Buying onsite retail media technology should not be this hard.

Yet for many RMN buyers, it is. The pitches sound remarkably similar. Every vendor claims cutting-edge AI. Every deck promises relevance, personalization, and incremental revenue. And because the underlying technology is difficult to inspect, buyers often default to two imperfect signals: sales narrative and price.

The assumption is simple — higher price reflects superior AI. That assumption is no longer valid.

‍

The Infrastructure Has Changed

New AI and machine learning infrastructure have fundamentally altered the economics of retail media technology. Capabilities that once required proprietary systems built and maintained at significant cost are now accessible at a fraction of the price. The question is not whether your vendor has built sophisticated AI. The question is where that AI comes from — and what it is actually optimizing.

That distinction matters more than most buyers realize.

‍

Two Architectures. One Clear Difference.

The first approach — the traditional model — is vendor-owned AI that optimizes ad revenue in isolation. This is the model Criteo, Koddi, and Moloco follow today. There are meaningful differences between them — some rely on relatively simple algorithms, others are advancing their semantic understanding, and Moloco has made significant and well-documented investments in machine learning infrastructure, with AI credibility that is widely recognized in the industry. But they share a common structural limitation: the AI operates in a silo, optimizing sponsored results without regard for the full, unified page experience.

You can test the algorithm quality of any vendor yourself. Search for "vegan protein bar" on a retailer running traditional ad tech: if whey-based products appear in the sponsored slots, the system lacks semantic understanding. Search for "footwear" and you may find no shoes in the sponsored results at all, despite active campaigns — the query and the inventory simply were not connected.

The second approach — the unified model — leverages existing search and personalization infrastructure to optimize the complete page — across both retail and media revenue streams simultaneously.

This is the architecture Pentaleap pioneered. It is the foundation for how The Home Depot, Macy's, and CVS run retail media today, and it reflects the direction that sophisticated ecommerce players — including Zalando and ABOUT YOU — have already moved toward.  

Criteo is now building a similar capability, which it currently refers to as “Holistic Page Optimization.” That will likely accelerate broader adoption.

‍

Why the Traditional Model Is Redundant by Design

The legacy approach carries a structural cost that is rarely discussed openly.

Vendor-owned ad AI and a retailer's existing search and personalization systems are solving the same problems — independently, in parallel, at full cost. The overlap is not incidental. It is architectural.

Retailers are effectively funding two parallel AI stacks to solve the same problem on the same page. The unified approach eliminates that redundancy by design — not as a feature, but as the founding premise.

Take Macy's as an example. They use Google Vertex AI for Retail Search to power onsite search and personalization. Pentaleap's optimization layer integrates with Vertex, loads the full page context, applies bid signals to surface sponsored products, and renders a unified page across paid and organic results. No duplication. No siloed optimization.

‍

‍

Which Vendor Will Yield the Best Results?

That depends on what you are actually trying to achieve.

If your goal is to optimize retail and media as separate businesses — with a page divided into fixed paid and organic zones — the traditional approach remains a valid path. In practice, few retailers consciously choose this. More often, it is the path of least resistance — organizational silos, existing vendor contracts, or concern about disrupting ad revenue that is already flowing. But running two systems in parallel has a cost beyond the technology bill: when ad AI and organic ranking operate independently, they can actively work against each other, creating a fragmented experience for the shopper and leaving performance on the table for the retailer.

Choosing between Criteo, Koddi, Moloco, and Pentaleap's ad serving technology comes down to two things: the quality of the AI, and the price you pay. Criteo, Koddi, and Moloco have all built their own proprietary AI and operate in a broadly similar pricing range. Pentaleap operates on a more efficient infrastructure model, a cost advantage we pass directly to our customers. My recommendation: scrutinize the technology carefully, and put less weight on the pitch.

Disclaimer: Pentaleap is best known for unified ranking — but the technology is flexible and works very well within fixed tile constraints. Even in that mode, we still leverage search and personalization AI to serve more relevant sponsored results than a traditional ad server would. In practice, most of our customers take a step-by-step approach: starting with fixed tiles, proving the results, and progressively expanding unified ranking category by category. I expect the same will apply to Criteo's new product once it is officially released.

If your goal is to bring retail and media closer together — to optimize total company performance rather than two separate P&Ls — the unified ranking approach will deliver better outcomes. By scoring sponsored and organic products within a single system, you drive better results on both sides simultaneously. Relevance improves. Ad performance improves. And the page works harder as a whole.

The key decision here is architectural.  

Criteo's Holistic Page Optimization comes bundled with their ad server and demand offering — a more convenient, integrated package. Pentaleap operates as an independent optimization layer, giving you the flexibility to work with multiple ad servers and any campaign management interface you choose.  

Which architecture you choose also has meaningful price implications. As a startup and a disruptor in this space — without the margin pressures of a publicly listed company — we pass the efficiency gains of the new infrastructure directly on to our customers.

The best AI in retail media no longer has to be the most expensive.

‍

Make the Market Work for You

Retail media technology is not a transparent market. Vendors control the narrative, benchmarks are rarely independent, and it is genuinely difficult to separate signal from pitch. Given the strategic and financial weight of this decision, we believe buyers deserve better than that.

Start by asking the right questions:

- Where does your AI come from? Is it proprietary, or does it leverage existing search and personalization infrastructure?

- What is your AI actually optimizing? Ad revenue only, or total page performance across retail and media?

- What is your pricing model, and what fees do you charge? Retailer-side only, or also brand-side? Are there API partner fees?

Then let the numbers do the talking. Ask vendors to prove their results on your inventory, against your KPIs, in a controlled test. The technology exists to run those comparisons directly — and a live test will tell you more than any sales deck.

As a buyer, you have more leverage than you might think. A new generation of infrastructure has made high-quality retail media AI more accessible and more affordable. For retailers willing to ask the right questions and let results guide the decision, that is a significant opportunity.

The best outcomes in retail media will go to the buyers who demand evidence — and act on it.

‍