Free report · 2026 edition

Every ad platform grades its own homework. Here are the grades.

Last quarter we compared what each ad platform reported with what it earned, across Roivenue clients. Meta claimed twice its share, TikTok more than three times. In half of the clients, the platforms together claimed more revenue than the business booked.

Grade your own setup →

What we found

Four numbers from our client base, and where each one comes from.

2.1×

what Meta reported against the revenue Roivenue credits it, the median across our clients. Google Ads: 1.8×.

Client base, Q3 2026
3.4×

for TikTok, the highest of the large platforms. It claimed more than it earned for every client that runs it.

Client base, Q3 2026
1 in 2

clients whose ad platforms, added up, claimed more revenue than the business booked, organic and direct included.

Client base, Q3 2026
2 in 3

conversions take two or more touchpoints. Each platform grades only its own.

Client base, September 2026

1. The grades, and how each one is earned

Five students, and each one marks its own paper.

MetaA+
Counts
A purchase within 7 days of a click on a Meta ad, or 1 day after someone saw one.
Why it flatters itself
A video scrolled past the evening before a brand search earns the whole order. Gaps left by iOS consent are filled with modelled conversions.
What it cannot see
The Google search, the comparison site and every other platform’s ads.
Google AdsA
Counts
A conversion within 30 days of a click on a Google ad, plus engaged views on YouTube.
Why it flatters itself
Brand search and Performance Max collect buyers who were already looking for you, and bill for it.
What it cannot see
Anything seen on Meta or TikTok. Anything outside Google.
TikTokA+
Counts
A purchase within 7 days of a click, or 1 day after a view.
Why it flatters itself
The same view-through logic as Meta. A view shortly before the sale takes all of it.
What it cannot see
Every touchpoint that happened somewhere else.
Affiliate networkA+
Counts
The last affiliate click before the order, inside a cookie window that often runs 30 days.
Why it flatters itself
Voucher and cashback sites step in at checkout, when the buyer was already paying.
What it cannot see
Whatever brought the buyer to the checkout in the first place.
GA4B
Counts
Sessions and clicks on your site, credited by its own data-driven model or by last click.
Why it flatters itself
Nothing deliberate. It sells no Meta or TikTok media, but it gives the credit to whatever brought the last click: brand search, direct, Google.
What it cannot see
Every ad that was seen and not clicked, on Meta, TikTok and YouTube.

Defaults as each platform documents them in September 2026. Nobody is lying: each one applies its own rules to its own data, and the rules are built to find its own contribution.

2. Add up the report cards

Every platform is internally consistent. The trouble starts when someone adds them up for the budget meeting.

Each report adds up. Together they do not.

In our worked month, Meta claims €1.24M, Google Ads €1.46M, TikTok €0.38M and the affiliate network €0.27M. That is €3.35M for a month that booked €2.10M. The extra €1.25M is the same revenue claimed twice, three times or four times, because every platform that touched an order can claim all of it. Across our clients the size varies: in the middle half, the platforms together claim between 0.6 and 1.6 times the revenue booked.

  • Overlapping windows a click in one and a view in another both count.
  • Brand search collects demand that other ads created.
  • Nobody referees no platform sees the others’ claims.

The gap is different on every channel

Across our clients, Google Ads reports 1.8 times what it earns when every order is counted once, Meta and affiliate networks 2.1 times, and TikTok 3.4 times. A single discount factor on the platform numbers corrects none of them.

  • Client base Roivenue clients, July to September 2026.
  • Your account the self-audit measures your overlap in a minute.
Run the self-audit →

3. One order, five verdicts

Self-grading does not only inflate. It also starves the ads that start journeys, which is why the grades are wrong in both directions.

Over-counted and under-counted at once

Order #48211 was worth €240. Google Ads claims all of it for the brand search click on day 29, and the comparison site claims all of it for a click on day 22. Meta’s video on day 1 and TikTok’s on day 15 fall outside their one-day view windows, so their own reports give them nothing. Credited once, the Meta video earns €56.

  • Inflated €480 claimed for a €240 order.
  • Starved the two ads that started the journey get €0 in their own reports.

What a referee does with the same order

Every touchpoint shares the €240, the views included, and the channels add up to exactly what the customer paid. That is the number you can put in front of Finance.

  • Once every order counted once, never twice.
  • Views count the ads seen and not clicked earn their share.

4. What nobody grades

Some touchpoints never reach any report card. The channels they belong to lose budget without anyone deciding it.

Ads people viewed and never clicked

Meta, TikTok and YouTube do not export impression-level data. For one client’s month, Meta’s own report gave it 7.1× the credit GA4 did: 708 against 100. With Synthetic Impressions in the model it came out at 286. Across the clients where Roivenue models these views, counting them adds a viewed ad to about one conversion in ten, and moves 6 more points of credit away from the last click.

  • GA4 sees the click, never the view.
  • Meta sees its own views, and nothing else.
How many sales did Meta drive? Three answers for one month.
GA4can't see the ads people viewed
0
Meta's own reportclaims every sale its ads touched
7.1×the credit GA4 gives Meta
0

One sees too little. The other claims too much.

The independent read
Roivenuecounts views, splits each sale
0

Roivenue models in the Meta views GA4 misses, then splits each sale across the channels that earned it.

The phone and the laptop

Someone researches on a phone on the way home and buys on a laptop that evening. To a click-based tool, that is two strangers, and the research gets nothing.

  • Customer ID joins the two when the buyer logs in or orders.
  • Path Stitching scores the rest, without third-party cookies.

The visitors who said no

Consent refusals remove visits from every report. A report that guesses them back in grades itself again.

  • Counted visits with full tracking.
  • Measured visits in anonymous mode, without identifying anyone.

5. Five questions that make a platform rep go quiet

Ask them at the next quarterly review. Write down the answers.

  1. Of the conversions you report for us, how many also appear in Google Ads’ report?

    Why it worksNo platform can see what the others claim, so none of them deduplicates. The honest answer is “we cannot know”.

    A good answer“We cannot see that. Compare the total with your booked orders.”

  2. How many of these are view-through, and what window are you using?

    Why it worksA one-day view window hands the order to an ad scrolled past just before a search the buyer was going to make anyway.

    A good answer“A split of click and view conversions, and the exact setting.”

  3. What share of this number is modelled rather than observed?

    Why it worksSince iOS 14.5, part of every Meta and TikTok count is a statistical estimate, not an event your pixel fired.

    A good answer“A percentage, and how the model is checked.”

  4. If we pause this campaign in two regions for four weeks, what drop do you expect? Will you put it in writing?

    Why it worksIt turns a claim into a prediction you can test. A holdout measures incrementality directly.

    A good answer“Yes, with a test design and the number to beat.”

  5. Does this reconcile with our booked revenue after returns and cancellations?

    Why it worksPlatforms count the purchase event. Finance counts the money you keep.

    A good answer“No, and here is how to adjust for it.”

6. Grade your own setup

Ten statements, one honest grade, and the three fixes that pay back first.

1When we add up the platform reports, we know how far the total is from booked revenue.
2Our attribution counts ad views on Meta, TikTok or YouTube, not only clicks.
3We can join a phone visit and a laptop purchase from the same person.
4We know our consent rate, and what happens to the visitors who decline.
5The tool that credits our channels does not grade its own ads.
6We can explain to our CFO in two sentences how credit is assigned.
7Revenue in our reports is net of returns and cancellations.
8We have run a holdout or geo test in the last twelve months.
9Our lookback window is at least as long as a typical buying journey.
10The budget meeting uses the same number as the performance report.

7. How we graded

Where every number in this report comes from, including the ones that are examples.

Platform defaults
Windows and counting rules as each platform documents them in September 2026. Your account settings may differ; the report cards show the defaults most accounts run on.
The worked month
One composite retail month, €2.10M booked, and order #48211 are the example used across this site. They show the mechanics. The gap in your own account will differ: the self-audit measures it in a minute.
Meta’s index
One Roivenue client, one month, with GA4 set at 100. From the figures in our guide to multi-touch attribution vs last-click, rebased.
Client base
Roivenue client accounts with data for July to September 2026, test and demo accounts excluded. Platform revenue is what each platform reports; Roivenue credit uses each account’s AI Data Driven model; every figure is the median across clients. A platform counts only where it reports a revenue value. Roivenue counts the orders it can track, so part of each gap is visitors who declined consent and whom the platforms model in. Comparing platforms inside the same account removes that part.
Journey lengths
Converting journeys across Roivenue clients in September 2026.
Orange
The published case study with Digiline: return per placement from 0.9 to 14.0.
Our own grade
We build an attribution tool, so grade us too. Roivenue does not see offline media, and attribution describes journeys: it does not prove what would have happened without an ad. That takes experiments, and the total budget question takes marketing mix modelling, which ScanmarQED, our parent company, builds.

Keep reading

See the same month graded once.

45 minutes with a consultant, on relevant demo data: every order counted once and split across clicks and ad views, and a one-page summary for your CFO.

See what your attribution misses →See pricing
See what you missPricing