Most writing about AI in procurement is aimed at enterprises with a procurement department, a category management team and a seven figure software budget. This is for businesses with a purchasing officer, a spreadsheet and an ERP, which is most of them.
Six examples that get built
Invoice matching and price variance. Supplier invoices are matched to purchase orders and receipts line by line and coded to the right job or account. Where the price differs from the agreed rate, the invoice is flagged before it is paid. This is the highest return use case in most businesses because overcharges are common and almost never caught by hand. One distributor we worked with found six figures of supplier overcharges in the first six months.
Purchase orders from requests. A request from site or the floor arrives by email, text or WhatsApp. It is drafted into a purchase order against your catalogue and preferred suppliers, and the buyer approves it. The person who needs the thing does not have to learn the ERP.
Supplier selection from your own history. Why did we go with this supplier over that one on the last job? What did they quote, how did they perform, were they late? A Sydney builder we worked with made years of tenders, quotes and job files askable. That question now takes a minute instead of a day.
Supplier compliance documents. Insurance certificates, licences, safety documentation and declarations are read and filed against the supplier. When they lapse, the system chases them. Nobody enjoys this work and it is entirely automatable.
Quote comparison. Subcontractor and supplier quotes are read and laid out against the scope, with the gaps and exclusions highlighted. The buyer compares like with like instead of reading twelve PDFs.
Landed cost and receipt reconciliation. For importers and wholesalers, freight, duty and supplier invoices are pulled together. Landed cost is known when the stock arrives rather than months later.
What generative AI changed
Older procurement AI predicted demand and scored suppliers from structured data. Useful, but it needed a data team.
Generative AI reads documents. Invoices, quotes, certificates, emails, tenders. That is the shape of almost all procurement admin in a mid-sized business. It was impossible to automate until a model was able to do the reading. Now the reading can be automated and the judgement can be left with a person.
The limits
AI in procurement proposes and a person decides. Nothing is paid, ordered or approved without the same person signing off as today. That is the design rather than a limitation to work around.
Supplier pricing and commercial terms are sensitive. Where they are involved, the model must run privately inside your own environment so no document or question leaves the business.
And messy inputs cost more. A supplier who sends invoices as photos of paper is harder than one who sends structured PDFs. Discovery must surface this before anyone quotes a build.
Where to start
Invoice matching, almost always. It has the most hours in it, the errors are expensive, and the data it needs already exists in your accounting system. Purchase order drafting and supplier history usually come next.
More on how we approach AI in procurement. If you are looking at software, read how to evaluate the vendors.