What AI agents need from a supplier

What an agent is doing

A person buying a product calls, browses, asks a question and uses judgment about the answer. A purchasing agent acting on a buyer's instruction does the equivalent at scale: it gathers candidates, checks each against the requirement, compares price and lead time, and prepares an order for approval. It does this by reading what is published. It cannot overhear a salesperson say the figure is out of date.

Four requirements

• A catalog with stable identifiers. Manufacturer part number, the supplier's SKU and a cross-reference between them, in consistent formats. Without them the agent cannot be sure two listings are one item.
• Specifications with units, basis and revision. A dimension is a number, a unit and a basis (nominal or actual, with a tolerance where it applies). A rating names its standard. A spec sheet carries the revision it describes.
• Availability and lead time with a timestamp. A stock value without the time it was observed cannot be used for a decision. Unknown must be representable.
• Authority. Which facts the supplier is authoritative for, which it is repeating from the manufacturer, and what the supplier is authorized to sell and install.

Where suppliers drop out

An agent does not rank a supplier it cannot read. A catalog locked in PDFs, a price visible only after a login, a specification that exists on one page and contradicts another, a lead time that is a phone number: each is a point where the comparison continues without that supplier.

This is less visible than a lost search ranking. The buyer never sees the omission, and the supplier never sees a missed lead.

What the agent cannot be trusted to do

Reading is one thing. Acting is another. An agent can prepare a quote request or an order, but the commitment should pass through a person until the data it relied on has been measured against the record. The boundary between automation and approval is part of the design, not an afterthought.

Making a supplier readable

The work is mostly in the data a supplier already has: reconciling identifiers, putting units and revisions on specifications, dating availability, and stating who is authoritative for what. It is described in what makes a B2B catalog agent-ready and in the product information chain.

Questions