Criteo connects commerce AI to shopper acquisition and retail media

Criteo applies AI to a specific commercial loop: understanding shopping signals, selecting relevant products and deciding where an advertisement should appear. Its buyer-side tools help advertisers acquire and retain customers, while Commerce Yield helps retailers sell advertising on and around their own commerce properties. These are different business decisions under one company.

01 / ProductBuying ads and operating a retail-media business are separate jobs

Commerce Growth supports acquisition, retention and retargeting using commerce data and campaign automation. The retailer buying ads wants sales or customers at an acceptable cost. That use case differs from selling sponsored placements to brands whose products already appear in the retailer’s own store. Criteo’s breadth makes it important to identify which side of the advertising transaction the team is evaluating.

Commerce Yield addresses the latter job through inventory and data monetization. Its sponsored-product offer places native ads in commerce environments, including search, category and product pages. The retailer is considering an additional revenue business, with responsibilities for seller relationships, ad placement and shopper experience. An attractive advertising revenue figure alone does not resolve those responsibilities.

The Commerce AI page describes learning from commerce activity to improve audience decisions, recommendations, creative and bidding. Those are vendor-described capabilities, not evidence that a particular retailer will generate more profitable sales. The practical mechanism is a feedback loop: product and event information inform ad decisions, and observed outcomes supply another round of signals.

02 / AudienceUseful where product data and advertising ownership are clear

An ecommerce advertiser with a maintained catalog, functioning purchase events and a person responsible for performance marketing has a concrete starting point. It can test whether Criteo reaches additional customers or improves an existing acquisition mix. A marketplace with established seller relationships has a different opportunity: helping sellers fund relevant placements while preserving the usefulness of the shopping experience.

The fit is weaker when catalog prices or availability routinely disagree with the storefront. AI cannot turn an unavailable product into a good shopping experience. Similarly, a small retailer without someone to reconcile spend and orders should account for the operating work before interpreting automation as a substitute for campaign ownership.

The AppLovin blueprint is a relevant paid-acquisition comparison, especially when the question is how a performance platform expands an advertiser’s reach. The Klaviyo blueprint addresses customer engagement and marketing from owned customer relationships. Paid retargeting and permissioned lifecycle messaging can complement each other, but should not receive duplicated credit for the same sale.

03 / WorkflowA proposed catalog campaign tests the data before the optimization

Consider a proposed campaign for a homewares retailer with many products but uneven repeat purchasing. The first objective is to establish whether a defined acquisition campaign brings profitable new orders. This example is an evaluation workflow, not a tested implementation or a forecast of Criteo performance. Retail-media monetization would be a later, separately scoped decision.

Begin with a small product subset whose stock, pricing and product identifiers are stable. The GO product-feed guide explains that feed files become the product catalog used for dynamic ads, supports scheduled imports and provides maintenance steps. Using an existing supported feed can save setup work, but the team should still compare representative records with the live store, including sale prices and unavailable variants.

Next validate the event stream. The GO onboarding guide requires events for campaign creation and separates billing, product-feed and event tasks. Create a controlled purchase and cancellation sequence and inspect the resulting records. The aim is to understand exactly what the optimization process will see, particularly when an order is amended, refunded or placed on another device.

Define new-customer status in the retailer’s own terms. Someone absent from one browser’s recent history may still be a longtime customer. Use the agreed customer and order records to assess the pilot, and document how unmatched shoppers are treated. Otherwise, a campaign that reaches existing purchasers can appear to solve an acquisition problem it has not actually tested.

Approve a creative and inventory scope that reflects the selected products. Review prices, delivery claims and landing destinations as they will appear to the shopper. For a product with several sizes or colors, decide whether the ad should lead to the exact variant or the broader product page. This small operational choice affects whether a relevant recommendation becomes a completed purchase.

Launch with a defined budget and observation window, then compare platform reporting with orders, refunds and contribution. Examine customer mix and total sales alongside attributed return on ad spend. A useful result is evidence of additional profitable demand after advertising costs, not simply more conversions credited to a new platform. Preserve a comparison group or other defensible incrementality design where feasible.

The onboarding guide says AI-assisted setup is available to selected participants. Do not make that optional path a dependency of the pilot. The standard data tasks still need verification even when the system detects a feed or event configuration automatically. Detection can simplify setup without proving that all commercially important fields have the intended meaning.

04 / PricingCommercial terms follow the product and buying route

The reviewed product pages route buyers to Talk to Criteo and do not provide a universal subscription tariff. The Commerce Growth FAQ specifies insertion-order contracts with cost-per-click or cost-per-thousand-impressions pricing. GO and a retailer’s Commerce Yield implementation still require their own commercial scope; these are not one interchangeable monthly plan. Request a proposal that identifies the operating model, actual billing basis and any minimum commitment for the selected route.

Predictive bidding describes adjusting bids using predicted user value. An optimized bid is a decision inside a campaign, not a promise of a fixed acquisition cost. Auction prices and conversion rates can change, and an advertiser’s target return should not be confused with a guaranteed commercial outcome. Assess the model using delivered spend and business results.

For Commerce Yield’s sponsored-product offer, the published page identifies a cost-per-click campaign model. That describes advertiser campaign charging, not necessarily the retailer’s full contract with Criteo. Ask separately about platform economics, service scope and the division of responsibilities. A retailer can earn advertising revenue while still taking on integration and seller-support costs.

In the proposed homewares pilot, keep implementation work separate from media investment. A product-feed repair benefits multiple channels; charging all of that work against one trial can distort comparison. Equally, excluding ongoing catalog maintenance makes an automated campaign look cheaper to operate than it is.

05 / DistinctionsCommerce context makes catalog quality commercially important

Criteo’s distinct focus is the connection between shopper behavior, products and media. That is useful when a buyer needs recommendations tied to a real inventory rather than generic audience categories. It also creates a specific dependency: catalog attributes, transactions and event quality influence which products and shoppers the system can meaningfully connect.

The company spans both advertiser demand and retailer monetization. An organization may therefore encounter Criteo as a media buyer, a retailer selling placements or a brand buying retail media. These relationships can coexist, but their performance metrics differ. Acquisition cost matters to one team; ad revenue and seller participation may matter to another; shoppers experience the combined result.

For a retailer building an ad business, the useful distinction is between filling more ad slots and creating valuable placements. A sponsored result should still satisfy the shopper’s query. The team should monitor search success, purchase behavior and seller concentration alongside ad revenue, rather than assuming that every additional advertising dollar represents an equal improvement to the store.

06 / QuestionsThree unresolved details can change the evaluation

Is the account using the expected product and setup route?

Confirm the exact Criteo product, available country and onboarding path with the representative or account. The public GO guide is concrete evidence of its setup requirements, but it does not establish that every prospective advertiser receives agentic onboarding. Avoid applying a feature or billing assumption from one interface to another.

Can the team reconcile ad recommendations with the catalog?

Ask how rejected products, delayed imports and unavailable items are surfaced. Keep representative product identifiers in the test record so that an incorrect creative can be traced back to its source. A successful feed import is less useful than confidence that the fields driving a real advertisement match the store.

What does reported return include?

Agree on attribution windows, order cancellations and existing-customer treatment before the pilot. Public product descriptions and vendor customer stories do not establish incremental lift for this business. This research did not operate an authenticated campaign or inspect a private commercial proposal; the workflow is designed to resolve those remaining questions with evidence.

07 / DecisionChoose one side of the commerce loop first

Evaluate Criteo against a specific job: acquire customers, retain them through paid media or build a retailer advertising business. The company’s AI is most useful when that job has reliable product and event inputs. Begin with the smallest commercially meaningful experiment and expand only when its economics and operational responsibilities are understood.