Third in a three-part series on agentic commerce readiness from Suhas Gudihal, CTO of Athos Commerce. Part 1: Your Data Is Your Storefront | Part 2: AI Agents Are Already Skipping Your Products

I asked a merchant last year to tell me exactly what they wanted from AI. The answer came without hesitation: “I don’t know, but there is a lot of hype. If I don’t have it, will I lose my users?”

That conversation wasn’t unusual. In the first two posts in this series, we explored why product data has become the new storefront and the specific data shortfalls that cause AI shopping agents to skip your products. But the more I talk with merchants, the more I realize the problem isn’t just in their product feeds. It’s in how they’re framing the problem.

Most of the ecommerce leaders I speak with know that something fundamental is shifting. They see the headlines about AI shopping agents, hear about Google’s [Universal Commerce Protocol], and feel the pressure to respond. But they keep asking “Do you have AI?” when the shift they’re facing isn’t really about AI as a feature. It’s about whether their data architecture is built for a world where machines, not humans, are the primary evaluator of their products.

This has happened twice before in my career, in completely different industries, and both times, the organizations that asked the right question early were the ones that grew through the shift.

What a food labeling mandate taught me about data readiness

In my first company, we grew 200–300% in a single year. We hadn’t predicted a market shift. Our data infrastructure just happened to be built for one.

We made food distribution software. Our tagline was “from the farm to the fork,” and the platform tracked ingredients across every step of the supply chain, from the producer through distribution to the retail shelf. We built traceability into the system because our customers needed it for their daily operations. If a product needed to be pulled from shelves, they had to know exactly where every ingredient came from and where it went. It was a practical requirement, not a strategic bet on regulation.

Then 9/11 changed the landscape. Country of Origin Labeling became mandatory, and every food distributor in the United States suddenly needed to trace ingredients from source to shelf. The large enterprise software vendors scrambled to retrofit their ERPs for traceability. They had the resources and the engineering teams. But their systems were built around human reporting workflows, not machine-readable traceability queries, and rebuilding that foundation under regulatory pressure took time they didn’t have.

Our system already captured that data. We didn’t need to rebuild anything. When the mandate arrived, we were ready, and distributors who had been evaluating us as a niche player started signing contracts within weeks. The companies that lost ground weren’t slow to respond. They were structured wrong for the question the market was now asking.

Then it happened again, in a completely different industry.

What cancer diagnosis taught me about structured data

Human pathologists diagnose cancer correctly 66% of the time. The FDA recognized this decades ago and began requiring a second independent pathologist to review every diagnosis—two experts, two interpretations, and then a judgment call.

My second company took a different approach. We scanned tissue samples using a digital scanner at 40x magnification and applied machine learning to distinguish healthy from cancerous cells. The system staged the disease in two minutes, a process that previously took pathologists three to four weeks.

The FDA required 2 million verification tests before granting approval. Our system achieved 97% accuracy. It is now deployed across nearly all 46 major cancer centers in the United States, and the James Cancer Center in Columbus used it to enable remote second opinions from clinicians in Saudi Arabia and South Korea, increasing their revenue by 25–30%.

The algorithm wasn’t more sophisticated than what the pathologists knew. The advantage came from consuming clean, normalized data. Digital scans with consistent resolution and magnification provided the machine with a reliable foundation for pattern recognition. A pathologist working from variable-quality slides under a microscope had noisier inputs and produced less consistent results.

AI shopping agents operate on the same principle. They aren’t smarter than a human merchandiser who knows the catalog inside and out. But they are faster and more consistent when the data they consume is structured for machine evaluation. An unstructured product description gives an agent the same problem a blurry slide gave a pathologist: not enough reliable signal to make a confident recommendation.

The pathologists who resisted the transition eventually adopted the system because the diagnostic accuracy was undeniable. Merchants who resist restructuring their product data will face similar pressure, but the feedback will manifest as lost visibility and revenue rather than as a clinical error.

Who tells your brand story when the agent does the shopping?

In my experience across food distribution, cancer diagnosis, and now ecommerce, every system that performs well under machine evaluation shares the same structural property. It separates structured truth from human narrative into two distinct layers. I call this the two-plane architecture.

• Agent consumption layer. Structured truth. Canonical, normalized product attributes and fulfillment rules. Materials, dimensions, shipping constraints, return policies, and inventory status. This is what AI shopping agents evaluate when deciding whether to recommend your products. Every attribute needs to be machine-readable, unambiguous, and verifiable.
• Brand expression layer. Human narrative. Positioning, emotional resonance, visual storytelling, and the brand voice that creates preference when shoppers land on your site. This layer is essential for converting the humans who still browse, compare, and buy based on how a brand makes them feel.

These two planes cannot be mixed in the same feed or API. Marketing language embedded in your product feed makes it harder for agents to extract clean attributes. But stripping brand voice from your product pages to make them machine-readable costs you the emotional connection that drives conversion with human shoppers. A single data layer forced to serve both audiences degrades both experiences.

This is the same separation that made food traceability data reliable for regulators and digital pathology data reliable for the FDA. Structured truth and human narrative serve different audiences. Combining them creates compromise, not efficiency.

Where should ecommerce leaders start?

When merchants ask me what they should do about AI, I give them three steps, and none of them involve buying an AI feature.

Separate your brand content from your product feed data. Pull up your Google Shopping feed and your product detail pages side by side. For every field in the feed, ask one question: does this help a machine evaluate my product, or does it persuade a human to buy? Fields like materials, dimensions, shipping speed, and return policies belong in the agent consumption layer. Brand messaging, lifestyle imagery, and emotional positioning belong in the brand expression layer. If you find marketing language mixed into your structured feed fields, that’s the first thing to fix.

Treat your feed as infrastructure, not an export. Most merchants treat their product feed as something they generate and send to Google or Amazon once a week. In agentic commerce, the feed is your agent-facing operating system. It needs the same investment cadence, ownership, and quality standards you give your website or your ERP. Agents require continuous, real-time data, and any delay in updating your feed reduces both your trust score and your ranking.

Start before the mandate arrives. In food distribution, the mandate was Country-of-Origin Labeling. In cancer diagnosis, the FDA approved digital pathology. In ecommerce, the mandate is already here. AI shopping agents are evaluating your product data today, and merchants with well-organized feeds are capturing visibility that compounds every week while competitors with incomplete catalogs fall further behind.

Before the food-labeling mandate took effect, I didn’t know it was coming. But because our data was already structured for traceability, we were ready when it hit, and we grew while our competitors scrambled. The ecommerce equivalent of that mandate is already underway. Every week that your product data stays unstructured is a week that competitors with cleaner feeds widen their lead. The merchants who act now will look back on this period the same way I look back on the months before Country of Origin Labeling. The window was open, and the early movers captured growth that the latecomers never recovered.

If you want to see what your product feed looks like through an agent’s eyes, my team at Athos Commerce can run a feed audit that shows you exactly what’s missing.

Request a personalized feed audit.

Watch our webinar on agentic commerce readiness.