Retailmedia foundations and networks

Retail media foundations and networks

Pick the right commerce media network and the right surface (onsite, offsite, in-store) for a brief, then set the expectation for how closed-loop measurement and new-to-brand will read. Retail media is programmatic buying where the seller owns first-party purchase data and can attribute ad exposure to a sale, so the network choice is really a data and audience choice, not just an inventory choice.

This skill is the map and the selection logic. It does not run campaigns inside any one network. For Amazon buying in depth, hand off to the amazon-dsp-* skills and amazon-marketing-cloud. For packaging retail audiences onto open-web inventory through sell-side deals, hand off to curation-and-curated-marketplaces. For trimming the bid-to- impression path, hand off to curation-and-supply-path-optimization. For KPI math (ROAS, CPA, incrementality), see programmatic-foundations.

When to use this skill

• "Which retail media network should we buy?" / "Amazon vs Walmart vs Kroger vs Instacart?"
• "What is the difference between commerce media and retail media?"
• "Onsite or offsite? Do I need in-store screens?"
• "Sponsored products vs sponsored display vs a DSP buy on this network."
• "How does closed-loop measurement work?" / "Why does this network report different sales?"
• "What is a good new-to-brand rate?" / "Is this campaign actually acquiring customers?"
• "Can I use Walmart or DoorDash data off their site?" (offsite activation and clean rooms)
• "What is AMC / a retail clean room and when do I need one?"

Boundaries with sibling skills:

• Running Amazon DSP campaigns, audiences, bidding, reporting: the amazon-dsp-* skills.
• Amazon clean-room analysis and audience building: amazon-marketing-cloud.
• Buying retail audiences as a curated Deal ID on open-web supply: curation-and-curated-marketplaces.
• Reducing hops and fees on the supply path: curation-and-supply-path-optimization.
• CTV pod mechanics when a retail network sells streaming: ctv-supply-and-ad-pods.
• Stitching retail conversions into one warehouse: marketing-analytics-warehouse-and-dashboards.

Quick reference

Brief	Lead network(s)	Lead surface	Why
Sell more of a product that lives on Amazon	Amazon Ads (Sponsored Products, then DSP)	Onsite search, then onsite/offsite display	Purchase intent and the buy button are on the page
Grocery/CPG, big-basket, US mass retail	Walmart Connect, Kroger Precision Marketing	Onsite + offsite + in-store	Loyalty-card purchase data and store footprint
Same-day grocery delivery, recipe/basket context	Instacart	Onsite sponsored + offsite (Carrot Ads network)	First-party cart and SKU-level data at add-to-cart
Food delivery, restaurant/CPG, high new-customer rate	DoorDash Ads, Uber Advertising	Onsite sponsored + offsite + CTV	Logged-in intent, strong new-to-advertiser reach
Reach a retailer's shoppers off their property	The retailer's offsite program or a curated deal	Offsite open web, social, CTV	Activate the audience where the impression is cheap
Non-endemic brand (auto, finance, travel) wanting purchase data	Chase Media Solutions, retailer offsite/clean room	Offsite + card-linked offers	Spend data without a product on the retailer's shelf
Prove exposure drove sales, privacy-safe	Any network with a clean room (AMC and peers)	Measurement layer, not a buy	Aggregated, log-level joins without exporting PII

Three surfaces, three jobs:

• Onsite: ads on the retailer's own app or site (search results, product pages). Highest intent, closest to the purchase, most limited scale.
• Offsite: the retailer's first-party audience activated on the open web, social, or CTV. Bigger reach, lower intent, needs a clean measurement join back to sales.
• In-store: screens, audio, and self-checkout in physical stores. Reaches the shopper at the shelf but is the hardest to attribute at the individual level.

Core process

1. Name the objective and where the product is bought. If the product sells on the retailer, the onsite search ad sits next to the purchase and usually wins first dollar. If the goal is reach or acquisition, weight offsite. The network that owns the transaction owns the attribution, so pick the network whose shoppers actually buy this product.
2. Match the audience to the network's data. Amazon has broad shopping and browsing signal; Walmart and Kroger have loyalty-card grocery baskets and store trips; Instacart has SKU-level cart data; DoorDash and Uber have logged-in food and mobility intent; Chase has card-spend data across merchants. Choose the network whose first-party data best describes your buyer.
3. Pick the surface per the Quick reference. Start onsite for conversion, add offsite for reach and new customers, add in-store only when the shelf moment matters and you accept weaker attribution.
4. Decide onsite ad type before DSP. On most networks the ladder is sponsored search first (keyword or category, closest to the buy button), then sponsored display on product pages, then the network DSP for audience-based display, video, and offsite. Do not jump to the DSP before the sponsored-search shelf is defended.
5. Set the measurement frame up front. Confirm the attribution window, whether reporting is click-based, view-through, or clean-room modeled, and whether new-to-brand (or new-to- advertiser) is available. Networks report their own numbers on their own methodology, so agree the metric definition before launch, not after.
6. Decide if you need a clean room. If you must join the network's exposure data to your own first-party or CRM data for incrementality, audience building, or cross-media dedup, plan a clean-room workflow (AMC on Amazon, equivalents elsewhere). Aggregated outputs only, no PII export. For AMC specifically, hand off to amazon-marketing-cloud.
7. Human-gate the spend. Recommend the network, surface, budget split, and target metric, then stop and let a human approve budgets and go live. Never launch spend or push a live config change automatically.

Decision rules and thresholds

Commerce media vs retail media

Retail media is the subset owned by retailers (Amazon, Walmart, Kroger, Target, Instacart). Commerce media is the wider set that includes delivery, mobility, travel, and financial players that sell media against first-party transaction data (DoorDash, Uber, Chase, airlines, hotels). DoorDash calls its expanded onsite-plus-offsite program a commerce media platform. Treat them the same way operationally: whoever owns the transaction owns the closed loop. The label matters less than "does this network's data describe my buyer, and can it attribute a sale."

Onsite first, offsite for reach

• Endemic brand, product on the retailer's shelf, conversion goal: onsite sponsored search is first dollar. It is the highest-intent, lowest-waste surface because it sits on the query.
• Reach, awareness, or new-customer goals: shift budget offsite once the onsite shelf is defended. Offsite buys the same audience on cheaper open-web, social, or CTV impressions.
• In-store: add only when the shelf decision is the moment that matters (impulse, category disruption) and you accept that individual-level attribution is weaker than digital.

Non-endemic brands

If your product does not sell on the retailer (a bank, an automaker, a streaming service), you cannot run sponsored products there. Your play is the network's offsite/DSP program to reach their shoppers, a card-linked-offer network like Chase Media Solutions that monetizes spend data, or a curated deal that packages the retailer's audience onto open-web supply (see curation-and-curated-marketplaces).

Offsite activation and data leaving the property

Several retailers now let you activate their first-party audience and run closed-loop measurement on media you buy in your own platforms, not just on the retailer's site. Walmart Connect is expanding first-party data access to outside platforms (Yahoo DSP as a launch partner, VIZIO CTV inventory via Magnite supply, in a phased rollout as of May 2026). DoorDash Ads runs offsite through an acquired ad-tech layer and matches to advertiser data through a data-collaboration partner for privacy-safe measurement (June 2026). When you activate a retailer's audience offsite, insist on the measurement join back to sales, because offsite reach without the closed loop is just another audience segment.

New-to-brand as an acquisition gauge

New-to-brand (Amazon's term) or new-to-advertiser (used by delivery networks) counts buyers who had not purchased from the brand in a lookback window. Use it to judge whether a campaign acquires customers or just harvests existing ones. Rules of thumb, not guarantees:

• Offsite and delivery surfaces skew high on new customers by design. Reported figures in 2025 and 2026 include a median of 52% new-to-brand on Walmart Connect offsite display, and over 80% new-to-advertiser reach on DoorDash test campaigns. Treat vendor figures as directional.
• A very low new-to-brand rate on a prospecting line usually means you are retargeting existing buyers. Rebalance toward upper-funnel audiences or offsite.
• A very high new-to-brand rate on a line meant to defend the shelf can mean you are buying incremental reach you did not intend. Match the metric to the line's job.

Full benchmark ranges and how each network defines and windows the metric live in references/new-to-brand-benchmarks.md.

When a clean room is worth it

Use a retail clean room (AMC on Amazon, and clean-room equivalents on other networks) when you need to join exposure to your own data for incrementality, custom audiences, or cross-media dedup, and you accept aggregated, non-exportable outputs. Do not stand up a clean room just to pull standard campaign reports; the network console already gives those. AMC on Amazon is privacy-safe (data cannot be exported, outputs are aggregated) and carries a 25-month lookback and custom-model support. For AMC workflows, hand off to amazon-marketing-cloud.

Reference material

• references/commerce-vs-retail-media.md: the taxonomy (retail vs commerce media), the onsite/offsite/in-store surface map, endemic vs non-endemic, and closed-loop measurement mechanics. Read this when someone confuses the categories or asks how attribution works.
• references/network-set-and-surfaces.md: a network-by-network map (Amazon, Walmart, Kroger, Instacart, DoorDash, Uber, Chase) with the data each owns, the surfaces and ad types it sells, its DSP and clean-room posture, and its offsite/curation path. Read this when choosing a network or briefing a specific one.
• references/new-to-brand-benchmarks.md: how each network defines new-to-brand or new-to- advertiser, the lookback windows, directional benchmark ranges by surface, and the pitfalls in comparing the metric across networks. Read this when setting or defending an acquisition target.

Templates and examples

• CPG snack brand, product on Amazon and Walmart, goal is unit sales plus new buyers. Plan: Amazon Sponsored Products to defend branded and category search, Walmart Connect Sponsored Search for the mass-retail basket, then Amazon DSP and Walmart offsite display for reach. Target metric: onsite ROAS on sponsored search, new-to-brand on the DSP and offsite lines. Measurement: AMC for cross-line incrementality on the Amazon side. Human approves budgets.
• Regional grocery push, loyalty-driven, goal is trips and basket size. Plan: Kroger Precision Marketing onsite sponsored plus offsite against loyalty-card segments, in-store screens in target banners for the shelf moment. Target metric: sales lift on card-linked purchases.
• Auto brand (non-endemic), goal is qualified reach among high-income grocery shoppers. Plan: no sponsored products (no shelf); buy the retailer's offsite/DSP audience or a curated Deal ID packaging that audience onto CTV and open web via curation-and-curated-marketplaces; measure reach and new-to-advertiser, and store visitation if the network offers it.
• Restaurant chain, goal is new customers and orders. Plan: DoorDash Ads sponsored listings plus the homepage Spotlight format onsite, offsite through the network's ad-tech layer, matched to first-party data through the network's data-collaboration partner for closed-loop measurement. Target metric: new-to-advertiser share and cost per incremental order.

Common pitfalls

• Comparing networks on the same "ROAS" number. Each network attributes on its own window and methodology (click, view-through, or clean-room modeled), so a 6x on one is not a 6x on another. Normalize the definitions before you compare, or compare incremental sales in a clean room.
• Jumping to the DSP before defending the shelf. On endemic buys, sponsored search is first dollar. Spending on audience display while a competitor owns your branded search leaks sales.
• Treating offsite reach as closed-loop by default. Offsite is only closed-loop if the measurement join back to the retailer's purchase data is in place. Confirm it exists before you count offsite as retail media rather than plain open-web reach.
• Assuming in-store attributes like digital. In-store screens and audio reach the shopper at the shelf but attribute at the store or basket level, not the individual. Set that expectation with the client up front.
• Running a non-endemic brand toward sponsored products. There is no shelf to sponsor. Route to offsite, card-linked offers, or a curated audience deal instead.
• Taking vendor new-to-brand figures as guaranteed. They are directional and self-reported on the vendor's own definition. Validate against your own repeat-purchase data where possible, and see references/new-to-brand-benchmarks.md.
• Standing up a clean room for reports you can already pull. Clean rooms are for joins and incrementality, not for standard campaign metrics the console already returns.

Sources

• Walmart Connect: expanding access to Walmart first-party data across leading platforms (as of July 2026)
• DoorDash Ads becomes a global commerce media platform (as of July 2026)
• Amazon Certified Supply Exchange (as of July 2026)
• Amazon Marketing Cloud (as of July 2026)
• Amazon DSP Performance+ tactics (as of July 2026)
• Brand+ and Performance+: two powerhouse tools (as of July 2026)
• The Trade Desk OpenPath (as of July 2026)
• The trader's guide to supply path optimization (SPO) (as of July 2026)