Why a dashboard cannot answer this
Events Manager will happily show a pixel as active while sending two purchases per order, a wrong currency, or a value from the product page rather than the payment. The dashboard reports that something arrived; it does not compare that something to your order table.
The three numbers that must agree
| Number | Where it lives | It is wrong when |
|---|---|---|
| The order | Your store's order record: id, total, currency, timestamp | No id, a total that excludes shipping or tax while the ad platform includes it, or a currency that differs from the feed |
| The counted conversion | The ad platform: one purchase, same value, same currency, same day | Two purchases, a value in minor units, or a conversion attributed to a different day |
| The customer's payment | The processor: what was actually captured | The value sent to the platform is a displayed price rather than the captured amount |
When all three agree, you have a measurement you can scale on. When two agree and one does not, you have a report you will argue with in six weeks.
The pre-spend checklist
- 1
Confirm the pixel fires on the pages that matter
Home, category, product, cart, checkout, confirmation. Not just the homepage — a pixel that loads once and never fires an event is the most common false pass.
- 2
Confirm the purchase event fires after payment, not on redirect
A purchase fired when the customer is redirected to the payment provider counts orders that never completed. Fire it when the order exists and is paid.
- 3
Confirm value and currency come from the order
The value should be the amount actually charged, in the store's real currency, as a number rather than a formatted string. Minor units sent as major units is the most expensive silent bug in this list.
- 4
Confirm one id is shared by both paths
Browser and server must send the same event id and the same event name, or Meta will count two purchases. If there is no server event yet, add one before you scale, not after.
- 5
Confirm what fires before consent
If you sell to the UK or EU, note which events fire before a consent choice and whether consent defaults are set before any tag loads. This is a compliance question and a data-quality one.
- 6
Confirm attribution survives the journey
Land on the site with utm_source and fbclid set, then navigate to a product and to checkout. The click id should be captured into a cookie; the utm parameters should still be visible somewhere on the way.
- 7
Confirm GA4 tells the same story
If GA4 exists, the same order should appear there with the same value. Two systems that disagree by 40% mean one of them is wrong and you do not yet know which.
- 8
Write the baseline down
Before spending, record orders per day, revenue per day, sessions, and events per session for the last two weeks. Without a before, no after means anything.
What to check in GA4 once it is installed
GA4 fails differently from an ad platform. It will not tell you that a purchase was counted twice; it will quietly file the purchase under the wrong event, the wrong currency, or no campaign at all. Six checks cover most of it.
- DebugView, not Realtime. DebugView shows the event parameters as they arrive, which is how you confirm that purchase carries a value, a currency and a transaction id. Realtime only proves something arrived.
- Key events, not conversions. If purchase is not marked as a key event, it will not appear in the reports you are about to trust. Do this before you evaluate a single campaign.
- The transaction id is unique. Two purchases with the same transaction id means the event fired twice — usually a thank-you page that reloads, or a theme and an app both reporting.
- Currency and value match the order. Check that the currency parameter matches the store currency and that value is the amount charged, not the cart subtotal before shipping.
- Consent mode shows in the data. If you run a consent banner, look for the (not set) and modelled rows rather than assuming the banner is working. A consent implementation that blocks everything also blocks the orders you wanted to measure.
- Cross-domain is configured if checkout is elsewhere. When the payment or checkout step sits on another domain, GA4 needs the linker configured or the session breaks and every conversion reports as a new session with no source.
The check people skip
Open Acquisition, then Traffic acquisition, and look at the share of sessions arriving as direct or unassigned. A large share on a store that always has campaign tags is not a traffic insight — it is usually a broken linker or a lost utm parameter, and it makes every channel look worse than it is.
One habit makes all six checks cheaper: add an internal traffic filter and a test-order property now, so your own orders and your own testing never contaminate the baseline you are about to record.
What to fix before you spend a penny
- A shared event id across browser and server, with the same event name on both sides.
- Purchase fired once, after confirmation, with the order's real value and currency.
- A server-side path that survives a closed tab — otherwise every blocked or abandoned browser session is silently missing.
- Consent defaults set before any tag loads, if you sell into markets that require it.
- UTM and click identifiers captured on landing, not read at checkout from a URL that no longer has them.
What to record before the campaign starts
| Metric | Where to read it | Why it matters later |
|---|---|---|
| Orders per day and revenue per day | Store back office | The only truth the ad platforms can be checked against |
| Purchase events per day, per platform | Each ad platform | The ratio to real orders is your measurement health, and it should be close to 1.0 |
| Sessions and conversion rate | GA4 or equivalent | Distinguishes a traffic problem from a tracking problem when the numbers move |
| Events per session | GA4 | A sudden fall with flat orders is a tracking regression, not a demand one |
Red flags that mean do not scale yet
- The platform reports more purchases than your store has orders.
- The reported purchase value differs from the order total by a consistent ratio — usually 100×.
- Purchase fires on the payment redirect, so abandoned payments are counted.
- There is no server-side event and you are buying traffic in markets with heavy tracking protection.
- GA4 and the ad platform disagree by more than about a fifth with no explanation.
Each of these makes the cost per purchase look better than it is. At a true £36 cost per purchase, a doubled purchase count shows £18 — and £18 is a number a merchant increases budget on.
If you cannot answer the three-number question about your last order, the measurement is a guess. We can run that test on your store in an hour and tell you what it says.
