Skip to content
← All guides
Guide

How do I evaluate AI vendors for procurement?

Short answer

Evaluate AI procurement vendors on five things. Whether they worked with businesses your size and in your industry. Whether the product reads from and writes to the systems you already run. Where your data goes and whether the model trains on it. Whether pricing is clear before you start. And whether they can name a client whose deployment is a year old and still in use. For most mid-sized businesses, a build around the existing ERP beats an enterprise platform.

You are probably here because a vendor has sent a very good deck. This guide covers what to check in it.

Ask who they have done this for at your size

Most AI procurement platforms were built for enterprises with a procurement department. Their case studies are companies with thousands of staff and spend measured in hundreds of millions. Ask for a customer with two or three hundred staff and one purchasing officer. If there is not one, the product can still be good, but it was not built for you.

Ask for a customer in your industry too. Procurement in construction, with subcontractor claims and retention, looks nothing like procurement in wholesale.

Ask what it connects to

The value of any procurement AI is in what it reads and where it writes. Ask whether it reads from and writes back to your ERP, your accounting system and your document store. Or does it expect to become the system of record? A tool that needs you to migrate purchase orders into it before it can help is a platform decision rather than a productivity one.

Ask where your data goes

Your supplier pricing and commercial terms are among the most sensitive information in the business. Ask three questions and insist on straight answers. Where is the data stored, and in which country. Does the vendor’s model train on your data. If it needs to, can it run privately inside your own environment?

Vague answers here are disqualifying.

Ask for the pricing in writing before the pilot

Enterprise procurement software is often priced per user, or as a percentage of spend under management. The second model means your cost rises as your business grows, regardless of value delivered. Ask for total cost over three years, including implementation. Then ask how the price changes when you use it for one workflow rather than all of them.

Be wary of a free pilot with unclear pricing at the end. The pilot is where you become dependent.

Ask them to demo on your documents

A demo on the vendor’s sample invoices proves nothing. Send them twenty of your real supplier invoices, including the messy ones, and ask for the demo on those. How it handles the photo of a paper invoice from your smallest supplier tells you more than any slide.

Ask for an old reference

Anyone can show you a customer from three months ago. Ask for one whose deployment is a year old. Ask that customer who operates it now, how often it needs the vendor, and what they stopped using.

The alternative most mid-sized businesses should consider

You already have the data. Purchase orders, invoices, receipts, quotes and supplier correspondence are sitting in your ERP, accounting system and email. What is missing is a layer that reads that data and does the repeated work with it.

For a business with a few hundred staff, build that layer around the systems you already run. Start with invoice matching and expand from there. That usually delivers most of the value of a platform at a fraction of the cost. There is no migration and no per user fee that grows every year.

We say that as a consultancy that builds those layers, so weigh it accordingly. The examples and use cases guide describes what gets built, and our approach to AI in procurement covers how.

Related questions

Questions

What does AI procurement software cost?

Enterprise procurement platforms with AI features are typically priced per user or as a percentage of spend under management. For a mid-sized business they usually land well into six figures a year once implementation is included. A build around your existing ERP for the two or three highest value workflows typically costs a fraction of that as a one off. The ongoing cost to run it is small.

What are red flags when evaluating AI procurement vendors?

A demo on their sample data rather than yours. Vague answers on where your data goes and whether the model trains on it. Pricing that depends on spend under management. No customer at your size who will speak to you. And any promise of a specific saving before they have seen your invoices.

Should a mid-sized business buy an AI procurement platform or build?

Usually build, on the ERP and accounting system you already have. The platforms are designed for procurement departments and their value depends on scale you do not have. The data you need already exists in your systems and just needs to be made usable.

What are best practices for implementing AI in procurement?

Start with one workflow, usually invoice matching. Test it against your own history before it goes live so you know the error rate. Keep every approval exactly where it is. Name the person inside the business who will own it. Only expand once the first workflow has run for a month.

Next step

We map where your hours go and price the work that is worth automating.