TheCardinal Source
Start an engagement →

The Analyst Note   Retail series  ·  piece 5 of 8  ·  Last updated September 2026

Retail media networks in 2026: buying the tech stack vs. building it.

Retail media crossed $50 billion in North American ad spend and remains the fastest-compounding channel in commerce marketing. Mid-market retailers evaluating whether to stand up an in-house RMN in 2026 systematically underestimate the identity graph, the measurement pipeline, and the campaign management surface by a factor of five to ten. The buy-side platform stack is more capable than most builders admit.

12 min · Deep-dive · Retail series, piece 5 of 8

Questions this article answers

  • What is a retail media network actually made of, technically?
  • What is the honest capex and opex floor for building one?
  • Which off-the-shelf platforms fit mid-market scale?
  • What is the identity graph and clean-room requirement?
  • How is measurement — MMM, incrementality, closed-loop — evolving in 2026?
  • What does campaign management and self-service look like at mid-market volume?
  • What sponsor-vs-endemic ad ratio economics should a retailer plan for?

Retail media is no longer a category question. It is a build-vs-buy question. The IAB and IAB Europe have released a stack of standards through 2025 and into 2026 — the In-Store Retail Media Definitions and Measurement Standards, the Guidelines for Incremental Measurement in Commerce Media, and the Retail Media Advanced Measurement and Data Collaboration guide — that collectively formalize what an RMN must do. Mid-market retailers reading those documents and pricing an in-house build against them are the audience for this note. The finding is consistent: the buy-side stack has matured faster than the build-side economics.

A retail media network is six coupled systems, not one.

An RMN combines an ad server or supply-side platform, a demand integration layer that plugs into DSPs and direct advertisers, an identity graph that ties the loyalty file to online and in-store behavior, a clean room that lets advertisers bring their own audiences without exchanging raw PII, a measurement pipeline that produces incrementality and marketing-mix outputs, and a self-service campaign management portal for the advertiser side. The first two look like ad tech. The last four are commerce data problems. Mid-market retailers pricing a build usually scope the first two and treat the last four as follow-on work. That is where the five- to ten-times underestimation comes from.

The honest capex floor for a build is $2M in year one, plus a ten-person team.

For a retailer at 100 to 500 stores with a functional loyalty program and a serviceable ecommerce site, an in-house build lands around $2M year-one platform capex — largely engineering — plus an ad server license, a data warehouse expansion, and integration cost against the site and app. The operating team floor is roughly ten: two ad ops, two engineering, two data science and measurement, two account management, one legal and privacy, one commercial lead. Anything below that headcount produces an RMN that cannot service more than a handful of endemic brand advertisers. By contrast, the buy path — a licensed platform from Criteo, Microsoft, Epsilon, or Koddi — starts around $500K first-year all-in and can be operated by three to four people plus the commercial lead. The platform vendor carries the ad server, the identity plumbing, and much of the measurement pipeline.

Four platforms cover mid-market retail scale.

The Criteo Commerce Media Platform is the most complete mid-market option: a demand-side platform for advertisers, a supply-side platform for retailers, and a monetization stack that runs onsite sponsored product, offsite audience extension, and open-web commerce placements against a shared identity substrate. Microsoft's PromoteIQ is the sponsored-product infrastructure that many retailers licensed when standing up their first onsite offering — the product has been rolled into Microsoft's broader retail advertising portfolio and remains a workable onsite-only anchor. CitrusAd, now inside Epsilon (Publicis), is the closest peer to Criteo on the retailer monetization side and often wins where a retailer wants a co-managed operating model. Koddi is the self-service advertiser platform of choice for retailers that want to control the operating model and expose a branded advertiser portal on their own front-end. For demand extension, Amazon Ads DSP is available to retailers with sufficient endemic inventory, and The Trade Desk's OpenPath appears where the retailer wants direct integration into an independent DSP without a supply-side intermediary. Molecule from LiveRamp, Cognitiv, and PubMatic Convert appear in the identity, prediction, and supply-path optimization layers rather than as end-to-end RMN platforms.

The identity graph is the piece a build cannot short-cut.

The identity graph inside an RMN has three surfaces: the loyalty ID tied to transaction history, the onsite and app device graph tied to browsing behavior, and the offsite graph the retailer needs to activate against display, video, and connected-TV inventory. Building the first is possible with the loyalty vendor and a customer data platform. Building the second is engineering-heavy but achievable. Building the third is not a mid-market build — it requires either LiveRamp, Experian, TransUnion, or Epsilon's identity spine, or a licensed platform that already resolves against one of them. The 2026 identity conversation is also a privacy conversation. Data clean rooms — Amazon Marketing Cloud, Google Ads Data Hub, LiveRamp Clean Room, Habu (now LiveRamp), Snowflake Data Clean Rooms, InfoSum — have become the required interface for advertiser-side audience collaboration and closed-loop measurement. A retailer standing up an RMN without a clean-room story cannot service most CPG advertisers by 2026.

Buyer-side. Supplier-paid. Buyers pay zero. Compensation has zero weight in the Cardinal Index scoring inside the Cardinal Method. Specific commercial terms live only in the private Decision Memo a buyer signs, never on this page.

Measurement in 2026 is incrementality, MMM, and closed-loop — not just ROAS.

The IAB Guidelines for Incremental Measurement in Commerce Media, published in 2025, are the standard mid-market retailers should be pricing their measurement stack against. Advertiser CPG teams expect three measurement outputs: incrementality (holdout-based lift on sales, not just clicks), a marketing-mix contribution that lets the RMN sit inside their aggregate media plan, and closed-loop attribution against a purchase event that ties the ad exposure to the loyalty transaction. Building this pipeline in-house requires a data science function that most mid-market retailers do not have. Buying it inside Criteo, CitrusAd, or Koddi comes with methodology defaults that the retailer's advertiser sales team can defend without hiring for the discipline. The one caveat: the IAB in-store measurement framework is still maturing, and in-store retail media (endcaps, shelf-edge digital, audio) is where the platforms differ most in 2026. If in-store is a meaningful share of the retailer's floor plan, the platform selection should be sequenced against the in-store roadmap, not the onsite-only offering.

Self-service portals are what advertisers will benchmark you against.

CPG media teams evaluate retailer RMNs on a small number of operational criteria: how quickly can a campaign be launched, how granular is the audience targeting, how transparent is the reporting, and how self-service is the workflow. Amazon Ads set the bar. Walmart Connect, Kroger Precision Marketing, Target Roundel, and Best Buy Ads set the mid-tier bar. A mid-market retailer's advertiser portal is being compared to those references, not to what the retailer's own tech team can build in eighteen months. The buy-side platforms ship with a portal that clears the mid-tier bar. A build path that does not budget for portal parity — reporting UI, self-service creative upload, taxonomy, campaign scheduling, audience builder — will not retain advertisers past the first renewal.

Sponsor-vs-endemic ad ratio economics decide the operating model.

The gross-margin economics of an RMN depend on the mix between endemic advertisers (CPG and category brands whose products the retailer already sells) and non-endemic advertisers (auto, finance, travel, entertainment). Endemic inventory is high-yield onsite sponsored product, sold on a self-service basis, at low incremental cost. Non-endemic inventory is offsite display and CTV extension, sold on a managed basis, with a demand-side buy-through fee. Most mid-market retailers plan for an 80/20 endemic-to-non-endemic mix in year one, moving toward 60/40 as the offsite offering matures. The platform choice reflects this: Criteo, CitrusAd, and Koddi cover both; PromoteIQ is onsite-endemic first; the DSP integrations sit on the non-endemic side. The economic implication is that the more the retailer needs non-endemic revenue to hit an EBITDA target, the more the platform decision favors an end-to-end vendor over an in-house build.

The 7-question retail media build-vs-buy framework

  1. Loyalty ID coverage. What percentage of the retailer's transactions are attached to a loyalty ID? Below 40 percent, the identity spine is not there and neither path will produce closed-loop measurement CPG advertisers will trust.
  2. In-house data science headcount. Does the retailer have two or more full-time data scientists who can own incrementality and MMM methodology? If not, the build path adds a permanent hiring commitment.
  3. Ecommerce share of revenue. If digital is under 15 percent, the onsite inventory alone will not support an in-house build. A licensed platform can still monetize what is there.
  4. In-store retail media roadmap. Are shelf-edge digital, in-store audio, or endcap media in the next 24-month capex plan? The platform decision should be sequenced against in-store, not onsite-first.
  5. Clean-room story. Which clean-room technology will host advertiser-side audience collaboration — LiveRamp, Habu (LiveRamp), Snowflake, InfoSum, or a platform-native option? A retailer without a clean-room story cannot service enterprise CPG.
  6. Non-endemic revenue target. What share of RMN revenue must come from non-endemic advertisers by year three? Above 30 percent, the demand-side extension is material and a platform with integrated DSP relationships wins.
  7. Advertiser portal reference. Which retailer's advertiser portal is the internal benchmark — Amazon, Walmart Connect, Kroger, Target? The build path must budget for portal parity against that reference, or advertisers will not renew.

Where the build path is defensible.

The build path is defensible in three narrow cases. First, a retailer with an existing ad-tech engineering function, typically inherited from a marketing services acquisition, where the incremental build cost is smaller than it looks from the outside. Second, a retailer whose category economics — pharmacy, home improvement, specialty grocery — produce enough sponsored-product yield to justify the operating team on onsite alone. Third, a retailer whose category concentration makes a platform vendor's shared identity graph a competitive risk. In every other case, the buy path with a phased operator handoff is the faster, cheaper, and lower-risk path to a working RMN in 2026. The build path is not wrong. It is just under-priced by mid-market retailers when they scope it against the six coupled systems that the IAB standards now formalize.

This is the fifth piece in the Retail Analyst Note series. The anchor is How mid-market retailers should source technology contracts in 2026. Every vendor named here is in The Cardinal Source's active supplier pool.

In short

  • An RMN is six coupled systems — ad server, demand layer, identity graph, clean room, measurement pipeline, advertiser portal — not one. Mid-market builders scope the first two and underestimate the last four.
  • The honest build floor is roughly $2M year-one plus a ten-person team. The buy floor is roughly $500K plus three to four operators.
  • Four platforms cover mid-market: Criteo Commerce Media, Microsoft PromoteIQ, CitrusAd (Epsilon), and Koddi. Amazon Ads DSP and The Trade Desk OpenPath appear on the demand side.
  • The identity graph is the piece a build cannot short-cut. Without LiveRamp, Experian, TransUnion, or Epsilon underneath, offsite activation does not work.
  • Measurement in 2026 is incrementality, MMM, and closed-loop against the IAB commerce media standards — not ROAS. Data clean rooms (LiveRamp, Snowflake, InfoSum) are now the interface with advertisers.
  • Advertiser portals are benchmarked against Amazon, Walmart Connect, Kroger, and Target. A build without portal parity does not retain advertisers past renewal.

Sources

All linked sources were live at time of publish (September 2026). Verify before quoting in a procurement document.

Standing up an RMN in 2026?

Run the Tier 1 benchmark.

Send us your loyalty coverage figure, ecommerce share of revenue, and current onsite ad server (if any). We return a benchmark PDF in five business days showing whether the buy path or the build path fits, which platforms match your scale, and where the identity and clean-room gaps sit. Free. No follow-up sales drip.

Run the Tier 1 benchmark →

Series  ·  Retail deep-dive · 8 pieces

Anchor: How mid-market retailers should source technology contracts in 2026. Coverage: Retail. Method: The Cardinal Method.