10 min read
In brief
- Meta and Shopify will never match exactly, even when nothing is broken. Attribution windows, view-through and refunds move the number in the same direction as real tracking loss, which is why the first job is separating the two.
- One ratio routes the whole diagnosis: reported purchases divided by real orders. Consistently below 0.9 is loss; above 1.0 is duplication. The fixes are opposite, so guessing costs a month.
- The Conversions API closes the loss side, not the duplication side. Installing it on a setup with no shared event id makes the over-reporting worse, not better.
The gap runs in two directions
Almost everyone arrives at this question with the same assumption: Meta is under-reporting. Sometimes that is true. Sometimes Meta is reporting more purchases than the store has orders, which is a worse problem and much easier to miss, because a higher number looks like good news.
The two directions have opposite causes and opposite fixes. Loss means the browser could not send what happened. Duplication means the browser sent it and the server sent it too, without a shared id, so both were counted.
So the first job is not fixing anything — it is establishing which direction you are in.
One number, one direction
Take the last full week: purchases reported by Meta, divided by orders in Shopify (excluding cancellations). Below 0.9 is loss. Between 0.9 and 1.05 is probably normal measurement noise. Above 1.05 is duplication. Write the number down before you change anything, because it is the only baseline you will get.
Why they will never match exactly, even when nothing is wrong
- Attribution windows. Meta credits a purchase to an ad interaction within its window — seven-day click and one-day view by default. Shopify credits the order to whatever referred the session. Different questions, different answers.
- View-through. A purchase from someone who saw an ad and returned later by direct visit belongs to Shopify's direct channel and to Meta's view-through attribution. Both are correct in their own frame.
- Timing. A purchase on the 31st can land in Meta's reporting on the 1st, depending on the timezone each system uses. Daily comparisons always show noise; weekly comparisons mostly do not.
- Refunds and cancellations. Shopify counts an order the moment it is placed. Meta counts a purchase the moment it is reported. If you cancel the order afterwards, one system keeps the sale and the other keeps the conversion.
Those four explain a difference of a few per cent in either direction. They do not explain a 40% gap, and they do not explain a conversion count that is double your order count.
The five causes, ranked by how often they are the answer
| Cause | Direction | One-line test |
|---|---|---|
| Browser event never sent — blocker, closed tab, in-app browser, restricted storage | Loss | Compare purchases reported on desktop vs mobile; a heavy mobile share with a low count points here |
| No server-side path at all | Loss | If the only purchase source in Events Manager is the browser, every blocked session is simply missing |
| Event name or id mismatch between browser and server | Duplication | Both copies present in Events Manager as separate events rather than one |
| Two base codes or two purchase codes (theme plus an app) | Duplication | The same purchase appears twice with the same id, or two PageViews per navigation |
| Value or currency sent incorrectly | Neither — but worse | Meta's purchase value divided by your order total is a constant ratio, usually 100 |
The last row deserves its own sentence. A value sent in minor units — 8450 instead of 84.50 — does not create a count difference, so it survives every check that compares counts. It makes every cost metric wrong by a factor of one hundred and it is the most common single defect we find on stores that believe their tracking is fine.
What the Conversions API actually is
The browser pixel runs in the customer's page and reports what happened there. The Conversions API is a second path: your server sends the same event directly to Meta, from the order record, over an authenticated request.
The difference is what each one can see. The browser sees the click that arrived with an fbclid, the cookies written on landing, and the pages visited before the product. It cannot see an order that completes after the tab closes, and it cannot send anything from a device that is blocking it.
The server sees the order that actually exists: the id, the amount captured, the currency, the timestamp. It cannot see the click unless your landing page stored the identifiers, which is why a server-only setup is worse than a browser-only one for attribution.
Together they cover both halves — the browser has the click, the server has the order — provided the two copies of each event are joined so Meta counts one.
What it does not do
The Conversions API does not fix duplication; it creates the risk of it. It does not improve ROAS by itself, and it does not recover events that were never sent from the browser if your server event is fired from the same incomplete data. It closes one hole and opens one, and the event id is what keeps the second one shut.
Do you need server-side tracking? A decision, not a sale
| Your situation | Verdict |
|---|---|
| Spending under a few thousand a month, single market, desktop-heavy audience | Not yet. Fix the browser path and the value first — that is where the error is |
| Spending meaningfully, or scaling this quarter | Yes. At that point a 15% under-count is a budget decision made on wrong numbers |
| Selling into the UK or EU with consent requirements | Yes, and do the consent defaults at the same time, not later |
| Already duplicating purchases | Fix the join first. Adding a server path to a broken join doubles the duplicate |
| Running shoppable ads in apps with restricted browsers | Yes — the in-app browser is where the browser path fails most often |
The honest version: server-side tracking is not a growth lever, it is a correction. It does not make a bad offer work; it stops you from making decisions on numbers that are wrong in one direction.
Why this comes up more on Shopify
Three Shopify-specific things make the browser path weaker than merchants expect, and none of them are visible from the admin.
First, checkout and the thank-you page may live on a Shopify-controlled domain rather than yours, depending on your plan and setup. The pixel can still fire there, but the cookies and the click identifiers are being read in a context you did not build, and any app that hooks the same events can fire alongside yours.
Second, pixels arrive from more than one place: the theme, the Shopify customer-events layer, and any app that sends conversions. Two of those firing a purchase on the same order is common, and it looks exactly like a correct setup until you compare counts.
Third, product and cart events are frequently templated by apps rather than authored, so the value sent can be the product price rather than the cart total — which is right for one item and wrong for every basket with two.
How to check whether your pixel is actually firing
There are four levels of checking, and each one answers a different question. Most merchants stop at the first, which is the one that proves the least.
- 1
Events Manager: is the pixel receiving anything?
Test Events shows events arriving from your browser in real time as you click through the store. This proves a connection exists. It does not prove the event carries the right name, value or id.
- 2
The browser network tab: is the request actually sent?
Filter the network panel for the pixel's request and watch it fire on page load and on add-to-cart. A pixel that appears installed in an app but sends nothing from the page is a theme conflict, not a Meta problem.
- 3
The event payload: does it carry what you think?
Open the request and read the parameters: event name, value, currency. This is where a purchase arrives as PageView, or a value arrives in minor units, and both look perfectly healthy in every dashboard above.
- 4
Your own order table: does one order equal one purchase?
Place a real order and compare. One order should produce one purchase with the same value. This is the only check that catches duplication, because duplication looks like success everywhere else.
Firing and working are different claims
A pixel that fires is not a pixel that works. The three failures that cost the most money — a duplicated purchase, a value in the wrong units, and an event fired before payment completes — all pass the first two checks and fail the last two.
The thirty-minute diagnosis
- 1
Calculate the ratio for the last full week
Reported purchases divided by Shopify orders, cancellations excluded. Below 0.9, above 1.05, or inside — each leads somewhere different.
- 2
Read the purchase value against the order total
If Meta's average purchase value is a suspicious round multiple of your average order value, you have a units bug and you can stop diagnosing and start fixing.
- 3
Open Events Manager and count sources
Look at whether the same order appears once, twice, or once per source. Two sources are fine only when they are joined; two events are not.
- 4
Place one real order and trace it
Note the id, then find the browser request and the server event. Same id, same name, same value, same currency — or you have found your answer.
- 5
Only then decide what to build
Loss with a clean value usually means a server path or a consent fix. Duplication means the join. A units bug means neither — it means a number is wrong.
What to fix, in what order
- Value and currency first. A number that is wrong by 100× invalidates every other comparison you are about to make.
- Then the join: one id, one event name, both paths.
- Then the loss: a server event fired from the order record, so a closed tab no longer removes the sale.
- Then consent defaults, if your markets require them, before any tag loads.
- Then attribution plumbing — UTM capture, click ids on landing — because this is what makes the report worth reading afterwards.
One rule for the whole job
Change one thing, then re-measure the ratio. A tracking fix you cannot observe in the ratio is a change you cannot prove happened, and you will be back here in a month.
If you want the number before you decide anything, the audit is one week of your order data and one traced order. We will tell you which direction you are in and what it will take to close it.
