A framework for structuring commerce data so that AI systems can interpret it.
Presence is no longer the bar. Interpretability is.
Maintained by NSOLVIA™ · Author: Juan Carlos López Castaño, Founder Status: V1.0 · Document, not code
Full overview: What is a Semantic Commerce Layer?
Commerce infrastructure was built for databases and human browsing. A growing share of product discovery now runs through AI systems that read structure, not pages. Most catalogs are adequate for humans and traditional search but were never structured for machine interpretation.
The Semantic Commerce Layer™ is the interpretability layer between merchant catalogs and intelligent systems. It does not replace catalogs, feeds, or platforms — it sits alongside them and turns fragmented product data into a form machines can read consistently.
It makes no claim about rankings or sales. The claim is narrower: products a machine cannot interpret will struggle to participate in machine-driven commerce.
Catalogs were designed for two readers: the human browser and the traditional search engine. Their fields — title, description, brand, price, category, images — serve those readers well.
A third reader has arrived and reads differently. Recommendation, comparison, and retrieval systems need more than "a product exists." They need context: when to surface it, why a buyer would choose it, which use cases it serves, how it relates to alternatives.
A product can be fully present in a catalog and still be unreadable beyond basic attributes. The gap is not between products that exist and products that don't — it is between products that are listed and products that are structurally interpretable.
Several shifts arrive together: AI search answering directly instead of returning links; conversational commerce; recommendation systems deciding what a shopper sees before they search; and agents acting on a buyer's behalf.
In each case the entity evaluating a product is a system parsing structure, not only a human scanning a page. The readers are changing faster than the catalogs are.
Market context: McKinsey/ICSC project up to ~US$1T in U.S. B2C orchestrated retail revenue by 2030 (moderate scenario), ~US$3–5T globally, with ~50M daily shopping-related queries already on ChatGPT. Estimates vary widely by definition (~US$144B to multi-trillion); the direction, not the figure, is the point. Sources at end.
Three terms are easy to conflate. The distinction is where the value sits:
| Term | Meaning |
|---|---|
| Normalization | Inconsistent data → consistent data (same attribute, expressed the same way across products). |
| Enrichment | Adding information that was implicit or missing (use cases, intent, structured attributes). |
| Interpretability | The outcome of both: a product a machine can actually understand and act on. |
Normalization and enrichment are means. Interpretability is the end. A catalog can be perfectly normalized and still uninterpretable; partially enriched and still ambiguous. This framework is organized around the end state, not around any single technique.
Merchant Catalog → Semantic Commerce Layer™ → Feeds · APIs · AI Systems
(built for humans (interpretability: (the readers that now
and databases) structure + meaning) evaluate the product)
The layer improves product clarity, category precision, contextual understanding (use cases, intent), structured attributes, and discoverability potential.
We do not replace the catalog. We make it understandable to machines. No platform migration, no catalog rebuild — the existing catalog stays where it is and becomes interpretable.
What the layer produces is the subject of this document. How it produces it is out of scope.
A framework that works on only one platform is an integration, not infrastructure.
The transformation described here has been applied across Shopify, WooCommerce, Wix, Squarespace, and direct-to-consumer storefronts. It operates on product meaning, not on any one platform's format. Consistency across platforms is itself part of the evidence — it is harder to dismiss as a quirk of one ecosystem.
A fair objection: isn't this feed enrichment that existing feed tools already do?
Feed tools format and route existing catalog data to channels — they move data and largely preserve its structure. This layer addresses a different question: not where the data goes, but whether a machine can interpret it once it arrives. Formatting an ambiguous product for a channel does not make it interpretable. A clean feed of unstructured meaning is still unstructured meaning.
The layer sits upstream of distribution — it is the interpretive substrate the other pieces depend on.
A proposed, directional measure of how interpretable a product's data is to modern systems — not an industry standard, and explicitly not a prediction of rankings, traffic, or sales. If a better measure emerges, we adopt it.
Three dimensions:
- Structural — identifiers, images, structured attributes.
- Semantic — category precision, product type, use cases, functional intent. The dimension most closely associated with machine interpretability.
- Discoverability — signals platforms use to surface and recommend a product.
| Range | Reading |
|---|---|
| 80–100 | Appears well to AI systems |
| 50–79 | Partially interpretable |
| 0–49 | Largely invisible |
The internal weighting is intentionally not published; the bands are sufficient to use the score directionally.
Observed results, not a controlled study. Merchant identities anonymized except where NSOLVIA owns the store. Two examples come from PonteBella, the live merchant NSOLVIA operates as its reference deployment; the rest are independent merchant catalogs audited by NSOLVIA.
| Vertical | Platform | Before | After | Lift |
|---|---|---|---|---|
| Shapewear (PonteBella) | Shopify | 36 | 83 | +47 |
| Beverage | WooCommerce | 15 | 80 | +65 |
| Apparel | Wix | 20 | 68 | +48 |
| Breakfast / supplements | DTC | 15 | 75 | +60 |
| Skincare (PonteBella) | Shopify | 29 | 80 | +51 |
Worked example (shapewear, PonteBella, named with permission):
category "Shapewear" (generic) → apparel_shapewear; product type waist trainer; use cases everyday, postpartum, gym; functional intent waist shaping, breast lifting; size/color variants structured.
In every example, the largest gain occurred in the Semantic dimension. A generic title became information a machine can act on.
Across 80 audited catalogs, one factor stood out: whether a product's category could be resolved into a structured value.
- 65 catalogs with category resolved → average lift +48.4 (avg after-score 68).
- 15 catalogs where category could not be resolved → average lift +7.3 (avg after-score 24).
Both groups started from nearly the same average before-score (≈19 vs ≈18); the divergence appeared after enrichment. Where category remained ambiguous, the Semantic dimension consistently collapsed toward zero regardless of initial structural quality.
Category Resolution Hypothesis (observational): In this sample, catalogs whose category could be resolved showed substantially larger interpretability gains than those whose category remained ambiguous. Category resolution appears to be one of the strongest observed predictors of semantic interpretability in our sample.
This is a correlation in a sample of 80 — a finding, not a law. Category is one of several signals the layer structures (alongside use cases, functional intent, materials, purchase signals); its relative weight is a question continued auditing will refine.
Built to interoperate, not replace. Designed to coexist with Schema.org, Merchant Center and platform feeds, marketplace catalogs, commerce APIs, and emerging AI commerce protocols (e.g. MCP, ACP, UCP).
These protocols define how agents and platforms exchange and transact data. This layer concerns what they exchange — whether the underlying data is interpretable at all. The layer feeds the protocols; it does not compete with them.
Protocols will evolve — some consolidate, some get replaced, new ones appear. The interpretability problem exists independently of which protocol prevails. A product machines cannot understand remains a problem under any standard. That is why this framework is defined around interpretability, not around any single protocol of the moment.
- No guaranteed rankings on any platform.
- No guaranteed recommendations or surfacing by any AI system.
- No guaranteed traffic or sales increases.
- No preferential treatment from any AI platform, protocol, or marketplace.
No such guarantee is currently possible — commerce AI protocols are still being defined, and ranking behavior is controlled by the platforms, not by any data provider. Scope is limited to the quality, structure, and interpretability of commerce data.
Catalogs were built for databases. The next generation of commerce requires catalogs machines can understand. Participation will belong to merchants whose products can be interpreted — not merely those whose products are online. Presence is no longer the bar; interpretability is.
The merchants of the next decade may not be distinguished by who has the largest catalog, but by whose catalog can be understood.
- Retail Dive — https://www.retaildive.com/news/agentic-commerce-us-one-trillion-2030/818936/
- Digital Commerce 360 — https://www.digitalcommerce360.com/2025/10/20/mckinsey-forecast-5-trillion-agentic-commerce-sales-2030/
- MetaRouter — https://www.metarouter.io/post/the-agentic-commerce-opportunity-for-retailers
- eMarketer — https://www.emarketer.com/content/consumers-cozy-up-agentic-commerce-retailers-face-big-decisions
The Semantic Commerce Layer™ and the Agentic Catalog Readiness Score™ are trademarks of NSOLVIA™. © NSOLVIA™ · Juan Carlos López Castaño, Founder.
When an AI system reads a catalog, it isn't browsing — it's asking questions. The Semantic Commerce Layer structures product data to answer them:
- Product Discovery — surfacing the right product when a system or shopper looks for it.
- Goal-Oriented Shopping — matching products to an outcome a buyer is trying to reach, not just a keyword.
- Product Comparison — giving machines the structured attributes needed to compare options.
- Occasion / Use-Case Matching — connecting a product to the context it serves.
- Agentic Commerce Retrieval — making products retrievable and actionable by AI agents shopping on a buyer's behalf.
Is a semantic commerce layer the same as semantic search? No. Semantic search is how a system finds relevant results based on meaning rather than exact keywords. A semantic commerce layer is the structured, interpretable product data that makes that possible in the first place — it sits upstream. Semantic search reads the data; the semantic commerce layer is what makes the data readable.
How do I make my product catalog AI-readable? A catalog becomes AI-readable when each product carries structured, machine-interpretable data — a resolved category, a clear product type, use cases, and functional intent — not just a title and description. The Semantic Commerce Layer™ transforms an existing catalog into that structured form without rebuilding it.
What is product data enrichment for AI? Product data enrichment is the process of adding the information a machine needs but a catalog usually leaves implicit: use cases, purchase intent, structured attributes, and a precise category. Enrichment turns a listing a human can read into data an AI system can interpret and act on.
How do I get product visibility in AI search and agentic commerce? AI search and shopping agents evaluate structure, not pages. Products with normalized, enriched, machine-readable data can be interpreted and surfaced; products that remain ambiguous are largely invisible to these systems. Structuring the data is the prerequisite for participation.
What is a semantic layer for commerce? A semantic layer for commerce sits between a merchant's catalog and the AI systems that read it, translating fragmented product data into a consistent, interpretable form. It complements feeds, APIs, and protocols rather than replacing them.
What is the difference between a scanner and an enrichment engine? A scanner tells a merchant what is missing. An enrichment engine produces the corrected, structured data. Diagnosis identifies the gap; enrichment closes it.