A French brand ordering ten creator videos spends between 1,500 and 4,000 euros, and in most cases will never know which of the ten earned anything. This is not a tooling problem: the data already sits in the ads manager and in the analytics. It is a badly framed question.
The question "UGC ROI" does not answer
Asking for the return on investment of UGC is like asking for the return on investment of video. The answer depends on what you do with it, and the same ten videos give three different results depending on whether they run as paid ads, organically, or on a product page.
The useful question is always narrower: this video, in this placement, against what? Without the comparison term there is no measurement, only a number.
The benchmark that is almost always missing
A brand that has never run its usual creative alongside does not know whether UGC did better. It knows the campaign worked. The benchmark does not need to be sophisticated: same audience, same budget, same week, and the in-house creative facing it. Without that, all you measure is the season.
Three measurements, three horizons
| Measurement | What it answers | Time to read | Where to read it |
|---|---|---|---|
| Cost per acquisition | Does this video sell more cheaply than another | 7 to 14 days | Ads manager |
| 3-second retention | Does the hook hold | 48 hours | Creative statistics |
| Share of revenue | Does the channel weigh anything | 30 to 90 days | Analytics and back office |
The three do not replace one another. A video can have the best 3-second retention in the batch and the worst cost per acquisition: it attracts, it does not convince. Another can be expensive per click and come first on share of revenue, because it brings buyers rather than browsers.
Why 3-second retention is read first
Because it arrives within 48 hours and costs little to get. A hook that holds nobody makes the rest of the video pointless, and knowing that after two days rather than after three weeks changes what you order in the next batch.
It says nothing about selling, and that is its limit. A brand that optimises on this measurement alone ends up with videos that hook and do not convert, which is an expensive and well documented problem.
Share of revenue, the slowest and the most useful
It needs thirty to ninety days, which discourages most brands, and yet it is the only one of the three that answers the question a director asks: does this channel weigh anything. The other two compare creatives to each other, this one compares a channel to other channels.
It is read in the back office rather than in the platform, because the back office knows the orders actually collected, the returns and the refunds. A campaign showing an excellent cost per acquisition and a thirty per cent return rate has not done what the advertising dashboard claims.
What France makes harder to measure
Consent. Since the CNIL guidance on cookies, a significant share of French visitors refuses advertising trackers, and that share therefore does not appear in the platform's attribution. A campaign that seems to earn nothing may well be earning, without the pixel seeing it.
Two practical consequences. First, comparing a UGC campaign to a standard campaign remains valid, because both suffer the same loss. Second, comparing a platform figure to a back-office figure does not: they do not count the same population, and the gap between them is not an error, it is the measurement of what consent removes.
The brand that progresses fastest is the one that notes the gap once, turns it into a coefficient, and stops rediscovering it every campaign. The coefficient is recalculated twice a year, not per campaign: it moves with consent habits and with browsers, not with the creative you have just launched.
The minimum protocol, in five steps
- Pick a single question, for instance: does this hook beat ours.
- Run both creatives on the same audience, the same budget, the same week.
- Change nothing during the test, not the budget, not the targeting, not the landing page.
- Read retention at 48 hours, cost per acquisition at 14 days.
- Write the result somewhere the next brief will read it.
The fifth step is the one missing everywhere. A brand that tests without recording runs the same test six months later, with a different supplier, and believes it has learned something new. A shared three-column file is enough: the question asked, the result, the date. What costs money is not the measurement, it is redoing it because nobody wrote down what it gave.
The volume objection
"We do not have enough traffic to test." That is often true, and it is a reason to test fewer things, not to measure nothing. A low-volume brand drops the A/B test and keeps the two measurements that do not need significance: 3-second retention, readable on a few hundred views, and completion rate. They do not say what sells, they say what gets watched, and that is already more than nothing.
The other route for a low-volume brand is to give up simultaneous testing and compare periods instead. That is methodologically weaker, season and weather invite themselves in, but two consecutive months with the same offer and the same ad budget give an indication nobody should mistake for proof. The rule is to say so when presenting it internally, rather than letting a chart pass it off as a demonstration.
The attribution window nobody sets on purpose
Most platforms default to a seven-day click window, which suits impulse purchases and flatters short cycles. A brand selling a considered product, a piece of furniture, a course, an appliance, will see a campaign look weak simply because the buyer took eleven days to decide.
Setting the window to match the real decision time, and keeping it the same across every test, costs one click in the settings and removes an entire category of false conclusions. What matters is not which window is right in the abstract: it is that the same one is used every time, so that two campaigns are actually comparable.
What cannot be measured, and has to be owned
A creator video also produces effects that appear on no dashboard: the brand becomes familiar, a comment reassures a hesitant buyer, a piece of content is reused by customer service. None of these attaches to a line of revenue.
The honest reflex is neither to invent them nor to ignore them. It is to name them as unmeasured, decide in advance what share of the budget you accept spending there without proof, and measure the rest seriously. A brand that demands a numeric proof for everything ends up producing only what proves itself quickly, which is another way of investing badly.
Sources
- CNIL, cookies and other trackers
- CNIL, guidance on audience measurement tools
- Meta Business, about attribution
Checked on 6 September 2026. Rules and tools change: where this guide and the official source disagree, the source prevails.
