Joe Pizarro is an eCommerce executive and AI strategist specializing in digital transformation, catalog intelligence, AI systems, and business growth.

Kato is a governed product intelligence layer for supplements and other complex health products. It sits directly between what a shopper actually means and what a commerce catalog contains, so recommendations rest on reviewed evidence instead of marketing copy and product titles.

How does Kato's three-pillar architecture work?

Kato separates three concerns that general-purpose systems usually merge: governed knowledge (structured facts about ingredients, health claims, and safety), universal product intelligence (what the market offers, read through connections such as Shopify partner APIs), and merchant intelligence (a single operator's private information). The relationship runs one way, so Kato's universal understanding never depends on any one merchant's private operational data.

Why does Kato understand far more products than it recommends?

Understanding a product and earning permission to recommend it are two separate gates. In testing across real Shopify catalogs, Kato reached 73.7% product-level understanding but only about 9.1% verified governed fit for a specific shopper need. That gap is the architecture working as designed, because recognizing a bottle of vitamins is not the same as proving it works or responsibly telling someone to take it.

What does 'nothing autopublishes' mean?

New knowledge only enters Kato through a reviewer-gated process that requires evidence quality, clear provenance, safety context, and human approval. Nothing is added by scraping the web and nothing publishes itself, so Kato can distinguish what it recognizes, what the evidence supports, what it can safely conclude, and what remains uncertain.

How does Kato handle hard constraints like 'no melatonin'?

Kato's language intelligence converts natural shopper language, such as wanting something for sleep without melatonin, into structured intent, preferences, and non-negotiable constraints. On its strict testing benchmarks Kato recorded 0% hard constraint violations, escalated to safety protocols 100% of the time when required, and chose to abstain rather than guess in 96% of the cases where it should.

What is agentic commerce, and why does it need an intelligence layer?

Agentic commerce is shopping carried out by AI agents. In the emerging stack, shopper intent goes to an agent that checks a catalog and completes a transaction. In high-stakes categories like health and wellness that stack needs one more step, with intent routed through governed product intelligence before it ever reaches the catalog, so agents can recommend with a defensible basis rather than on commercial signals alone.

Where does Kato stand today?

The proof of concept is running: 51 core canonical ingredients, 65 ingredient forms, 63 strictly published governed statements, and 548 benchmarked Shopify catalog product cards. Kato measures not only how often it answers, but whether its decisions are better and safer, with every decision built to be explainable and traceable.

Back to Studio

Video · Kato

Why AI Recommends Without Understanding — An Intro to Kato

What AI shopping systems actually base their recommendations on, why accessible product data is not the same as understanding, and how Kato's three-pillar architecture separates what a product is, what the evidence supports, and what can be responsibly recommended.

Full transcript

Click any timestamp to play from that point.

Why an AI's recommendation is shallow

Have you ever actually stopped to think about what happens when an AI recommends, say, a sleep formula or maybe a prenatal vitamin? Like, what is that AI's decision actually based on?

Well, for most systems right now, the answer is surprisingly shallow. They're literally just reading the exact same marketing copy, product titles, and reviews that you and I see.

But here's the thing. When we're talking about our health, or the health of our pets and babies, taking marketing claims at absolute face value just isn't safe, and honestly, it isn't smart.

So, welcome to today's explainer. Today, we're diving deep into the architecture behind Kato Intelligence.

We're going to unpack exactly how this specialized system is being built to bridge that massive gap between what a shopper actually intends to buy and the wild raw data of massive product catalogs.

We are looking at a foundational shift in how AI actually understands complex products.

All right, here is our road map for today. One, the AI commerce problem. Two, the missing intelligence layer. Three, Kato's three-pillar architecture. Four, understanding versus recommending. And finally, five, the agentic commerce future. Let's jump right in.

The AI commerce problem

Section one, the AI commerce problem, or why accessibility definitely does not equal understanding.

Okay, so currently systems like Shopify have done an absolutely incredible job building infrastructure that makes the entire product universe accessible to machines.

But in high-consideration categories, think wellness and supplements, these commercial signals just describe a product. They really only tell us what the merchant says the product is.

They absolutely do not establish what the scientific evidence actually supports. They don't check if a shopper's strict dietary constraints are being met, or frankly if there's even enough valid information to recommend the product in the first place.

And that brings us to a crucial paradigm shift. Traditional search and recommendation systems basically ask, hey, what products match these words, or what's this person likely to buy?

But Kato asks a totally different set of questions. It asks, given what this person actually wants, what are these products? What meaningfully differentiates them? What does the evidence actually support? And what can we responsibly conclude? It's just a completely different bar for success.

The missing intelligence layer

Moving on to section two, the missing intelligence layer. Introducing Kato.

Because of that massive gap we just talked about, there's this missing layer between what the shopper intends and what the catalog actually holds. Kato is being built specifically to become that layer.

And it's defined as governed product intelligence. Now, I want to be super clear on what this actually means.

Kato is not another storefront. It is not another search engine. And it is definitely not just some general-purpose AI chatbot.

It's a specialized, dedicated intelligence layer sitting directly between what a human means and what the commerce ecosystem contains.

Kato's three-pillar architecture

Section three, Kato's three-pillar architecture. Let's dissect the system.

So Kato actually separates three major concerns that general-purpose systems usually just mash altogether.

First up, governed knowledge. This is the structured intelligence, the hard facts about ingredients, health claims, and safety context.

Second, universal product intelligence. This is Kato looking outward at the market, evaluating products across external catalogs using direct back-end connections like Shopify partner APIs.

And third, merchant intelligence. This looks inward at a specific operator using their private merchant information.

And here's the crucial part. This relationship is strictly one way. Kato's universal intelligence never relies on a single merchant's private operational data to form its baseline understanding.

The entire strict philosophy behind Kato's governed knowledge pillar can literally be summed up in three words. Nothing autopublishes.

Seriously, knowledge growth here isn't about just scraping the web at all costs. New knowledge only enters Kato through a rigorous, reviewer-gated process.

It absolutely requires evidence quality, clear provenance, safety context, and actual human approval.

Kato deliberately draws a line between what it recognizes, what the evidence supports, what it can safely conclude, and importantly, what remains uncertain.

To see how this plays out in the real world, think about how people actually shop. We don't talk using rigid database terminology, right? We talk like humans.

We say things like, "Hey, I want something for sleep, but I don't want melatonin." Or, "I already drink a ton of coffee. What's the difference between these two boosters?"

Kato's language intelligence is built to take that natural, messy shopper language and turn it into structured intent, preferences, and hard constraints.

The goal isn't just word matching. It's genuinely understanding what matters for a person's health decision.

And the results of that language understanding are honestly striking. On Kato's strict testing benchmarks, the system hit 0% hard constraint violations. Zero.

It also scored 100% in knowing exactly when to escalate to safety protocols, and 96% in correctly knowing when to simply say "I don't know" and abstain.

Basically, if you say no melatonin, Kato understands that as a rock-solid, non-negotiable rule. It's not just a keyword it playfully weighs in an algorithm.

Understanding versus recommending

Section four, understanding versus recommending. The core architectural philosophy.

When evaluating a product, Kato deliberately separates three very distinct questions.

Question one, what is this product? Question two, what does the evidence actually support? And question three, is Kato justified in recommending it for this specific decision?

Standard AI commerce tools treat those as the exact same question. If they can identify the product, they'll recommend it.

But Kato breaks them apart, because recognizing a bottle of vitamins is very different from proving it works. And that is very different from responsibly telling someone they should take it.

Let's look at the actual numbers here. In current testing across real Shopify catalogs, Kato achieved a 73.7% product-level understanding.

Pretty great. But the verified governed fit for a specific shopper need? That was only about 9.1%.

Now you might look at that steep drop-off and think, "Wow, is that a flaw?" But that difference isn't a limitation of the architecture. It is the architecture.

Kato can understand substantially more products than it is actually willing to recommend.

The big takeaway here is that understanding a product and getting permission to recommend it are entirely separate gates.

Look, a system optimized purely for making sales could easily just take that 73% recognition rate and turn it straight into recommendations. Kato deliberately does not do that.

Its goal isn't to maximize the sheer number of products it spits out. The goal is to continuously expand what it can responsibly understand without ever lowering that bar for evidence.

Every single decision, including the decision not to recommend something, is built to be fully explainable and completely traceable.

And Kato is putting this architecture into practice right now. The proof of concept is actively running.

As of today, the system manages 51 core canonical ingredients, 65 ingredient forms, and 63 strictly published governed statements. Plus, it has benchmarked 548 real Shopify catalog product cards.

They are literally building the measurement infrastructure right alongside the intelligence itself.

They aren't just tracking how often Kato gives an answer. They're tracking whether Kato is making better, safer decisions based on actual governed evidence.

The agentic commerce future

Which brings us to section five, the agentic commerce future. Reshaping the stack.

So, how does this build a completely new framework for how we buy things? The standard AI commerce stack we're seeing emerge is pretty simple. Shopper intent goes to an agent, which checks a catalog, and boom, you get a transaction.

But in high-complexity categories like health and wellness, we absolutely need this new critical step we've been talking about.

The stack must become shopper intent to an agent, then routed through governed product intelligence before ever hitting the catalog and moving to that transaction.

And that brings us to this closing thought. Shopify and platforms like it are doing the incredible heavy lifting of making the product universe accessible.

But Kato is doing the entirely different job of making complex health product decisions understandable. We need both of these jobs to exist.

Agentic commerce, where AI agents are actually out there making purchase decisions for us, can only work safely in these categories if both the accessibility layer and the intelligence layer are fully functioning.

So I'll leave you with this question to chew on. As AI continues to weave its way into literally every facet of our shopping experience, what happens to AI systems in high-stakes categories?

If this intelligence layer simply doesn't exist, if we're relying solely on commercial signals and marketing copy for our health and wellness decisions, what are the real-world consequences of an ungrounded recommendation?

It's a question the e-commerce world desperately has to answer. And as we've seen today, Kato's architecture is actively building that solution.

Thanks so much for joining me for this explainer. And remember to keep questioning the architecture behind the tools you use.

Frequently asked questions

Short answers to the questions this explainer raises.

What is Kato?: Kato is a governed product intelligence layer for supplements and other complex health products. It sits directly between what a shopper actually means and what a commerce catalog contains, so recommendations rest on reviewed evidence instead of marketing copy and product titles.
How does Kato's three-pillar architecture work?: Kato separates three concerns that general-purpose systems usually merge: governed knowledge (structured facts about ingredients, health claims, and safety), universal product intelligence (what the market offers, read through connections such as Shopify partner APIs), and merchant intelligence (a single operator's private information). The relationship runs one way, so Kato's universal understanding never depends on any one merchant's private operational data.
Why does Kato understand far more products than it recommends?: Understanding a product and earning permission to recommend it are two separate gates. In testing across real Shopify catalogs, Kato reached 73.7% product-level understanding but only about 9.1% verified governed fit for a specific shopper need. That gap is the architecture working as designed, because recognizing a bottle of vitamins is not the same as proving it works or responsibly telling someone to take it.
What does 'nothing autopublishes' mean?: New knowledge only enters Kato through a reviewer-gated process that requires evidence quality, clear provenance, safety context, and human approval. Nothing is added by scraping the web and nothing publishes itself, so Kato can distinguish what it recognizes, what the evidence supports, what it can safely conclude, and what remains uncertain.
How does Kato handle hard constraints like 'no melatonin'?: Kato's language intelligence converts natural shopper language, such as wanting something for sleep without melatonin, into structured intent, preferences, and non-negotiable constraints. On its strict testing benchmarks Kato recorded 0% hard constraint violations, escalated to safety protocols 100% of the time when required, and chose to abstain rather than guess in 96% of the cases where it should.
What is agentic commerce, and why does it need an intelligence layer?: Agentic commerce is shopping carried out by AI agents. In the emerging stack, shopper intent goes to an agent that checks a catalog and completes a transaction. In high-stakes categories like health and wellness that stack needs one more step, with intent routed through governed product intelligence before it ever reaches the catalog, so agents can recommend with a defensible basis rather than on commercial signals alone.
Where does Kato stand today?: The proof of concept is running: 51 core canonical ingredients, 65 ingredient forms, 63 strictly published governed statements, and 548 benchmarked Shopify catalog product cards. Kato measures not only how often it answers, but whether its decisions are better and safer, with every decision built to be explainable and traceable.