How to choose an AI development partner: a buyer's checklist

Most AI vendors can give you a good demo. Far fewer have run an AI system with real users for months. If you're choosing a partner for an AI project, the questions below help tell the two apart. We hold ourselves to them too.

1. Ask to see something they run today

Look for an app, platform or dashboard that the team operates for real users. Slides and recorded demos don't count. Running software teaches you about monitoring, rate limits, running costs and failed deploys, and that experience shapes the architecture of your project. For SmartCherry, those are Rank Cherry and Stock Cherry, both on Google Play.

2. Ask how they run AI in production

• What happens when the model provider rate-limits you or goes down?
• How will we know whether the answers are still good after launch?
• What will this cost to run each month, and what keeps that cost in check?
• Who fixes it when it breaks, and how do they find out?

A team that has run AI in production will have specific answers to each of these.

3. Settle ownership and data before the SOW

The deliverables built for you should be yours, while the vendor's existing tools stay theirs. Your data should be used only for your project. Get all three in writing.

4. Agree the scope before the price

A well-defined build can be priced up front. An AI project with open questions about data quality or accuracy usually can't be, and a fixed quote at that stage hides a lot of assumptions. A short discovery that looks at your actual data first protects both sides.

Warning signs

• Accuracy promises made before anyone has seen your data
• A portfolio of demos with nothing in production
• Reluctance to say who will actually do the work
• No opinion on what you shouldn't build

We're happy to be judged against this list. Book a free 30-minute call.