Table of contents
About Attribution
In retail media advertising, attribution is the process through which one identifies which sets of actions led to the sale of a product.
No matter the types of demand accounts in which campaigns are running, sales attribution is measured. It determines what part of the advertiser's retail media activity led to purchases by the shoppers that they reached.
We measure retail media attribution in our internal reports through Attributed Sales and ROAS (Return on Advertising Spend) metrics.
Our attribution methodology is made of three things:
User Matching: We only attribute purchases made by shoppers reached with the advertiser's ad(s).
SKU Matching: We only attribute purchases of products that were related to advertisers’ ad(s)
Look Back Window: We only attribute purchases made within a selected time after the advertiser's ad were either clicked on or seen.
For campaigns you manage, attribution settings can be changed at any time, and all data and dashboards in the platform will update retroactively to match your new settings in just a few hours.
User Matching Rules
Criteo's attribution models will rely on four user identifiers to determine whether a purchase is related to advertiser's ad(s) or not:
The retailer’s CRM ID
The retailer’s 1st party cookie ID
The Criteo cookie ID
The Criteo X-device ID
All of our user matching rules are purely deterministic.
We never use IP addresses or probabilistic models to determine attribution at the user level.
Lookback Windows
The conversion cycle can be drastically different between a shopper buying groceries and a shopper buying a TV. ROAS measurement can also be different depending on campaign objectives (lower or upper funnel) and the type of ads that are running.
As a result, to give advertiser's more flexibility, we provide flexible attribution windows that can be selected at the campaign level:
Post-click: 7, 14, 30 days
Post-view: None, 1, 7, 14, 30 days
In the beginning, we recommend using the default attribution settings, to compare apples to apples. Then, as more campaigns are turned on, advertisers can consider changing the model to see the impact of those changes.
Product Matching
The attribution model works differently for Demand Brand and Demand Seller accounts. It considers products as attributable if:
Demand Brand Accounts
Same Product
The same product as the one promoted in the ad is purchased by a shopper who was reached. We call this same product attribution.
Same Category
One of the products that is within the same category and brand as the promoted one is purchased by a shopper you’ve reached. We call this same category attribution.
Same Brand
Any product within the same brand as the promoted one is purchased by a shopper you've reached. We call this same brand attribution.
Demand Seller Accounts
Same Product
It refers to situations where the advertised product and the purchased product are exactly the same, and both are sold by the same advertising seller.
Same Category
It refers to cases where the advertised and purchased products are different but belong to the same product category, and both are sold by the same advertising seller.
Same Seller
It refers to situations where the advertised and purchased products are different and belong to different categories, but both are sold by the same advertising seller.
Sales Deduplication
A shopping journey can involve multiple advertising events (such as clicks and impressions) that lead to an eventual conversion to purchase. Sales deduplication means that a sales conversion is attributed once to a single advertising event (deduplication) rather than multiple events (duplication).
The attribution model ensures that this single attributed event is the most relevant event that led to the purchase.
To determine which of an ad’s advertising events is the most attributable event, the attribution model applies the following prioritization rules in descending order of priority:
Clicks take precedence over impressions.
Same SKU events take precedence over Same Category events.
If SKU events and user events yield the same characteristics, the event closest to the time of purchase takes precedence.