A retail media lead opens a vendor deck and finds a diagram of her own stack drawn back at her. An ad server here. A search engine there. A campaign tool, a measurement vendor, a spreadsheet holding the reporting together. The next slide replaces all of it with one box. It is a good pitch, because the diagram was accurate.

The simplicity being sold is real and worth wanting. What it costs is not on the invoice, and it does not come due this year. It comes due the first time you want to change one thing and discover that one thing is four things.

Worth being fair about who is in this conversation. Plenty of teams are not swinging back toward the bundle after trying something else. They never left. Retail media started as a two-vendor decision, the all-in-one model was the only model on offer. And nothing since has forced a reckoning. That is not a failure of judgement. It is what happens when a category changes faster than any one team's reason to re-examine it.

TL;DR

• Bundling gets sold as simplicity. What it actually does is recreate lock-in under a friendlier name, and the cost shows up as switching cost later rather than on this year's invoice.
• Three years ago retail media ad tech was a two-vendor category. It is now ten to fifteen specialized vendors, and the real choice has moved from vendor versus vendor to business model versus business model: bundled or decoupled, single-source or multi-partner demand.
• Bundled versus decoupled is not a features comparison. It is a question about who owns the decision layer and what it costs you to change your mind.
• The bundle also carries a pricing effect worth naming: when decisioning, demand and campaign management are sold as one thing, you are not able to price any of them separately or put any of them out to competitive tender.
• What decoupling protects is specific: you can replace the ad server, the search engine, the campaign interface, or a demand source, without rebuilding the other three.
• Simplicity and independence are not the same purchase. Both are legitimate. It is worth knowing which one you are signing.

‍

How did we end up with fifteen vendors?

Three years ago retail media ad tech was a two-vendor category. Today it is an ecosystem of ten to fifteen specialized platforms. And the competition has moved from vendor versus vendor to business model versus business model.

That fragmentation looks like chaos from the inside, and mostly it is not. 

It is what a maturing category does, and it is why the composable versus monolithic argument has become the live debate in this category rather than a theoretical one. Ad serving, ranking, campaign orchestration, demand connection, measurement and in-store were once sold as one product. That’s because nobody had built a better version of any single piece. Now people have, and the pieces have specialized.

The stakes rose alongside the fragmentation. 

The most advanced retail media networks earn 6 to 8 percent of their ecommerce revenue from media. Most retailers are stuck at 1 to 3 percent. 

That spread is wide enough that the structural decisions you make about your stack now have a number attached to them, which is new. When everyone was earning roughly the same, architecture was a preference. It is not a preference anymore.

‍

Why does "just use one vendor" sound so good right now?

Because managing six vendors is genuinely worse than managing one, and anyone who tells a retail media lead otherwise has not sat through the Thursday status call.

The fatigue is real. Six contracts with six renewal dates. Six support queues. Six versions of the same number, none of which reconcile. A quarterly business review where thirty minutes goes to explaining why the ad server's impression count and the measurement vendor's impression count differ by four percent. Nobody in that meeting is thinking about architectural optionality. They are thinking about whether anyone can just give them one number.

So when a vendor offers to collapse all of that into one contract, one interface and one number, the appeal is not naivety. It is a reasonable response to a real operational cost.

The reframe is not "resist simplicity." It is that simplicity comes in two kinds and they get sold together as though they were one. Operational simplicity is one interface for your team, one contract to administer, one invoice, one support relationship. Architectural simplicity is one vendor owning decisioning, demand and the front end. 

You can buy the first without the second. Most bundles require you to buy both, and only the second one is hard to undo.

‍

Is this a features question or a lock-in question?

Vendor evaluations in this category tend to run as feature comparisons, and feature comparisons favour bundles, because a bundle has more rows filled in. That is the wrong axis.

The axis that matters is which parts of your stack you can change independently later. 

A bundled system answers that with "the whole thing, at renewal." 

A decoupled one answers with "any single component, whenever." 

That is the substance of the difference between a decoupled stack and legacy ad serving, and it does not show up as a row on a feature grid.

There is a commercial version of the same point that deserves saying plainly, since it is the part CFOs notice last and care about most. 

When decisioning, demand access and campaign management are sold as one line item, none of them is individually priced. That means no piece of your retail media stack can be competitively tendered on its own, and the vendor's fees are not exposed to comparison at the component level.

It also means the fee typically scales with your ad revenue rather than with the work being done, so the better your media business performs, the more the arrangement costs you. Whether that is the right trade is a real question with a real answer either way. 

The problem is that a bundle prevents you from asking it component by component.

A useful way to hear this in a pitch: when a vendor says integration is included, ask whether that means "we integrate with anyone" or "everything you need is already inside our system, so integration is not a thing you have to think about." The second answer is more comfortable and describes a closed stack.

‍

What would we actually want to swap, and when?

This is where the abstract argument for decoupling either earns its place or does not, so it is worth being concrete about the scenarios. 

All four of these are ordinary, not hypothetical, and none requires anything to have gone wrong.

Your search and personalization provider changes. Retailers re-evaluate this category regularly, and the decision is usually driven by ecommerce rather than by media. 

In a decoupled stack, media is unaffected. The ranking layer sits above the search provider and keeps working. 

In a bundled one, where relevance logic lives inside the vendor's system or depends on their model, a search migration becomes a media project too, and now two teams are negotiating one timeline.

A demand source you want appears. Your brands start asking to buy your inventory from the tools they already use rather than from your campaign interface. If your decisioning layer can connect additional demand sources independently, that is an integration. If demand access is bundled with decisioning, your vendor's commercial interest in routing demand through their own pipes is now sitting between you and that budget.

You go beyond onsite. Offsite, in-store and connected TV each have real specialists, and the orchestration tool your brands want to work in may not be the one your ad server's vendor sells. A decoupled stack lets you choose the orchestration layer separately from the layer that decides what ranks. A bundled one makes that an all-or-nothing migration.

Something better ships. This category is producing meaningful new capability every few quarters. If adopting any one of those requires renegotiating your entire stack, and your stack ends up waiting on a vendor's roadmap, the practical answer is that you will not adopt it. And the cost of the bundle is the thing you quietly stopped evaluating.

None of these scenarios requires your current vendor to be bad at their job. That is the part worth sitting with. You can be perfectly happy with a vendor and still be constrained by the shape of the contract you signed with them.

‍

What does it cost to unwind an all-in-one two years in?

Ask a retailer who has done it and the answer is rarely about the software.

The advertiser migration is the real bill. Every brand running campaigns in the incumbent's interface has to be re-onboarded. New logins, new workflows, historical campaign structures rebuilt, account teams retrained. Done all at once, this is the part that overwhelms support lines and strains brand relationships. And it is mostly invisible in a vendor evaluation because it is your cost, not theirs. This is also the strongest argument for treating the campaign stack and the ranking layer as two separate projects, which DocMorris ran deliberately as a sequence of independently reversible steps rather than one migration.

Reporting history breaks. Two years of performance data sits in the vendor's schema, measured their way. Whatever you export rarely reconciles cleanly with what comes next. This means year-over-year comparisons get an asterisk for a full cycle and every brand conversation starts with a caveat.

Relevance and merchandising rules get rebuilt. Category rules, brand exclusions, blocked terms, seasonal overrides. These accumulate quietly over two years, they are rarely documented anywhere outside the system they live in, and they are the reason a migration that looked clean in the plan produces a strange-looking results page in week one.

Everything renews at once. This is the underrated one. A bundle means a single renewal date for decisioning, demand and the campaign interface. You do not get to renegotiate the piece you are unhappy with while keeping the pieces you like. You renegotiate all of it, at the same time, with the vendor knowing exactly how hard it would be for you to leave.

That last item is the tax, stated plainly. It is not a line on an invoice. It is the leverage you gave up, and it is charged at renewal.

‍

Three questions to answer before your next renewal

#1 Which single component of our retail media stack could we replace next quarter without touching the other three?

#2 If our ecommerce team changed our search provider next year, how much of our media setup would have to be rebuilt?

#3 Do we know what each piece of our current bundle would cost if it were priced separately?

‍

Key Takeaways

• Retail media outgrew the two-vendor market, and the real decision is now bundled versus decoupled rather than one vendor versus another.
• Operational simplicity, meaning one interface and one contract for your team, is worth buying. Architectural simplicity, meaning one vendor owning decisioning, demand and the front end, is what costs you later. They usually get sold as one thing.
• A bundle prevents component-level pricing and competitive tender on any single piece of the stack, which is a commercial effect independent of whether the total fee looks reasonable today.
• The scenarios that make decoupling pay off are ordinary: a search provider change, a new demand source, an omnichannel expansion, a better component shipping. None of them requires your incumbent to have failed.
• The cost of unwinding a bundle is mostly your cost, not the vendor's: advertiser re-onboarding, broken reporting history, undocumented merchandising rules, and a single renewal date that removes your leverage.
• Ask which one you are buying. Both answers are legitimate. Only one of them is easy to change your mind about.

‍

Frequently Asked Questions

Why not just use an all-in-one retail media platform? Because it answers an operational problem with an architectural commitment. One interface and one contract genuinely reduce overhead, and that benefit is real. The trade is that decisioning, demand access and campaign management become a single unit you can only change together, at renewal. If your stack is stable and you expect it to stay that way, that can be the right call. If you expect to change your search provider, add a demand source, or expand beyond onsite in the next three years, the bundle prices those changes higher than they need to be.

What is the actual switching cost of leaving an all-in-one vendor? Most of it is not software. It is re-onboarding every advertiser into a new interface, losing clean year-over-year reporting because historical data sits in the old vendor's schema, and rebuilding merchandising and relevance rules that were never documented outside the system. The structural cost on top of that is timing. A bundle renews as one contract, so you cannot renegotiate the weakest component while keeping the rest.

Does decoupled mean more vendors to manage? Usually yes, and that is the honest trade-off rather than a footnote. A decoupled stack adds at least one relationship. What it buys is that no single one of those relationships controls the others. Some of the operational overhead can be recovered by consolidating the interface layer separately from the decisioning layer. So your team still works in one place even though the underlying components are independently replaceable.

Is a bundled retail media stack ever the right choice? Yes, for a retailer committed to their current vendor's commercial model and roadmap for the long term, and without near-term plans to change search providers, add outside demand, or expand into offsite and in-store. The mistake is not choosing a bundle. It is choosing one without pricing what it costs to change your mind.

How do we evaluate bundled and decoupled vendors side by side when they are not the same shape? Stop comparing feature grids, since a bundle will always fill in more rows. Compare on replaceability instead. For each component in the proposal, ask what it would take to swap that one piece in two years, and require the answer in contract language. That single question produces more separation between vendors than any feature matrix, and it is the one a bundled proposal is structurally least able to answer well.

‍