There is a version of the AI-in-procurement pitch that is genuinely useful, and a version that is a screensaver with a subscription. Telling them apart is now one of the more valuable skills a buyer can have, because the wrong purchase does not just waste money, it buries your team in a tool they have to feed and cannot trust. The honest position is not that AI is hype, and not that it changes everything. It is that AI earns its place in some procurement jobs and not yet in others, and you should know which is which before a salesperson tells you.
Where it genuinely helps
AI is at its most useful in procurement doing the unglamorous work of making sense of your own data. Classifying and cleaning messy spend so you can finally see where the money goes. Spotting that you are buying the same thing from six suppliers at four prices. Forecasting demand so you order the right quantity. Reading a stack of contracts and flagging the unusual clauses. These are pattern jobs on large, dull datasets, which is exactly what the technology is good at, and they are the jobs humans do slowly and hate.
Where it burns you
The trouble starts when the promise moves from "help you see" to "decide for you". Autonomous negotiation, AI that sets strategy, tools that promise to run supplier relationships without a human, these are where the gap between demo and reality is widest. Procurement decisions carry judgement, relationships and risk that a model does not hold. A tool that recommends and lets a person decide is an asset. A tool that acts on its own in a domain full of exceptions is a liability waiting for the exception.
The solution, as a use-case maturity map
Score any AI tool before you buy it
| The job | AI's role | Verdict |
|---|---|---|
| Spend classification and visibility | Do the work, human checks | Buy, this is the strong case |
| Demand and price forecasting | Predict, human decides | Buy, with your own data |
| Contract and clause review | Flag, human judges | Buy, as a first-pass filter |
| Supplier risk monitoring | Alert, human investigates | Useful, verify the sources |
| Autonomous negotiation and strategy | Acts alone | Wait, keep the human in charge |
The rule of thumb: AI that informs a human decision earns its place. AI that replaces the human decision in a domain full of exceptions does not, yet.
The buying test
Before you sign, make the vendor prove it on your data, not their demo. A tool that shines on a clean sample and falls over on your actual, messy spend is telling you something. Ask what happens when it is wrong, who is accountable, and how you leave if it does not work. And be honest about the boring precondition: AI on top of bad data gives you confident, fast, wrong answers. The unglamorous work of getting your spend data clean is what makes any of this pay, which is why the best first use of AI in procurement is usually cleaning up the data itself.
Questions to ask any AI procurement vendor
- Will it run on our real, messy data, or only on your clean demo?
- Does it inform a human decision, or make the decision itself?
- What happens when it is wrong, and who is accountable then?
- What does it actually cost to feed and maintain, in our time?
- How do we get our data out and leave if it does not work?
Used well, AI is one of the better things to happen to procurement in years, because it finally makes the mountain of spend data legible to a small team with no analysts. Used badly, it is an expensive way to automate a mess. The line between the two is not the technology. It is whether you keep the human in charge of the decisions and put the machine to work on the data. Buy on that basis and AI earns its place. Buy on the brochure and it will cost you twice.
Notes and sources
From the book
Procurement That PerformsTurn supplier spend into growth and performance, using AI where it earns its place, with no procurement team.