Retail Media Network Platforms: Can You Prove the Sponsored Click Actually Sold Something?

Why a retail media network breaks the commerce platform underneath it

Picture a Tuesday call with a supplier who bought $40,000 of onsite sponsored placements for a summer push. They want three things. Why 60 percent of the budget spent in the first nine days. Why their product showed against a competitor brand term they never bid on. And what those clicks actually sold. Your answer to the first two is a CSV export. Your answer to the third is a click chart from the ad tool, which the supplier's own sell-through report flatly contradicts. The call ends with a make-good and a much quieter renewal conversation.

That gap is the entire category. Retail media is high margin because you are selling attention you already paid for, but what the supplier is actually buying is proof. They need a defensible line in a trade budget that a finance team will sign off. Every component exists to produce that line: the auction that picks which product shows, the pacing that spends evenly, the creative that renders inside your own page templates, and the attribution join that ties an exposure to a basket in your own transaction log.

Shopify, Salesforce Commerce Cloud and commercetools have no concept of a sponsored slot with a bid, a budget, a pacing curve and an advertiser owner. They have a merchandising rule engine. The distance between those two things is a platform, and it is the reason your head of retail media currently spends Monday reconciling spend by hand.

The auction is a ranking problem inside your own search, not an ad server bolted on top

The common first mistake is treating a sponsored placement as an ad unit dropped onto a results page. A sponsored product only earns money twice, once from the supplier and once from the basket, and it only does that when it is relevant. Relevance lives in your catalogue, your search index, your availability data and your margin structure. So the auction has to run inside the ranking pipeline, not beside it.

Concretely, a bid is not a price, it is one input to a score. The useful score is bid multiplied by predicted click probability, and that prediction needs the same features your organic ranker already has: the query, product attributes, historical conversion for that query and product pair, and availability at the fulfilment node serving this specific shopper. Serve a sponsored slot for something out of stock at the store the shopper selected and you have taken the supplier's money while degrading your own conversion in a single request.

CitrusAd and Criteo Retail Media both integrate into a retailer's search and do it competently for a fairly standard grocery site. They strain when your ranking carries genuinely retailer-specific rules: own-brand protection, category exclusivity you sold in a trading meeting, store-level rather than site-level availability, or a promotions engine already applying multibuys to the same page. Then sponsored and organic decisions are made by two systems that cannot see each other, and the visible symptom is a sponsored unit contradicting the promotion banner directly above it. Topsort sits architecturally closer to the retailer's own surfaces, which helps, but no vendor can encode a commercial commitment that exists only in a signed joint business plan.

What a custom build does: one ranking call returns the blended page. Sponsored and organic candidates score in the same pass, business rules apply once, and slot policy is yours. How many slots, where they sit, first price or second price, reserve price per category, whether a competitor may bid on your own-label terms. Those are commercial decisions and they change quarterly. In a vendor platform each one is a support ticket.

Pacing and budget control are where the make-goods come from

Every make-good you have issued traces back to the same failure. A campaign with a flat daily cap front-loads into the highest traffic hours, exhausts on a Saturday morning, and then the supplier's brand is absent for the promotion week they actually cared about. Or the reverse: budget underspends, the campaign closes at 70 percent delivery, and you refund.

Vendor platforms pace against their own traffic model. Yours is different, because your traffic has a Thursday grocery shop peak, a payday spike, a weather sensitivity in some categories, and a completely different curve on the app versus the site. Pacing that does not know your curve will always over-deliver early.

What a custom build does: pacing runs against a forecast built from your own hourly impression history per surface and per category, with a target delivery curve the trader can shape. A campaign booked for a two-week promotion gets a curve weighted to the promotion mechanic, not a flat line. Add a guaranteed-delivery mode for the large annual commitments your commercial team sells, because those are underwritten differently from auction spend and cannot share a pacing model with them. Then surface pacing health to the trader daily so a shortfall is a conversation in week one, not a credit note in week three.

Closed loop attribution is the product, and it is a data problem before it is an ad problem

Attribution is where the money is defended, and it is precisely where you cannot hand the work to an outside platform, because the join requires loyalty identity and transaction detail that should not leave your business. That is not a preference, it is what your privacy counsel will tell you the first time a supplier asks for shopper-level export.

The join itself is unglamorous. An exposure event carries a household or loyalty identifier and a timestamp. A transaction carries the same identifier, a basket and a channel. You define the window, and you define it per category, because a shopper exposed to a sponsored ad for a soft drink converts in a different window than one exposed to a nappy multipack. Then in-store redemption has to fold in, which is the part that decides whether suppliers take you seriously, because for most grocers the majority of the sale still happens in a store and an online-only attribution number understates your value.

The genuinely defensible version adds incrementality. Hold out a matched control audience, report lift rather than last-click, and publish the methodology. That is what turns a media line into a trade line in the supplier's budget. Deliver it through a clean room pattern where the supplier queries aggregates with a minimum cohort size and never receives raw rows. Koddi is strong on the measurement and reporting side and is a fair choice if measurement is your only gap, but it still needs your data to move to reach it, and the retailers who win here treat their transaction log as the asset it is.

Self service, trading terms and actually getting paid

The billing question kills more of these programmes than the ad tech does. Media spend from a supplier is not a credit card transaction. It is negotiated against existing trading terms, often netted against other supplier income, invoiced by your AR team on agreed payment days, and sometimes funded from a promotional allowance rather than a media budget. A standalone ad platform with its own billing produces a second invoice stream your finance team has to reconcile by hand, and it will not net.

What a custom build does: campaigns carry the supplier's existing vendor number, spend accrues to a media revenue account, and the invoice is generated through the same AR process as every other supplier charge, with the credit limit and payment terms already on file. Self service sits on top for the long tail of smaller suppliers, with an approval workflow so a trader can gate creative and keyword bids on sensitive categories. Expect roughly the top 20 suppliers to keep buying through your commercial team on an insertion order regardless of what you build, and design for both paths from day one rather than pretending everyone will self serve.

What this costs and how long it takes

Across the 2,000-plus projects Digital Heroes has delivered, this is the honest shape. A first release covering the in-search auction, pacing, creative rendering in your templates and a supplier-facing reporting view runs $90,000 to $180,000 and ships in 14 to 20 weeks. That is a system your traders sell against, not a demo. A full platform adding supplier self service with approvals, offsite audience extension, clean room reporting, incrementality holdouts and AR billing integration runs $250,000 to $650,000 phased over 9 to 15 months.

What pushes cost up: the number of distinct surfaces, because search results, category pages, product detail, app home and email each need their own slot logic and their own pacing history. Store-level availability, which turns a single stock check into a per-node one. Existing loyalty identity resolution quality, because if your online and in-store identities do not reliably join, attribution work becomes an identity project first. And auction latency budget, since a sponsored decision that adds 80 milliseconds to search is a conversion cost that can exceed the media revenue.

What keeps cost down: launching on search results only, with a fixed three-slot layout, auction plus pacing plus reporting, and nothing else. That covers most of the revenue and teaches you your own numbers.

When buying beats building

Buy if your media revenue is under roughly $5M a year, if you have fewer than about 40 active supplier advertisers, or if your ecommerce site is a small share of total sales and no supplier is asking hard measurement questions yet. CitrusAd, Criteo Retail Media and Topsort will all get you live faster than a build and the revenue share is cheaper than an engineering team at that scale.

Build when two or more of these are true. Your suppliers are challenging your attribution numbers and you cannot answer with your own data. Your commercial commitments require ranking rules a vendor cannot express. Your in-store sales are the majority and online-only measurement is costing you rate. Your finance team is reconciling a second billing system by hand. Or the revenue share you are paying a vendor has quietly become larger than the fully loaded cost of a team. That last one arrives faster than people expect, because revenue share scales with your success and an engineering cost does not.

How to choose a developer for a retail media build

Ask them to describe the request path for a search page with sponsored slots, including where the auction runs and what the added latency is. If they describe a separate ad call made client side after the page renders, they are building an ad widget and your conversion will pay for it.

Ask how they handle attribution identity across online and in store, and what happens for a shopper with no loyalty ID. A credible answer includes household-level matching, a stated coverage rate, and honesty that unmatched traffic is reported separately rather than silently dropped.

Ask what they have integrated on the billing side. Netting media revenue against supplier trading income through an existing ERP (Enterprise Resource Planning) is a different problem from charging a card, and the team that has never touched an AR ledger will discover this in month four.

Ask who owns the code, the models and the data pipelines, and get it in writing before kickoff. Your transaction log and your attribution methodology are the durable asset here. At Digital Heroes the client owns the repository from the first commit, and we would tell you to walk away from any developer who hedges on that.