Every Monday, somewhere, a marketing lead opens three tabs. Meta says last week's return on ad spend was 4.1. Google Ads says 3.6. The store, or GA4, says the whole paid programme returned 2.2. Someone asks which one is right, and the honest answer is that each is right about a different question.
The gap is rarely a tracking bug. It comes from a handful of settings and conventions that are perfectly reasonable on their own and incompatible with each other. Once you can name them, the argument stops being about whose dashboard to trust and starts being about which question you are asking.
1. Every platform grades its own homework
Meta and Google each count conversions they can connect to their own ads. Neither knows, or cares, that the other also touched the same customer. A shopper who clicks a Google search ad on Tuesday and an Instagram ad on Thursday, then buys on Friday, can be counted as a full conversion by both. Add the two platform figures together and you have sold that order twice.
That is not dishonest. It is what a platform-level report is for: telling you whether ads on that platform are contributing. It just means platform figures are not additive, and their sum will almost always exceed what actually landed in the bank.
2. Attribution windows are different by default
An attribution window is how long after someone interacts with an ad a purchase still counts. The defaults differ:
- Meta defaults to 7-day click and 1-day view: a purchase within seven days of a click, or within a day of merely seeing an ad, is credited.
- Google Ads defaults to a 30-day click window for most conversion actions, and newer accounts use data-driven attribution to split credit between ad interactions.
- GA4 distributes credit across every channel it sees, paid and organic, so email, direct and organic search all take a share that the ad platforms never give up.
Longer windows find more conversions. Shorter windows find fewer. Neither is wrong, but comparing a 30-day number with a 7-day number tells you about the windows, not the ads.
3. View-through is doing a lot of quiet work
The “1-day view” part of Meta's default means someone who scrolled past an ad and bought the next day, perhaps after searching for your brand, is credited to Meta. For brands with heavy existing demand, view-through can be a large share of reported conversions. Some of it is real influence. Some of it is people who were going to buy anyway and happened to be on Instagram first.
You do not have to turn view-through off. You do have to know how much of the number it represents. Break the column out by attribution setting before you decide what the figure means.
4. The sale lands on different days
Platforms do not all book a conversion on the same date. Google Ads reports conversions against the day of the ad interaction, so last week's figure keeps growing for days as late purchases arrive. Analytics tools and your store report against the day the order was placed. Pull “last week” on a Monday morning and the two are measuring different slices of time, and the ad platform figure is still incomplete.
5. Nobody agreed what “revenue” means
The pixel fires with whatever value the checkout passes it. That might include tax and shipping. It probably does not subtract refunds, cancellations or the order that was paid for with a gift card. The finance team's revenue figure does all of those things. Two ROAS numbers built on different definitions of revenue will never meet, however well the tracking works.
6. Some conversions are modelled, not observed
When people decline cookies or opt out of tracking on their phones, platforms fill some of the gap with modelled conversions. That is a sensible approach, and the models are often good. But it means part of a platform's figure is an estimate that your store will never confirm order by order.
How to get to one number everyone accepts
You will not make the three figures agree, and you should stop trying. What you can do is give each one a job.
Pick a source of truth for revenue
Revenue comes from the store or the order system, net of refunds, in one currency. Nothing else is allowed to be called “revenue” in a board pack. Platform figures are “platform-reported conversion value”, and they are labelled that way.
Report blended ROAS to the business
Blended ROAS, sometimes called marketing efficiency ratio, is total revenue divided by total paid media spend for the same period. It ignores who gets the credit and answers the question the business actually asks: for every unit we spend on ads, how much comes back? It will be lower than every platform figure, and it is the one that survives a board meeting.
Use platform ROAS inside each platform
Platform figures are still the right tool for decisions within that platform: which campaign, which audience, which creative. They are consistent with themselves, which is what matters when comparing one Meta ad set with another. They are the wrong tool for moving budget between Meta and Google, because each is measured with its own ruler.
Write the definitions down
One page: which attribution settings each platform uses, whether view-through is included, how revenue is defined, which day a conversion lands on, and when the weekly figures are pulled. Agree it once. Most recurring arguments about ROAS are really arguments about an undocumented setting someone changed.
Test the big questions with holdouts
When the question is “does this channel make money at all?”, attribution cannot answer it. Switch a channel off in some regions, or hold out a share of an audience, and compare. It costs a little revenue for a few weeks and settles debates that dashboards never will.
How we handle it in FORTVISION: spend and platform-reported value from Meta, Google Ads and DV360 sit in the same table as revenue from your store, each clearly labelled with its source, so the blended figure and the platform figures are side by side instead of in three tabs.
The goal is not one perfect number. It is one agreed number for the business, a set of consistent numbers for the people optimising each platform, and nobody confusing the two.