Retail Media
Retail media network architecture: the nine components, how data flows, and a diagram you can reuse
Published · Last updated
Next scheduled review:
A retail media network architecture has nine parts. Five handle data and decisions: an ad server and auction, a product catalog, identity and loyalty data, an audience builder and a clean room. Four reach shoppers and report: offsite activation, in-store screens, measurement, and a self-serve console with an API. Shopper data stays at the centre.
Contents
Key facts
- Onsite ads are becoming programmatic. The Trade Desk began selling onsite retail placements through Koddi on 9 October 2025, with Gopuff first (PPC Land, October 2025). Pentaleap and Teads launched real-time bidding for sponsored product ads on 24 July 2025 (PPC Land, July 2025).
- Offsite data now flows into the buyer's own tools. Walmart Connect audiences became available in Google Display & Video 360 for YouTube on 11 June 2026 (PPC Land, 11 June 2026). Kroger's SKU-level sales reports reached the same tool in March 2026, built on LiveRamp and MetaRouter (PPC Land, 24 March 2026).
- The matching standard is PAIR. IAB Tech Lab has finalised PAIR version 1.1 and says it replaces the earlier OPJA specification for matching advertiser and publisher first-party data in clean rooms (IAB Tech Lab, accessed 9 November 2026).
- Clean rooms are not always needed. An IAB framework from September 2026 sorts 14 commerce data use cases and finds that nine, including attribution and incrementality, can often run on existing tools (PPC Land on IAB, 9 October 2026).
- Conversion data for clean rooms has a format. IAB Tech Lab published final guidelines for sending server-side conversion events to clean rooms on 3 November 2026 (PPC Land, 4 November 2026).
- Stores are now a large ad surface. CRI took over Albertsons' in-store network of about 3,000 screens, serving roughly one million ads a day (invidis, 18 August 2026).
Definition
Retail media network architecture: the software, data stores and integrations a retailer uses to sell ads on its own and partner surfaces, target them with shopper data and report sales to advertisers.
For the business definition of the category, see the retail media hub and the glossary entry for a retail media network.
The nine components at a glance
9 of 9 rows
1 | Ad server and auction | Picks which sponsored product, banner or video fills each slot, and prices it | Ad requests from site and app, campaigns, bids, catalog, audiences | Ad decisions, impression and click logs | Usually licensed; the largest networks build |
2 | Catalog and product feed | Keeps every SKU, price, stock level and category in sync with the ad server | Product information system, store stock | SKU list with eligibility flags | Built; it already exists in e-commerce |
3 | Identity and loyalty data | Links online and in-store purchases to one pseudonymous shopper | Loyalty sign-ups, logins, card transactions | Pseudonymous IDs, purchase history | Built; identity partners for offsite matching |
4 | Audience builder | Turns purchase history into segments such as "bought oat milk in the last 60 days" | IDs and purchase history | Segments, counts, lookalikes | Often built on a CDP |
5 | Clean room | Lets a brand match its own data with the retailer's without either side seeing raw records | Retailer IDs, brand IDs, exposure logs | Aggregated overlaps, reach, sales results | Usually licensed |
6 | Offsite activation | Sends retailer audiences to DSPs and SSPs so ads run on other websites, apps and TV | Segments or matched IDs | Campaign delivery and exposure logs | Partnered; a few build their own DSP |
7 | In-store screens | Plays ads on store screens and audio and proves each play | Playlists, store IDs, campaign schedule | Proof-of-play logs, footfall estimates | Licensed screen software plus owned hardware |
8 | Measurement and reporting | Ties exposures to sales and runs incrementality tests | Exposure logs from 1, 6 and 7, sales from 3 | Attributed sales, new-to-brand, lift | Built logic on licensed tools |
9 | Self-serve console and API | Lets brands and agencies plan, buy, pay and pull reports themselves | Campaign setup, budgets, creative | Reports, invoices, API feeds | Licensed, then branded |
1
- Component
Ad server and auction
- What it does
Picks which sponsored product, banner or video fills each slot, and prices it
- Data in
Ad requests from site and app, campaigns, bids, catalog, audiences
- Data out
Ad decisions, impression and click logs
2
- Component
Catalog and product feed
- What it does
Keeps every SKU, price, stock level and category in sync with the ad server
- Data in
Product information system, store stock
- Data out
SKU list with eligibility flags
3
- Component
Identity and loyalty data
- What it does
Links online and in-store purchases to one pseudonymous shopper
- Data in
Loyalty sign-ups, logins, card transactions
- Data out
Pseudonymous IDs, purchase history
4
- Component
Audience builder
- What it does
Turns purchase history into segments such as "bought oat milk in the last 60 days"
- Data in
IDs and purchase history
- Data out
Segments, counts, lookalikes
5
- Component
Clean room
- What it does
Lets a brand match its own data with the retailer's without either side seeing raw records
- Data in
Retailer IDs, brand IDs, exposure logs
- Data out
Aggregated overlaps, reach, sales results
6
- Component
Offsite activation
- What it does
Sends retailer audiences to DSPs and SSPs so ads run on other websites, apps and TV
- Data in
Segments or matched IDs
- Data out
Campaign delivery and exposure logs
7
- Component
In-store screens
- What it does
Plays ads on store screens and audio and proves each play
- Data in
Playlists, store IDs, campaign schedule
- Data out
Proof-of-play logs, footfall estimates
8
- Component
Measurement and reporting
- What it does
Ties exposures to sales and runs incrementality tests
- Data in
Exposure logs from 1, 6 and 7, sales from 3
- Data out
Attributed sales, new-to-brand, lift
9
- Component
Self-serve console and API
- What it does
Lets brands and agencies plan, buy, pay and pull reports themselves
- Data in
Campaign setup, budgets, creative
- Data out
Reports, invoices, API feeds
Table by AdSovereign, CC BY 4.0.
Vendor capabilities and customers are documented with sources in retail media network platforms compared. The table above lists examples, not a ranking.
How data flows between the components
Six flows run the network. Each one carries a different kind of data, and that decides where it can go.
- Catalog sync. The catalog sends SKUs, prices and stock to the ad server many times a day, so ads never promote items that are out of stock.
- Identity build. Loyalty and checkout systems send transactions to the identity layer, which links them to a pseudonymous shopper ID. Names and emails stay in the retailer's systems.
- Ad decision. When a shopper searches, the site asks the ad server for ads. The ad server runs an auction among eligible campaigns and returns the winners in the page response. Osmos says its ad server answers in under 30 milliseconds (company claim, Osmos, accessed 9 November 2026).
- Audience export. The audience builder sends segments to the clean room or straight to a DSP as hashed or pseudonymous IDs. Kroger's route into DV360 uses LiveRamp's RampID, so "neither party" handles raw personal data, according to PPC Land's account of the launch (PPC Land, 24 March 2026).
- Exposure return. Onsite logs, offsite delivery logs and in-store proof-of-play files flow back to measurement. MetaRouter routes Kroger's online and in-store events back into DV360 reports (PPC Land, 25 March 2026).
- Reporting and billing. Measurement joins exposures with sales and writes results to the console and the API. Billing reads the same logs, so invoices and reports agree.
The RTB architecture explainer covers the auction mechanics in flow 3 in more depth.
What an onsite ad request looks like
There is no common standard for sponsored product calls, so each platform defines its own fields. The example below is illustrative only. It shows the minimum an ad server usually needs and what it returns.
{
"request_id": "7f3c2e19",
"page_type": "search",
"query": "oat milk",
"slots": 2,
"store_id": "0412",
"shopper_id": "lp_9a41c0de",
"consent": {"personalised_ads": true}
}The response names the SKU, the campaign that paid for it and the tracking addresses that feed measurement.
{
"request_id": "7f3c2e19",
"ads": [
{"slot": 1, "sku": "012345678905", "campaign_id": "cmp-881", "pricing": "cpc", "impression_url": "https://ads.retailer.example/i/881", "click_url": "https://ads.retailer.example/c/881"},
{"slot": 2, "sku": "098765432109", "campaign_id": "cmp-417", "pricing": "cpc", "impression_url": "https://ads.retailer.example/i/417", "click_url": "https://ads.retailer.example/c/417"}
]
}Note what is missing: the shopper's name, email and full basket never leave the retailer. The shopper ID is a pseudonym the ad server can use for frequency and targeting rules.
The components one by one
Ad server and auction. This is the heart of onsite media. It must rank sponsored products next to organic results without making search worse, so relevance scoring matters as much as price. Selling these slots through outside DSPs is new: The Trade Desk and Koddi opened programmatic buying of onsite placements in October 2025 (PPC Land, October 2025). A network that wants that demand needs an ad server that speaks OpenRTB 2.6, the current version of IAB Tech Lab's bidding protocol (IAB Tech Lab on GitHub, accessed 9 November 2026).
Catalog and product feed. Ads sell SKUs, so the catalog is the inventory list. Stock and price changes must reach the ad server fast. A promoted product that is out of stock wastes the brand's money and the shopper's time.
Identity and loyalty data. Loyalty data is what makes retail media different from other ad networks. It links a card swipe in a store to a login online. Tesco, for example, says 84% of its transactions carry a Clubcard (Digiday, 2 December 2025). The engineering choices behind this layer are covered in identity graph architecture.
Audience builder. Segments are built from what people buy, not what they browse. Retailers usually keep this in-house because it is where their data science sits. Kroger's runs through its 84.51° unit (PPC Land, 25 March 2026).
Clean room. A clean room lets a brand join its customer list with the retailer's shoppers and see only aggregates. It is not always needed. The IAB's September 2026 framework says a clean room adds value mainly when data sits across separate organisations (PPC Land, 9 October 2026). When one is needed, the IAB Tech Lab's November 2026 guidelines define how conversion data should be packaged (PPC Land, 4 November 2026). Vendor options and costs are in clean room architectures compared.
Offsite activation. Offsite media runs on other publishers through a DSP or an SSP. The trend in 2026 is to send retailer data into the buyer's existing DSP rather than force a new one. Walmart did this with DV360 in June 2026 (PPC Land, 11 June 2026). The flows and their clean room needs are covered in offsite retail media explained.
In-store screens. In-store media needs screen software, players, content rules and proof that each ad played. Albertsons moved its network of about 3,000 screens to CRI's content system in 2026 (invidis, 18 August 2026). Broadsign and Mirakl Ads announced a joint offer in June 2026 to book online and in-store campaigns together (PPC Land, 27 June 2026). Formats and measurement are covered in in-store retail media.
Measurement and reporting. Measurement joins exposure logs with sales, reports attributed sales and runs tests. The IAB/MRC guidelines of January 2024 and the IAB in-store standards of 3 December 2024 set the rules (IAB, 3 December 2024). Stores make good test units: Albertsons compares stores with screens against matched control stores (Modern Retail, 20 January 2026). The full rulebook is summarised in retail media measurement standards explained.
Self-serve console and API. A console lets smaller brands buy without a sales call. Tesco Media launched its Ad Manager for sponsored products at its 2026 Upfront and plans to add display and video (Grocery Trader, 9 October 2026). An API matters as much, because large agencies run campaigns from their own tools.
The retail media network ecosystem
The network does not sell alone. Around it sit four groups: brands and agencies that buy; DSPs, SSPs and exchanges that carry offsite and programmatic demand; clean room, identity and measurement vendors; and screen operators in stores. IAB Europe's April 2026 guide lists the systems a retailer must connect: customer data platforms, clean rooms, personalisation engines, analytics, retail ad servers and in-store networks (PPC Land on IAB Europe, 17 April 2026).
Some retailers also rent the whole stack from another retailer or a delivery app. Instacart powers more than 310 grocery ecommerce sites (PPC Land, 29 October 2026). That model trades control for speed. The side-by-side view of onsite, offsite and in-store is in onsite vs offsite vs in-store retail media.
Build or buy, component by component
Most retailers mix. A useful rule is to build what holds your data and your relationships, and license what is generic.
- Build: the catalog link, identity and loyalty data, the audience builder and the measurement logic. These hold the data and the methods buyers will question. Ahold Delhaize USA went further and built its own ad platform, Edge, for onsite display, sponsored search and store screens. Its banners reach more than 26 million shoppers a week (Supermarket News, 21 January 2026).
- License: the ad server for most retailers, screen software, and the clean room engine. These are hard to build well and the vendors are many.
- Partner: offsite DSPs and SSPs, where the buyer's choice of platform matters more than the retailer's.
- Decide case by case: the self-serve console. A white-label console is fast; an owned one keeps the brand relationship.
The trade-offs, costs and timelines are set out in retail media infrastructure: build vs buy. Whether the network will earn enough to fund any of it is the subject of retail media network revenue benchmarks.
Example. Picture a regional grocer with a few hundred stores. It licenses an ad server and a screen system. It keeps loyalty data and segments in its own warehouse. It uses a licensed clean room only for brands that bring their own customer lists. And it sends segments to the two DSPs its largest suppliers already use. That covers all nine boxes with three contracts and no new copies of shopper data outside its control.
Retail media network diagram
- Layout. Three vertical zones from left to right. Left zone, shaded, labelled "Retailer-controlled". Middle zone labelled "Partners". Right zone labelled "Advertisers and agencies".
- Left zone boxes (top to bottom). "2 Catalog and product feed"; "3 Identity and loyalty data" (drawn with a padlock icon and the note "names and emails stay here"); "4 Audience builder"; "1 Ad server and auction"; "8 Measurement and reporting". Draw a store icon at the bottom of the left zone with "7 In-store screens".
- Middle zone boxes. "5 Clean room" straddling the border between left and middle zones; "6 Offsite activation: DSPs and SSPs"; a small box "Screen software" next to box 7.
- Right zone boxes. "9 Self-serve console and API" straddling the middle and right zones; "Brands and agencies".
- Surfaces. Above the left zone, three small surface tiles: "Website", "App", "Email". Above the middle zone: "Open web", "CTV", "Social".
- Arrows 1 to 3. (1) Box 2 to box 1, solid, label "SKUs, price, stock". (2) Store and website icons to box 3, solid, label "transactions". (3) Website and App tiles to box 1 and back, double-headed, label "ad request / ad decision".
- Arrows 4 to 6. (4) Box 4 to box 5 and box 6, dashed, label "pseudonymous IDs or segments". (5) Boxes 1, 6 and 7 to box 8, dotted, label "exposure logs, proof of play". (6) Box 8 to box 9, solid, label "reports and invoices".
- Unnumbered arrows. From "Brands and agencies" to box 9, label "campaigns and budgets". From box 9 to box 1, label "campaign setup".
- Legend. Solid line: retailer data inside the perimeter. Dashed line: pseudonymous data leaving the perimeter. Dotted line: logs returning for measurement.
- Caption. "Shopper data stays in the retailer zone; only pseudonymous IDs or aggregates cross into partner systems, and exposure logs return for measurement."
- Alt text. "Diagram of a retail media network. On the left, inside a shaded retailer zone, are the catalog, identity and loyalty data, audience builder, ad server and measurement, with in-store screens below. A clean room and offsite DSPs sit in a partner zone in the middle. A self-serve console connects brands on the right. Numbered arrows show catalog sync, identity build, ad decisions, audience export, exposure logs returning and reports going to brands."
FAQ
What are the components of a retail media network?
What does a retail media network diagram show?
Do you need a clean room to run a retail media network?
Should a retailer build or buy its ad server?
How do in-store screens connect to the rest of the network?
What is the retail media network ecosystem?
How this works on owned infrastructure
The diagram above has a shaded retailer zone and a partner zone, and the main design choice is which boxes sit inside the shading. AdSovereign is an enterprise RTB stack (DSP, SSP, exchange, DMP and clean room) designed to run in a retailer's own data centre or private cloud tenancy. Deployed that way, it can fill the auction, audience, clean room and offsite boxes inside the retailer zone, while the catalog, loyalty system and screen software stay with the systems the retailer already runs. Pseudonymous IDs then cross the boundary only when the retailer chooses to send them. The architecture page maps its components to boxes like these, and the integrations page lists the connectors to catalogs, DSPs and screen systems.
Sources
- PPC Land, "TTD enables programmatic retail media buying through Koddi partnership" (October 2025, on the announcement of 9 October 2025). https://ppc.land/ttd-enables-programmatic-retail-media-buying-through-koddi-partnership/
- PPC Land, "Retail media networks embrace RTB for sponsored products" (25 July 2025). https://ppc.land/retail-media-networks-embrace-rtb-for-sponsored-products/
- PPC Land, "Walmart Connect brings real purchase data into Google DV360 for YouTube ads" (11 June 2026). https://ppc.land/walmart-connect-brings-real-purchase-data-into-google-dv360-for-youtube-ads/
- PPC Land, "Kroger joins Google's Commerce Media Suite to unlock SKU-level sales data on YouTube" (24 March 2026). https://ppc.land/kroger-joins-googles-commerce-media-suite-to-unlock-sku-level-sales-data-on-youtube/
- PPC Land, "MetaRouter: the hidden engine inside Google's Kroger ad deal" (25 March 2026). https://ppc.land/metarouter-the-hidden-engine-inside-googles-kroger-ad-deal/
- IAB Tech Lab, "Publisher Advertiser Identity Reconciliation (PAIR)" (version 1.1, accessed 9 November 2026). https://iabtechlab.com/pair/
- PPC Land, "IAB says 9 of 14 commerce use cases often work without a clean room" (9 October 2026). https://ppc.land/iab-says-9-of-14-commerce-use-cases-often-work-without-a-clean-room/
- IAB, "Data Clean Rooms in Commerce: Common Use Cases" (September 2026). https://www.iab.com/guidelines/data-clean-rooms-in-commerce-common-use-cases/
- PPC Land, "IAB Tech Lab clean room rules allow unhashed emails under technical controls" (4 November 2026, on guidelines of 3 November 2026). https://ppc.land/iab-tech-lab-clean-room-rules-allow-unhashed-emails-under-technical-controls/
- IAB Tech Lab, OpenRTB 2.x specification repository, "from 2.6 onward" (accessed 9 November 2026). https://github.com/InteractiveAdvertisingBureau/openrtb2.x
More sources (12)
- invidis, "Retail Media: CRI Takes Over Albertsons from Stratacache" (18 August 2026). https://invidis.com/news/2026/08/retail-media-cri-takes-over-albertsons-from-stratacache/
- Modern Retail, "Albertsons to put digital screens for ads in 800-plus stores" (20 January 2026). https://www.modernretail.co/technology/albertsons-is-putting-digital-screens-for-ads-in-more-than-a-third-of-its-stores/
- PPC Land, "Broadsign and Mirakl Ads want to end the online vs in-store retail media split" (27 June 2026). https://ppc.land/broadsign-and-mirakl-ads-want-to-end-the-online-vs-in-store-retail-media-split/
- IAB and IAB Europe, "In-Store Retail Media: Definitions and Measurement Standards" (3 December 2024). https://www.iab.com/guidelines/in-store-retail-media/
- IAB, "Final IAB/MRC Retail Media Measurement Guidelines" (January 2024). https://www.iab.com/guidelines/final-iab-mrc-retail-media-measurement-guidelines-available-for-download/
- Grocery Trader, "Tesco Media unveils new ad platform, shoppable formats and smarter in-store media" (9 October 2026). https://grocerytrader.co.uk/tesco-media-unveils-new-ad-platform-shoppable-formats-and-smarter-in-store-media/
- Digiday, "Tesco wants to be comparable to traditional media partners" (2 December 2025). https://digiday.com/marketing/u-k-retailer-tesco-wants-advertisers-to-see-it-as-equal-to-any-traditional-media-partner/
- PPC Land, "IAB Europe retail media guide: what's actually holding retailers back" (17 April 2026, on the guide of 15 April 2026). https://ppc.land/iab-europe-retail-media-guide-whats-actually-holding-retailers-back/
- PPC Land, "Instacart ad revenue gains 23% to $331m in quarter before DoorDash deal" (29 October 2026). https://ppc.land/instacart-ad-revenue-gains-23-to-331m-in-quarter-before-doordash-deal/
- Supermarket News, "Grocery retail media enters a new growth phase: report" (21 January 2026). https://www.supermarketnews.com/retail-media/grocery-retail-media-enters-a-new-growth-phase-report
- Osmos, "Osmos vs CitrusAd" (undated, accessed 9 November 2026). https://www.osmos.ai/osmos-vs-citrusad
- IAB Europe, "Data clean rooms for retail and commerce media: two guides, one journey" (29 September 2026). https://iabeurope.eu/data-clean-rooms-for-retail-and-commerce-media-two-guides-one-journey/
Changelog
- 9 November 2026: first edition, nine components, six flows, diagram spec.