Most of the Criteo accounts we inherit look profitable on the surface and fall apart the moment anyone asks a second question. The platform reports a return figure, the finance team compares it to the number in the ecommerce backend, and the two disagree by a factor that nobody can explain. That gap is the problem. Criteo Ads ROI is not hard to calculate; it is hard to calculate in a way that two different departments will both accept. Retargeting sits closest to the purchase of any channel you run, which means it inherits credit for buyers who were already returning, and it means any measurement you build has to answer the incrementality question before it answers anything else.
We run Criteo for ecommerce clients with a real catalogue, real repeat customers and a real cost of goods, and the work splits in half. The first half is operational: feed quality, audience windows, exclusion rules, creative formats, bid governance. The second is measurement: deciding which conversions belong to the channel, building a report that reconciles with the backend, and running periodic holdouts so the answer is not opinion. Skip the second and you spend a year scaling something that would have happened anyway. Skip the first and you measure a badly built account very precisely.
Why the platform number and the backend number never match
Three things create the gap. Attribution windows differ: the platform counts a conversion inside its own lookback, your analytics counts it against whatever touched the session last. Models differ: view-through credit is standard in retargeting and absent from a last-click backend report. And deduplication differs: if the same order is counted by Criteo, paid search and email, channel revenue will exceed total revenue. None of these are faults. They are different questions with different answers, and the mistake is treating one as the truth.
Our practice is to publish both numbers side by side and label them. The platform figure is the optimisation signal, because that is what the bidding system responds to. The backend figure, reconciled against orders and net of returns, is the business figure. When someone asks about Criteo Ads ROI, the honest answer is a range with a stated method, not a single decimal.
Feed quality decides more than bidding does
Retargeting performance is mostly a function of what the algorithm has to work with. A feed carrying accurate stock status, clean category structure, correct pricing and images that render at small sizes will outperform a neglected feed however carefully bids are managed. We have seen accounts recover most of their lost efficiency from feed work alone.
The checks we run before touching campaign settings:
- Stock and availability sync frequency, because serving out of stock products burns budget and trains users to bounce
- Price accuracy against the live product page, including currency and tax treatment for Turkish and export storefronts
- Category and product type fields populated consistently, so audience segmentation is possible at all
- Image quality at the smallest rendered size, where most placements actually appear
- Exclusion of products with negative margin or chronic return rates, which the platform cannot know about
That last item is where margin enters the conversation. A platform optimising for revenue will happily scale your worst products. Excluding the items you do not want more of changes the economics more reliably than any bid adjustment.
Running a holdout without breaking the account
The only clean way to establish incremental value is to withhold ads from part of the audience and compare purchase rates. This is uncomfortable, because it means deliberately not advertising to people who might buy. We size the holdout so it finishes in a defined period, agree the success metric before launch, and freeze other changes for the duration. A test contaminated by a creative refresh tells you nothing.
What comes back is that some portion of retargeted revenue was incremental and some was not, and the ratio differs sharply by audience. Cart abandoners within a short window are frequently self-closing; browsers who never reached a product page are the opposite. Once you know that split, the audience strategy writes itself.
What a usable Criteo Ads ROI report contains
A report that survives a finance review has four layers. Spend and platform-reported return, unchanged, as the optimisation baseline. Backend revenue for the same period, net of cancellations and returns. Gross margin at product or category level, so the result is contribution rather than turnover. And the incrementality factor from the most recent holdout, stated with its date, because it ages.
We also record what changed and when. Criteo Ads ROI moves for reasons that have nothing to do with the channel: a site speed regression, a checkout change, a stock outage on a hero product, a competitor running a promotion. Without a change log, every fluctuation becomes an argument. With one, most fluctuations explain themselves in under a minute.
Where we would stop spending
Retargeting has a ceiling set by the size of your qualified audience. If traffic is not growing, more budget in the same pool produces more frequency and less value per impression. When we see frequency climbing while contribution flattens, the recommendation is to cap the channel and move the difference upstream into whatever is generating new visitors. That is an unpopular recommendation from an agency paid to manage the channel, and it is still the right one. Criteo Ads ROI is a measurement of efficiency, not an argument for scale, and the accounts that stay healthy are the ones where somebody was willing to say the budget had reached its useful limit.
Keep reading: Criteo Ads · Werbung