The next “buyer” won’t open your PDF
Procurement has spent years digitising approvals, workflows, and sourcing events—then we kept publishing the actual buying information as PDFs, email attachments, and “see website for details.” That works when a human is hunting and interpreting. It fails when the buyer is an autonomous agent that ranks options by structured attributes, constraints, and confidence.
Agentic commerce is starting to formalise around the idea that products and services must be represented in machine-readable formats so autonomous agents can discover, compare, and transact reliably. If you’re not represented in those formats, you’re not “hard to find.” You’re functionally absent.
Discoverability becomes a data problem, not a marketing problem
Humans tolerate ambiguity. Agents don’t. A category manager can infer that “industrial cleaner, 20L, citrus” is equivalent to another supplier’s “degreaser, 5 gallon, orange.” An agent needs explicit mappings: units, concentrations, hazard classes, compatibility claims, lead times by region, and what “equivalent” means in your organisation.
This is where many supplier portals quietly break down. They collect a vendor profile and a bank form, then stop. Agentic buying flips the priority: rich, structured offer data matters as much as onboarding data. The supplier that publishes clean, comparable attributes wins the shortlist before anyone asks for a quote.
A concrete scenario: why the agent skips you
Imagine an internal buying agent tasked with “monthly replenishment of nitrile gloves for Plant A,” constrained by: delivery within 48 hours, powder-free, specific thickness range, compliance with a named standard, and a maximum unit price. Supplier 1 exposes SKUs with thickness, AQL, certifications, pack sizes, and delivery promises as structured fields. Supplier 2 has a brochure that says “premium gloves, fast shipping.” The agent doesn’t negotiate with the brochure. It filters Supplier 2 out because it can’t prove fit.
Two-sided coin: build agents that buy—and become a supplier agents can buy from
Most teams treat “agentic procurement” as an internal automation topic: intake agents, sourcing agents, contract agents. That’s only half the story. The other half is external: your supply base must be machine-readable so your agents can shop effectively, and your organisation must be machine-readable so other companies’ agents can buy from you (or from your internal business units, if you sell cross-division).
The uncomfortable truth: your best negotiator can’t out-negotiate an agent that never invites you to the event. The shortlist is built upstream, by data quality and interoperability.
What “machine-readable” actually means in procurement terms
This isn’t about sprinkling AI on a catalog. It’s about representing your offers, constraints, and proof in a way software can parse without guessing. Think: structured product/service definitions, consistent identifiers, explicit units, policy-ready terms, and verifiable compliance artifacts.
Offer data: SKU/service IDs, names, descriptions, categories, attributes (dimensions, materials, performance), units of measure, pack sizes, substitutions/equivalents rules.
Commercials: price structures (tiers, volume breaks, surcharges), currency, validity dates, incoterms where relevant, minimum order quantities, cancellation rules.
Fulfilment: lead times by geography, shipping methods, cut-off times, backorder behaviour, service windows, capacity limits.
Risk and compliance: certifications, audit dates, insurance, safety data, ESG claims with evidence pointers, export controls, country-of-origin fields.
Contract hooks: standard terms, deviations, clause libraries mapped to common procurement requirements (e.g., data processing, confidentiality, liability caps).
Identity and matching: legal entity names, tax IDs, DUNS/other identifiers where used, address normalisation, bank details handled securely but linked correctly.
Notice what’s missing: “a beautiful website.” Agents don’t care. They care that “48-hour delivery” is a field with a scope and exceptions, not a slogan.
Prepare your buying agents: the supply base will be uneven for years
Even if standards mature quickly, suppliers won’t modernise at the same pace. Your agents must handle a messy middle: some suppliers will provide structured feeds; others will still send a rate card PDF. Early adopters will be tempted to ban the PDFs. That’s neat in theory and painful in operations.
A pragmatic approach: design your agent workflows with “confidence gates.” Let agents auto-award when data is complete and comparable; route to a human when key attributes are missing, or when the model is forced to infer. This avoids the worst failure mode: silent errors that look like efficiency.
Common mistake: automating the wrong bottleneck
Teams often start with an agent that drafts RFPs faster. The real bottleneck is usually downstream: comparing offers that aren’t expressed the same way. If you can’t normalise “hourly rate + travel” against “fixed fee per site,” your agent will either oversimplify or stall. Put effort into comparison schemas, not just generation.
Prepare to be found: supplier enablement becomes a competitive weapon
Procurement can push suppliers to provide structured data, but you’ll get better results if you treat it like enablement, not policing. Suppliers resist because it feels like unpaid admin work with unclear upside. Make the upside concrete: faster awards, fewer clarifications, fewer disputes, quicker payment.
Publish your required fields per category (not one generic template). A facilities services schema is not a chemicals schema.
Provide an example “golden record” offer and a validation tool so suppliers can test before submission.
Define acceptable evidence for claims (e.g., certification number + issuing body + expiry date), not just “upload certificate.”
Make exceptions explicit: which fields are mandatory for auto-award vs. which trigger a manual review.
Offer a migration path: start with top 20 suppliers and top 200 SKUs/services, then expand. Don’t boil the ocean.
If you sell, apply the same thinking internally. Many B2B suppliers still rely on account managers to “translate” their offer. Agentic buyers will reward suppliers that can be evaluated without translation.
Interoperability: standards are coming, but your data discipline has to come first
Emerging web standards for agent-to-retail interoperability are a signal, not a full solution. Waiting for a perfect standard is a comfortable excuse because it postpones hard work: cleaning master data, agreeing attribute definitions, and aligning category taxonomies.
Start with what you control: define canonical attributes per category, map them to your ERP/P2P fields, and insist on consistent units and identifiers. When standards settle, you’ll translate from something coherent instead of from chaos.
A near-term playbook (next 90–180 days)
Pick two categories: one SKU-heavy (e.g., MRO) and one service-heavy (e.g., temp labour, facilities). Agentic buying will stress them differently.
Define a minimum comparable dataset for each category: the 15–30 fields an agent needs to filter and rank without guessing.
Audit your current supplier data: identify where the truth lives (contract, catalog, email, invoice) and where it conflicts.
Add “machine-readability” to supplier performance: not as punishment, but as a path to preferred status and faster cycle times.
Build agent guardrails: confidence thresholds, escalation rules, and an audit trail that explains why a supplier was shortlisted or excluded.
The early-mover advantage here isn’t flashy. It’s quiet: your organisation becomes easier to buy from and easier to buy for. When agents become the default interface, “structured and trustworthy” will beat “well-known and persuasive” more often than people expect.