A homeware brand in Leeds used to lose the first twenty minutes of every monthly board meeting to the same argument. The agency slide said 180 conversions from Google Ads. The analytics lead's GA4 export said 94. The finance spreadsheet, pulled straight from the store, said 71 orders could be traced to paid search. Three people, three numbers, and a growing suspicion that someone was inflating results.
Nobody was lying. I run into this on most multi-channel accounts I take over, and the resolution is almost never a bug. The three systems are answering three different questions, and the business had been treating the answers as if they were the same question.
What each one is actually counting
Google Ads counts conversions it believes its own clicks caused, within an attribution window you set, and it credits them back to the date of the click rather than the date of the sale. So an order placed on the 3rd of November from an ad clicked on the 28th of October shows up in October. It also counts every conversion action you have marked as primary, so if both "purchase" and "begin checkout" are primary, one customer can be two conversions.
GA4 counts events in sessions, attributed by its own model, usually data-driven or last non-direct click. It does not credit the click date; it credits when the event happened. It also loses anyone who declined consent, blocked the tag, or arrived on one device and bought on another without being identified.
Your store counts orders. That is the only one of the three that corresponds to money arriving in a bank account, and it has no opinion at all about which channel deserves credit.
Once you say it in that order, 180 against 94 against 71 stops being a scandal and starts being arithmetic.
The four structural reasons the gap exists
Attribution window and model. A 30-day click window will always report more than GA4's default, and view-through conversions in Meta or Display will widen the gap further. Neither is wrong. They are different definitions of "caused".
Conversion date versus order date. This alone can swing a month by double digits at the start and end of a period, especially for products with a long consideration cycle. If you compare a platform's conversion-date report against a finance order-date report, the months will never reconcile cleanly and you will waste hours trying.
Double counting inside one platform. Multiple primary conversion actions, a thank-you page that can be refreshed, and both a tag-based and a server-side event firing without deduplication will each inflate the number. The last one is the most common and the easiest to miss, because the setup looks more sophisticated than the broken one.
Consent, blockers and cross-device. In the EU and UK, a meaningful share of visitors decline analytics cookies. Those people still buy; they just do not appear in GA4 in the normal way. Add iOS privacy behaviour and people who research on a phone and buy on a laptop, and client-side tracking structurally undercounts.
Which number to use for what
| Decision | Use this | Why |
|---|---|---|
| Should I raise this campaign's budget? | Platform conversions | It is the only signal the bidding algorithm can act on |
| Which creative or keyword wins? | Platform, same source for both | You need consistent relative comparison, not absolute truth |
| Is marketing profitable this month? | Store revenue and total ad spend | Only orders are real money |
| Is the channel mix working? | Blended MER, from store data | Immune to every attribution argument above |
| Did a change actually work? | A holdout or geo test | Attribution cannot prove causation, only correlation |
That table ended the Leeds argument more effectively than any tracking fix. The agency number was not a lie, it was an in-platform operating metric. The finance number was not pessimistic, it was the business result. They had simply been asking one number to do two incompatible jobs.
The reconciliation I run on a new account
Start from the store, always. Export orders for 90 days with the order date and whatever channel attribution your store records. That is your denominator and your sanity check.
Then calculate blended marketing efficiency ratio: total revenue divided by total marketing spend across every channel, month by month. No attribution model is involved, which is exactly the point. If MER improves while spend rises, the account is working. If MER sags while platform ROAS looks healthy, you are buying customers you already had.
Then, and only then, open the platforms, and look at them as instruments rather than scoreboards. Within Google Ads, compare one campaign to another. Within Meta, compare one creative to another. Do not add Google's revenue to Meta's revenue and expect the total to resemble your bank statement. It never will, because both of them claimed the same customer.
The fixes worth making
Set exactly one primary conversion action per platform, the one that corresponds to revenue. Everything else becomes a secondary action you can still see but which does not drive bidding. This single change fixes more reporting disputes than any other.
Deploy server-side tracking properly: the Conversions API for Meta, enhanced conversions for Google, with deduplication keys so a browser event and a server event about the same order are recognised as one. If you are running both without deduplication right now, your numbers are inflated and your bidding is learning from noise.
Implement consent mode correctly if you sell into the UK or EU, so that modelled conversions fill part of the gap rather than the data simply disappearing.
Agree on one reporting calendar, by order date, and put the platform numbers in a clearly separate section of the report labelled as in-platform metrics. The labelling matters more than it sounds. Half of these arguments are about trust, not arithmetic, and a report that explains its own definitions is a report people stop fighting about.
What good looks like afterwards
The Leeds brand did not end up with three numbers that match. They never will, and neither will you. What changed is that the monthly review now opens with store revenue and blended MER, then moves to platform metrics for the operating decisions, and the gap between them is a known, explained quantity rather than a monthly accusation.
That is the realistic goal. Not perfect attribution, which does not exist, but a clear agreement about which number answers which question.
Three articles pick up where this one stops. Why a healthy ROAS can still leave no profit is the margin half of the same problem. Performance Max claiming your brand searches is the most common single source of inflated platform revenue. And cheap leads that never close shows what happens when the conversion action itself is measuring the wrong thing.
If your platforms and your store are telling you different stories and you want someone to reconcile them properly, the free growth audit includes a full tracking and attribution review: what is firing, what is double counted, what is missing, and which number you should actually be managing to.