Build or buy AI agents for procurement?
Most companies are choosing to build their own AI agents. In most cases, that's the wrong call.
Should companies build or buy AI agents for procurement?
Buying will almost always be the better choice. A few reasons:
- A specialized vendor has already built this workflow 10 times. You'll start from zero and learn by making the same mistakes. The difference in speed and quality will be enormous
- Procurement is almost always at the back of the IT queue. In practice, that means 6 to 12 months for something that could be done in 30 days
- Maintenance is permanent and rarely makes it into the business case. Models evolve, processes change, integrations break. When you run the numbers, it almost never comes out cheaper.
I see one exception: companies that have a technical team embedded within procurement and know how to operate with agility — shipping fast, breaking things, iterating quickly. Someone strong in product needs to be watching the project closely. That combination is extremely rare in large companies in Brazil.
And even then, it probably doesn't make sense to build every type of agent. Some agents depend on expertise accumulated across many companies. Material master data registration is a good example of this.
We're still living in the initial AI agent hype cycle, and I see many companies diving in and taking far too long to build something that actually works. Unfortunately, this seems like a lesson most will have to learn the hard way.