Criteo review — the ratings are real, the question is which product they are rating

Three letters, and a company that has changed shape

Three letters first: on this page DSP means demand-side platform — the software that buys programmatic advertising — and never Amazon's Delivery Service Partner network, which franchises parcel-delivery vans. Both abbreviations attract reviews, and they answer completely different questions.

Criteo's own company page describes the business today as the Commerce Intelligence Platform, with 3,500+ employees, 17,000+ clients and more than 5 billion ads served a day, read on 20 August 2026. Its platform page still carries the older label, Commerce Media Platform. That is not something to score against anyone; it is a signal about how often this company renames what it sells, and that rate of change is what fractures a review corpus.

The advertiser-facing lines as they stand now are Commerce Growth for acquisition and retention, Commerce Max as the commerce media demand-side platform for enterprise brands, retailers and agencies, and Criteo GO for self-service. On the supply side sit Commerce Yield for retailers and marketplaces and Commerce Grid as a supply-side platform. A reviewer writing in 2021 used none of those words.

This page comes from Dr. DSP, Amazon DSP run as a managed product by Full Circle, a full-service Amazon management company with more than $500M in managed revenue across 100+ brands. We overlap with Criteo on one narrow axis and not on most of the others.

The corpus has two authors who have never met

Criteo runs a demand side and a supply side, and both write reviews under one logo. A performance marketer at a footwear brand rating Commerce Growth is grading a media buy — reach, cost per acquisition, creative control. A monetisation lead at a grocery chain rating Commerce Yield is grading a revenue engine — fill rate, demand connectivity, how much money the ad business returned to the retailer. Those two reviewers share a vendor and nothing else.

Directories model this unevenly, which is where the arithmetic goes wrong. G2 separates the advertiser products into their own listings, and on 20 August 2026 its Commerce Max (Retail Media) entry carried 18 reviews against 241 on the Commerce Growth entry. Capterra's principal Criteo listing held 22 reviews spanning April 2018 to April 2025 and reads throughout as a retargeting product; its separate Commerce Max listing had none at all. SourceForge shows the same pattern — a Commerce Growth page with no reviews, sitting beside a general Criteo page with a handful of ratings.

None of this is a fault of Criteo's, which does not decide how third-party directories model its catalogue. It is a warning about averaging. Blend those listings together and you have combined a decade-old retargeting product, a current retail media DSP with a thin sample, and a monetisation platform that no brand-side buyer will ever log into. The mean of those numbers describes nothing you can actually purchase.

A method for dating and sorting a review corpus

This works on any vendor whose product has been renamed, refactored or absorbed, which in retail media is nearly all of them. In this order it takes about fifteen minutes.

• Read the URL, not the page heading. Directories rename listings and keep the review history underneath. G2's Criteo Commerce Growth pages resolve from a criteo-dynamic-retargeting path, which tells you the reviews on that listing were written about the retargeting product that came before it.
• Sort by date, then cut everything older than the product name you are buying. A 2019 review of a line launched in 2022 is evidence about a predecessor. It may still be true about the company; it is not quotable about the thing in your proposal.
• Filter by firmographic before you read a word. Small-business reviewers dominate several of these listings. If you are placing seven figures a year, their experience of onboarding, minimums and support is not the one you will get.
• Separate employee reviews from customer reviews. A Glassdoor page carrying 1,441 employee reviews ranks for this exact query. It is a genuinely interesting document about a company and it is not a product review.
• Establish whether the reviewer was buying or selling. Retailer-side and brand-side reviewers of one vendor are grading opposite outcomes, and a five-star rating from one tells you nothing about the other.
• Weight anything that names a number. One review reporting a cost per acquisition, a fill rate or a reconciliation gap outranks fifty saying the interface is intuitive.

What survives that filter is usually a much smaller pile — often a dozen reviews rather than four hundred. Read those twice.

What Criteo is genuinely good at, in specifics

The strengths here are structural rather than promotional. Criteo's advertiser page, read on 20 August 2026, states access to 200+ retailers and 1,300+ direct publisher partners. No Amazon specialist can assemble that, and if your revenue is spread across grocery, mass, pet, DIY and your own site, holding all of it in one relationship has real operational value: fewer contracts, one creative pipeline, one reporting surface.

Two further things deserve explicit credit. The first is catalogue-driven advertising — Criteo built its name on dynamic product ads generated from a live feed, and reviewers from every listing and every era of the corpus agree that it does that part well. The second is measurement ambition. The same page advertises closed-loop measurement from ad exposure to point of purchase across retailer partners, SKU-level retailer-transparent sales data, brand lift studies, A/B testing and incrementality testing. A platform that puts incrementality testing on its own product page is not ducking the hard question.

Who should buy Criteo rather than us: a multi-retailer brand needing onsite and offsite reach across many banners at once; a retailer building or operating its own ad business; an advertiser in a market where the retailer network is the dominant surface and Amazon is not; and any team that wants a self-service route and employs traders who will actually use it week to week.

Reference-call questions that reveal how a media platform really runs

Ask for two references and take both calls yourself. Skip support responsiveness — every reference answers that identically. Ask these, and listen for whether the answer arrives with a number attached.

• Who did the trading? A named person at the vendor, an algorithm, or your own team. The answer decides what the reference is even evidence of.
• Show me your delivery reporting. Not performance — delivery. Impressions by placement, by retailer, by day. A reference who has never seen delivery broken out has been given summaries, not records.
• What independent measurement did you run, and what did it say? A holdout, a geo test, a media mix model — anything computed outside the platform that is reporting on itself.
• Were the platform's reported sales ever reconciled against your own books? Then ask by how much they diverged. Divergence is normal and expected. Not knowing its size is the finding.
• What share of your committed budget reached inventory? Most references will not know, and that is worth recording too. Take-rate visibility is a contract question, but whether customers ever see it is a culture question.
• What happened the last time you asked to change something mid-flight? Turnaround on a creative swap or a budget reallocation describes the operating model better than any case study will.
• What would have to be true for you to leave? The most revealing question on the list, and the one people answer most freely.

Where we fit, and when you should ignore this section

Dr. DSP does one thing: Amazon DSP, bought and operated as a managed product. There is no retailer network here and no monetisation platform. If Amazon is a minority of your revenue, we are the wrong purchase and the breadth described three sections up is the right one.

Where we do have an argument is proof. Last-click reporting cannot establish whether display added a sale or simply stood near one, and display is where that gap runs widest — every platform in this category, ours included, is grading its own homework. Our answer is to withhold display from a matched set of ASINs or regions for a defined window and reconcile in Amazon Marketing Cloud, the clean room where DSP and sponsored ads stop claiming the same orders. Amazon makes it available free to eligible advertisers.

A reference point, not a promise: across 30 advertisers in July 2026, a live API pull showed 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM, $1.42 blended cost per click, and $5.49 cost per acquisition across 57,137 attributed purchases, with 20.1% new-to-brand. Dr. DSP carries no published price — a demo, the first 30 days free, then priced on the call against real budget and scope, month to month, with Orbit included. Several vendors on your shortlist are more open about price than we are, and it is only fair to say so.

Two conditional redirects. If your Amazon problem is sponsored search rather than display, Dr. PPC is the right door and it publishes both its rate and its cap. If you want this bought as an agency engagement with a named team, across more channels than Amazon alone, that is reMKTR.