Short answerAgentic commerce describes AI agents that search, compare and buy products for consumers or procurement teams. Agents act on machine-readable data, so they need accurate price, live availability, complete specifications, compliance attributes and clear terms through feeds and APIs. Suppliers with structured, current product data become eligible for agent shortlists.
Key takeaways
- AI agents filter products by hard constraints such as specification, price ceiling, delivery date and certification, so a missing attribute removes a product from consideration.
- Emerging agent protocols and commerce APIs, alongside established standards such as schema.org, GS1 identifiers and BMEcat catalogues, are the routes through which agents read product data.
- Price and availability must refresh in near real time, because an agent that buys on stale data creates order failures that damage both channel trust and conversion.
Agents shortlist only what they can read
AI agents choose products from machine-readable data, so incomplete or stale data excludes a product before any comparison starts. A consumer agent asked to find a waterproof jacket under a set price with delivery by Friday applies each constraint as a filter. A procurement agent sourcing replacement bearings checks dimensions, load rating, certification and contract price. In both cases, a product without the relevant attribute fails the filter. Brand preference and persuasive copy carry less weight than data completeness. This makes agentic commerce a product data discipline first.
The effect differs by segment. Consumer agents emphasise price, delivery speed, reviews and returns, while procurement agents emphasise contract terms, approved supplier status, classification codes and documentation. Both read the same underlying record, which argues for one governed product master feeding every channel instead of separate datasets maintained per sales route.
Protocols and feeds expose data to agents
Agents reach product data through several channels, and most organisations will need more than one. Commerce and AI platforms have announced agent protocols and checkout APIs since 2025, and these remain young and fragmented. Established formats carry most of the load today:
| Channel | Typical use | Data carried |
|---|---|---|
| Schema.org markup on product pages | Discovery by AI assistants | Identity, price, availability, specifications |
| Merchant and marketplace feeds | Shopping surfaces and AI shopping features | Full offer data, images, variants |
| GS1 identifiers and GDSN | Trusted identity and master data | GTIN, brand, classification, logistics |
| BMEcat, cXML, punchout catalogues | B2B procurement systems | Contract price, units, classification codes |
| Commerce APIs and agent protocols | Real-time queries and checkout | Live stock, price, order actions |
A central PIM that publishes to all of these keeps facts consistent. The product data syndication guide covers feed management in detail.
Agents need four data domains
Agent-ready product data spans four domains, each with distinct owners and refresh rates.
- Commercial: price including VAT logic, currency, discounts, minimum order quantity, delivery cost and lead time. Owned by commerce or ERP, refreshed in near real time.
- Availability: stock per location, delivery promise and backorder status. Owned by ERP or order management.
- Specifications: dimensions, materials, performance values with units, compatibility and variants. Owned by PIM, sourced from PLM.
- Compliance: CE marking, safety warnings under GPSR (EU) 2023/988, substance data, certifications, packaging and sustainability attributes. Owned by regulatory teams.
Procurement agents often add classification codes such as ECLASS, ETIM or UNSPSC, which let buyer systems match products to internal categories.
Trust signals reduce agent risk
Agents acting on behalf of a buyer weigh the risk of a wrong purchase. Clear returns policies, warranty terms, verified identifiers, consistent data across sources and documented certifications all reduce that risk. Conflicting values between a brand site and a marketplace listing, or a GTIN that does not match the GS1 registry, give an agent reason to prefer a competitor. The same consistency principles that drive Generative Engine Optimisation apply to transactional agents.
Prepare in three steps
A practical preparation plan runs over an estimated 6 to 12 months for a mid-sized catalogue.
- Audit agent readiness. Measure attribute completeness per category for the constraints buyers apply, check identifier validity and compare price and availability latency across channels.
- Close the data gaps. Extend the attribute model with compliance, compatibility and use-context fields, enforce units and controlled values, and connect ERP for live commercial data. The product data quality governance framework provides the controls.
- Expose and monitor. Publish complete schema.org markup and feeds, evaluate emerging agent protocols with low commitment, and track agent-originated traffic and order errors as distinct metrics.
Governance should define which actions agents may complete, such as quote, reserve or order, and which require human confirmation, especially in B2B contracts.
Architecture determines speed of response
Agent channels reward organisations that can publish one consistent record to many endpoints quickly. Point-to-point exports from ERP or spreadsheets struggle with real-time price and stock. An API-first stack with PIM as the master for specifications, ERP for commercial data and an integration layer for distribution adapts faster as protocols mature. The composable MACH product stack describes this pattern, and suites suited to the role appear under /pxm/.
Frequently asked questions
What is agentic commerce?
Agentic commerce describes AI agents that search, compare and complete purchases on behalf of consumers or business buyers. The agent applies the user's constraints, such as budget, specification and delivery date, to machine-readable product data, then shortlists or orders products within permissions the user has set.
Which product data do AI shopping agents need most?
Agents need accurate price, live availability, complete specifications with units, valid identifiers such as GTIN, compliance attributes and clear delivery and returns terms. In B2B procurement, contract prices and classification codes such as ECLASS, ETIM or UNSPSC are also essential.
Should companies adopt new agent protocols now?
Agent protocols are young and fragmented, so broad commitments carry risk. A balanced approach completes structured data, feeds and schema.org markup first, since every protocol depends on them, then pilots one or two protocols relevant to key channels and monitors adoption.