Framework

The AI readiness audit framework

Readiness is auditable. Five dimensions, each scored on a one-to-five scale against a defined rubric, give a manufacturer's executive team a defensible map of where AI investment will pay and where it will burn. The framework is the same one Acro Commerce uses to seed the Celeste discovery diagnostic, written here so an internal team can run it on their own before booking time with anyone.

Dimension 1: product data completeness and structure

AI engines retrieve from product data. If attributes are missing, units are inconsistent across SKUs, or the controlled vocabulary varies by source system, the engine cannot disambiguate the entity. The score for this dimension answers a single question: can an AI engine extract the spec, the dimensions, the compatibility set, and the offer for any product on the catalogue without guessing?

Run the score against a representative slice. Pick 30 SKUs across the catalogue's range (top sellers, long tail, configured, kit). For each, check: does the PIM hold every attribute the PDP shows? Are units consistent (in versus inch, lb versus pound)? Is the compatibility set explicit and machine-readable? Does the PDP carry Product and Offer schema that match what the page shows? Score 1 if half the SKUs fail any criterion. Score 5 if every SKU passes every criterion.

The fix is rarely a new tool. It is usually attribute hygiene work inside the PIM, with the ERP as the ultimate reference for units and identifiers. The detail is in AI-ready product data. Most manufacturers we see score a 2 or 3 here and could reach 4 inside a single quarter with focused attribute work.

Dimension 2: content chunkability and schema coverage

AI engines extract chunks. A chunk that answers a buyer question in one or two sentences before context wins citation. A chunk buried four paragraphs deep loses to a competitor's cleaner page. This dimension scores whether the top 100 content pages on the site are structured so an engine can extract clean answers, and whether the right schema is present and valid.

Concrete checks: does every H2 have an answer paragraph immediately below it? Are FAQ blocks rendered as FAQPage schema, not as unstructured accordion HTML? Do TechArticle pages carry TechArticle schema with author, datePublished, and dateModified? Do PDPs carry Product and Offer with current price and availability? Does the schema validate against schema.org and Google's Rich Results test? Score 1 if most of the top 100 fail. Score 5 if every page passes every check.

This dimension is the highest-leverage early investment. The same work that earns AI citation improves classical SEO, internal search, and dealer self-service. The detail is in schema markup as AI fuel.

Dimension 3: workflow and ERP signal quality

AI agents that quote price, availability, and lead time call APIs. If the ERP returns inconsistent prices for the same SKU across customer accounts, or if availability is cached overnight while a buyer is asking now, the agent will misquote. This dimension scores whether the operational truths an AI agent would need are exposed cleanly, consistently, and at the freshness the buyer expects.

Checks: does the catalogue API return prices for an authenticated account that match the ERP's customer-specific price list? Does availability reflect the same allocation rules the rep tells the buyer? Does the quoting API respect the contract terms, surcharges, and freight rules a CSR would apply? If any of these is no, the AI agent question is moot. Fix the API contract first. The detail is in AI agent APIs.

Most manufacturers score a 2 here. The ERP holds the truth, but the API layer is brittle, incomplete, or inconsistent. Acro Commerce's ERP integration and expansion work is the path to a 4. A 5 is rare and usually requires deliberate API design rather than incremental fixes.

Dimension 4: first-party data and consent

Manufacturers hold strong first-party signals: order history, account hierarchies, configured product specifications, support tickets, and the engineering question log. Most of this signal is locked in systems that AI use cases never reach. This dimension scores whether the manufacturer can use its own data legally, structurally, and operationally to power AI features.

Legal checks: does customer consent cover the use of order data for personalization or recommendations? Is anonymization on data going into model context defined? Is the data residency clear for the AI provider's region? Structural checks: is the data accessible to the model context, or trapped in an export-only system? Operational checks: who owns the freshness window, and what happens when a contract changes? Score 1 if any of these is unclear. Score 5 if all are documented and enforced.

This is the dimension that catches manufacturers off guard most often. The data is there, but it is not usable. The fix is rarely glamorous: consent language tightened in contracts, data access patterns built deliberately, model context curated rather than auto-fed. None of that requires a new model. All of it pays the next time a customer asks why their data shaped a recommendation.

Dimension 5: brand presence on AI-cited platforms

AI engines cite Wikipedia, YouTube, Reddit, LinkedIn, and a small set of authoritative trade publications disproportionately. This dimension scores whether the brand is present on those platforms in ways that produce citation, not just impressions.

Checks: does the brand have a Wikipedia entry that meets notability and sourcing rules? Are there at least 10 YouTube technical videos with structured transcripts in the past 12 months? Are engineers visibly answering real questions in the relevant communities? Are there third-party authoritative mentions (analyst reports, trade publication articles) with current data points? Score 1 if most checks fail. Score 5 if all pass at a quality consistent with the brand's market position.

This is the slowest-moving dimension and the one most manufacturers underinvest in. The detail and the patterns that work are in AI-cited platforms. Plan for a quarterly cadence, not a sprint.

How to use the score

Score each dimension from 1 (broken or absent) to 5 (excellent). Sum to a 5 to 25 scale. Most mid-market manufacturers we audit score between 11 and 15 on first run. The number itself is less useful than the gap analysis: which dimension is the lowest, and is it the constraint on the next initiative?

The order of fix is usually data, then schema, then workflow, then first-party, then platform presence. Skipping ahead is tempting because platform presence looks more interesting than attribute hygiene. The compounding works in the other direction. Manufacturers who fix data and schema first see retrieval improvements within a quarter. Manufacturers who go after platform presence first often discover their on-site retrieval cannot capitalize on the improved visibility.

If you want an outside read on the score, Acro Commerce can run the framework with Celeste as part of an AI readiness discovery engagement. The deliverable is the same five-dimension score, with the sequence and the integration plan that follow. Whether the work is internal or external, the score is the right first artefact. Get an honest read before scoping initiatives.

Frequently Asked Questions

Next Step

Get the foundation right before you build.

For readers scoping a platform decision or wanting a full architecture recommendation.