Why Google, Meta and Shopify will never agree

You open Google Ads and it says last week brought 180 conversions. You open Meta and it claims 140. You open Shopify and it counts 240 orders. Three dashboards, three numbers, the same week.

The natural reaction is to assume somebody is lying. They are not, and it helps to understand why: as long as you hunt for the "right" number you waste time, when the real issue is that they are answering three different questions.

Every platform answers its own question

Google Ads answers: how many sales came after somebody saw or clicked one of my ads, by my rules. Meta does the same, by its rules. Shopify answers something simpler and more honest: how many orders did I receive.

The rules change everything. How far back you look after a click, whether a view without a click counts, how much credit to give when there were several touches before the purchase. There is no objective answer to those questions: they are conventions, and each platform picks its own.

Double counting is not a bug

Somebody sees a post on Instagram on Monday, searches your name on Google on Thursday, clicks the ad and buys. Meta claims that sale, Google claims it too. Neither is lying: both contributed, and each counts by its own rules.

The result is that the sum of platform-reported conversions almost always exceeds your real order count. Add those numbers up to see how you are doing and you are telling yourself a nicer story than the one you are living.

The number that does not argue

There is one figure nobody can game, because it depends on no attribution at all: total store revenue against total ad spend. It does not say which channel worked, and that is fine: it says whether everything you are doing together produces more than it costs.

That is the number for deciding whether to spend more or less overall. Platform numbers are for something else: deciding how to split that spend, knowing they are estimates made with partisan rules.

The worst trap: the blended figure

There is an error more insidious than double counting, because it looks like rigour. It is reading the combined figure across all platforms.

On one store we worked on, blended cost per click showed a 18.8% fall and click-through rate a 57.9% rise. Excellent numbers. Isolating Google Ads from the total, the real improvement was 3.6% and 29.4% respectively.

The difference was not a calculation error: it was Meta. In the recent window it weighed far more in the total, and Meta by nature has a very low cost per click and a high click-through rate. The blended figure was not measuring an improvement, it was measuring a budget shift between platforms.

It is the kind of error that is never visible, because the blended number looks exactly like a good result is supposed to look. The only way to catch it is to recalculate per platform, every time.

How to read a comparison without fooling yourself

Three rules, and they hold for any report.

Compare the same months a year earlier, not the previous quarter. Against the previous quarter, seasonality does the talking rather than the work you did.

Calculate per platform, never on the total, for the reason above.

Check the months exist. If a connector was linked halfway through the period, the earlier months are not zero: they are unobserved. Treating an unobserved period as a zero produces enormous, meaningless percentages.

Frequently asked questions

Which number should I use to set the budget?

For total budget, the ratio of total revenue to total spend: it depends on no attribution and nobody can inflate it. For splitting between channels, the platform numbers, knowing they are partisan estimates and treating them as clues rather than truth.

Does an attribution tool solve this?

It reduces it, it does not remove it. An external tool applies its own model, which is a fourth convention alongside the other three: more consistent, because at least there is only one, but still a choice rather than a measurement. The real benefit is having one rule instead of three.

Why does Shopify see fewer conversions than the platforms?

Shopify does not "see fewer": it counts orders, which are a fact. It is the platforms that count more, because each claims the sales it believes it contributed to, and several platforms claim the same sale.

How long does a period need to be to be reliable?

Long enough to contain a decent number of orders, and aligned to the same period a year earlier. At low volumes a week tells you nothing: random variation is larger than the effect you are looking for.

Is platform-reported ROAS useless?

No, it is useful for comparing campaigns inside the same platform, where the convention is the same for all of them. It becomes misleading when compared across platforms, or used to decide whether the business is making money.