The commerce media stack

I’m asked a lot about which tech vendors make up the commerce media stack, so I lay out a bit about what I’ve heard.

The Vendor Interview: FatTail

Marketecture Vendor interviews are free for a week, then require a subscription. Read more about subscription options at marketecture.tv.

Podcast: Mark Mannino on why Omnicom bought Flywheel for $825 million

The commerce stack

This week’s pod featured Mark Mannino formally the general manager of Flywheel, an e-commerce agency helping brands to optimize their commerce presence. This got me thinking about the nature of the commerce media stack and how buyers and sellers are interacting in this fast growing area.

The sell-side commerce stack

Building a sell-side commerce stack is significantly more complicated than monetizing traditional media properties. Off-the-shelf ad servers don't cover all the use cases and don't fit well with the native formats and SKU level targeting common to commerce media.

There are many components to the commerce stack, and not every company needs to invest in all of them:

1. Traditional ad serving

2. PLA marketplace

3. Data warehouse

4. The buyer portal / UI

5. Audience extension

For non-endemic ads on commerce sites many sellers choose a traditional ad server, like Google’s GAM. There’s no real reason to reinvent the wheel when monetizing banners, videos, and the like and targeting based primarily on first party data and context. And unlike SKU-level ads, programmatic demand can be a part of the revenue mix.

Serving Product Listing Ads (“PLA”s), on the other hand, is a problem that is not well suited for traditional ad stacks and differs in almost all respects:

• Creatives are always native

• There may be tens of thousands of creatives, and they need to be updated by the advertisers using APIs or feeds

• Targeting is either keyword or SKU, very rarely user-based

• Pricing is CPC or CPA

Many of the larger players in commerce media (e.g. Uber) have rolled their own systems for serving these ads. Kevel is a SaaS option for building a highly customizable ad serving stack that supports commerce media. Building your own is probably a good decision if most of the buyers in your network are unique to your platform.

Criteo, CitrusAd, and Microsoft’s PromoteIQ are vendors that specialize in offerings a PLA solution, along with in a network of marketer demand. If you are in a more common retail segment, like grocery, it probably helps to have a lot of demand lined up from your competitors. Since Citrus was acquired by Publicis, they have also expanded into audience extension with an Epsilon-powered solution.

Commerce media would be nothing without the data powering it. Retailers have a huge data advantage over the marketers who’s wares they sell, and controlling access to that data is what drives a lot of this business. Retailers need to get their data into flexible and actionable data warehouses to power on-site targeting, attribution, and audience extension. This work is less about the technology choices and more about making the data easily accessible in the programmatic environment.

It is then up to each company to decide how much transparency to offer their advertisers. Data used for targeting of onsite ads would generally be used in raw form. But reporting data on conversions and purchases might require some level of obfuscation to avoid both competitive data leakage, and privacy concerns. You may thus be forced into adding clean room technology into the mix to closely control granularity of data flows. For example, if you allow a marketer to upload a hashed email list, you may not also want to give that marketer back a report showing detailed purchase histories.

Audience Extension

Retailers can make their data available offsite to marketers through audience extension, and these partnerships often provides splashy headlines. Walmart has a well-publicized partnership with The Trade Desk to power the “Walmart DSP”. Target’s Roundel took a different approach and mostly makes its data available in various programmatic spots rather than trying to force you to use their DSP. And Amazon, of course, runs it’s own DSP, which is probably the third largest by volume.

For retailers there are a couple of considerations in the audience extension strategy:

• Should you sell data or media, or both? Roundel is an example of a data-only strategy, which is a lot easier operationally, but maybe captures less value.

• Should you go exclusive with a single DSP, or multiple? Kroger, for example, just launched on Yahoo, even when they were previously available on TTD.

• Should you also allow your data to be used for attribution in these platforms, or just for targeting?

The interesting wrinkle about off-site commerce media is the question of how signal loss and cookie deletion will impact this revenue stream. Retailer data has identity baked in, but the ads still need to appear on media that can be matched to that identity. For the most part retailers are relying on the platforms to figure this out.

Buy-side stacks

As a marketer with commerce relationships, the ability to utilize purchase and behavioral data is a godsend. But management of these relationships in many different UIs can be cumbersome.

The first consideration is how many retailers matter to your brand. While there are hundreds of retailers, and tens of thousands of marketers, the overlap between them is not smooth or regular. For non-CPG brands — say appliances or sporting goods — there may be very few retailers that matter to you.

The default option for marketers, to be bluntly honest, is to just buy ads on Amazon. eMarketer estimates that 75.2% of all retail media is just on Amazon. But Amazon’s tools have lagged in capabilities.

The leading independent vendors for working across commerce channels are Skai (formerly Kenshoo), Microsoft’s PromoteIQ, and Criteo’s Commerce Max. Skai boasts 100s of partners, including both Criteo’s marketplace and Amazon. Criteo’s Commerce Max product is a relative newcomer, having launched in 2023. Criteo has deep experience in working with SKU-level data and bringing in commerce feeds, but doesn’t (to the best of my knowledge) have an Amazon integration.

Of course, buying media across disparate closed channels makes the measurement problem more difficult. Each platform will be happy to attribute it’s success to the advertising it ran, which, in the case of PLAs might be justified. But what about all the search and social spend that drove the consumers to the retailer in the first place?

Retail and commerce media remains one of fastest growing and most interesting parts of advertising. The technology is evolving fast and new entrants could still shake things up. As with every thing, make sure your technology choices fit your business!

Reading list

• Roku to support UID2

• Google is a monopoly in search. AdExchanger article. Ari’s Monopoly Report summary.

• Uber on a billion dollar run rate, as they predicted (link)

• Nielsen and Innovid in big partnerships for cross-screen measurement (link)

• X (Twitter) sues GARM and WFA alleging an anticompetitive collusion to prevent advertisers from giving budgets to the social site.

• GARM suspends operations in light of lawsuit.

• Ziff Davis buys CNET for $100 million (link)

• Adalytics finds some really dirty stuff being supported by ads, with the appearance of DV and IAS tags.

• DV responds saying the methodology doesn’t prove anything.

• Ari does his first TikTok on this issue (really)