Using AI in public procurement the correct way

I recently watched a presentation about using AI in public procurement and, for once, there were a few ideas in it that felt genuinely useful.

Many procurement tasks are manual. You search for opportunities, open a bunch of tenders that looked relevant but turn out not to be, finally find one that might be interesting, and then start going through the documents to understand what the buyer actually wants and whether it's even worth spending time on.

After that comes the less exciting part: making checklists, finding requirements scattered across different annexes, checking whether some declaration is missing, and trying to figure out if the same requirement appears in three different places with slightly different wording.

I have done enough tenders in my life to know that this part does not become more exciting with experience.

Where AI already helps

Tender monitoring is an obvious example, because most companies still rely on CPV codes, keywords and filters, which work, but not particularly well. AI can actually read the tender and compare it with what your company does, which could save a lot of time.

The same applies to deciding whether you should participate at all. I think companies still spend far too much time on tenders they have very little chance of winning, partly because the logic is often simply that if we qualify, we should bid, which is not really a strategy.

Instead you should be asking: who is the customer, who else is likely to participate, have you worked with them before, is your price realistically competitive, is there a strategic reason you want this customer, how much time will the offer take, and what is the realistic probability that you could win the contract.

AI can help with that analysis, especially if you give it more than just the procurement documents. Previous tenders, earlier projects, information about the customer and competitors all add context, and most companies already have a surprising amount of this information sitting in different places.

Start with compliance checking

Then there is compliance checking, which is probably the use case I would personally start with. Give AI all the procurement documents, let it extract the requirements and tell you where each one comes from, then give it your offer and ask it to find what is missing, unclear or potentially risky.

AI can be very confidently wrong, and public procurement does not forgive confident mistakes.

I would still check everything myself before submitting. One mistake can make several weeks of otherwise very good work completely irrelevant.

What stays a human job

If AI takes away a large part of the repetitive work, experienced specialists can spend much more time on the things that actually matter: understanding the customer, deciding what to offer, finding the right partners, working on pricing and making the offer better.

They still need to be good at procurement and notice when AI is talking nonsense, which I suspect will remain a useful skill for quite some time.