Many companies want to build with AI but do not have an in-house team ready to do it. Working with an outside partner can be the fastest way to get started, but AI projects come with uncertainties that ordinary software projects do not. The right partner helps you navigate them. The wrong one can spend your budget on an impressive demo that never reaches real users.

These ten questions help you tell the difference.

Questions about understanding your problem

1. "What would you need to know before estimating this?"

A good partner asks about your users, your data, your systems, and what success looks like before talking about technology or price. Be cautious of anyone who quotes a fixed price after a single short call.

2. "Is AI actually the right tool for this?"

Sometimes the best answer is simpler automation, or a better process. A partner willing to say "you may not need AI for this part" is far more trustworthy than one who recommends AI for everything.

3. "What is the riskiest part of this project, and how would you test it first?"

Every AI project has uncertainties: whether the model can do the task well, whether the data is good enough, whether users will adopt it. Look for a plan that tests the biggest risk early and cheaply.

Questions about how they work

4. "How will we see progress?"

You should see working software regularly, not just slides and status reports. Ask what you will have in hand after the first few weeks.

5. "How do you measure whether the AI is good enough?"

Strong teams build a set of real test cases and measure answers against them. Vague answers like "we'll test it thoroughly" are a warning sign.

6. "What happens when the AI gets something wrong?"

Every AI system makes mistakes. Ask how errors will be detected, how users can correct them, and where a human stays in the loop.

7. "How will you handle our data and security?"

Ask which data will be sent to which services, how permissions are respected, and what happens to logs. A good partner has clear answers and puts them in writing.

Questions about the long term

8. "Who owns the code, the prompts, and the data?"

You should own what you pay for, including source code, prompts, configurations, and any data you provide. Make sure the agreement says so.

9. "What does it cost to run once it is live?"

Model usage and hosting create ongoing costs. A good partner estimates them up front and designs to keep them reasonable.

10. "What happens after launch?"

AI systems need monitoring and updates. Ask about handover, documentation, and options for ongoing support, and whether your own team could take it over if you wanted.

Warning signs

  • Guaranteed accuracy. No honest team can promise an AI that is never wrong.
  • Technology first, problem second. If the pitch is about models and buzzwords rather than your users, keep looking.
  • No early working version. Long periods without anything you can try are a risk.
  • Unclear ownership of code, prompts, or data.
  • No plan for running costs or for what happens after launch.

Good signs

  • They ask more questions than they answer in the first conversation.
  • They suggest starting small and proving value before expanding.
  • They show how they test quality, not just how they build features.
  • They are open about what AI can and cannot do for your case.
  • You understand their plan, and they explain trade-offs in plain language.

A simple comparison checklist

When you speak with potential partners, record the same information for each:

  • Problem and first milestone: What will the first version do, and what is deliberately out of scope?
  • Evidence: What relevant work or demonstration can they share, and which details can you independently verify?
  • Quality and safeguards: Which real examples will they test, and how will people review or correct errors?
  • Data and ownership: Where will data go, who can access it, and who owns the code and configuration?
  • Commercial assumptions: What is included in the estimate, what could change it, and what ongoing costs should you expect?
  • Handover: What documentation and support will your team receive after launch?

Treat these as discussion prompts rather than a universal scorecard: the right answers depend on your project and risk level.

Making the decision

A short, paid discovery or pilot phase is often the best way to choose. It lets you see how a team thinks, communicates, and delivers before committing to a larger project, and it leaves you with something useful either way.

We are always happy to answer these questions about our own work. If you are evaluating partners for an AI project, feel free to get in touch.

Learn how we approach AI projects or send us your questions.