On this page
Over the past two years, almost every software agency, consultancy and freelancer has added “AI” to their services. Some have deep experience taking AI into production. Many have built impressive demos but few systems that run a business process every day.
For a buyer, the difference can be hard to see in a sales meeting. This guide gives you twelve questions to ask, grouped by theme, with what a strong answer sounds like.
Understanding your business
1. What would you need to know about our process before recommending a solution?
A strong partner asks about your current workflow, the people involved, the systems used, the volume of work and how success is measured today. A weak one jumps straight to a technology or a tool.
2. How would you measure whether this project succeeded?
Look for a specific business metric, such as hours saved, response time, conversion rate or error rate, with a baseline measured before work starts. “The team will love it” is not a metric.
3. Would you ever recommend not using AI?
The best partners will tell you when a simple rule, a better process or an existing feature in your software is the right answer. If every problem gets an AI solution, be cautious.
Proof of delivery
4. Can you show us a system you built that runs in production today?
Ask what it does, how long it has been live, how many times it runs per month and what happens when it fails. Demos and prototypes are not the same thing.
5. What went wrong on a past project, and what did you change?
Experienced teams have clear, specific answers. Production AI always meets surprises: unusual inputs, changing data, model updates. How a partner handled them tells you more than any success story.
Quality and risk
6. How do you test the system before it goes live?
A strong answer includes a test set of real examples, measured accuracy against agreed criteria, and testing again after every change. Our article on testing AI automations describes what good looks like.
7. What happens when the AI is wrong?
Look for confidence thresholds, review steps for high-impact actions, clear escalation to people and logs of every decision. See human-in-the-loop design.
8. How do you handle our data and security requirements?
They should explain where data is processed, which AI providers are used and under what terms, how access to your systems is controlled, and whether they will sign your data processing agreement.
Integration and operations
9. How will this connect to the systems we already use?
The answer should name your actual systems, such as your CRM, help desk and ERP, and describe how the integration works. Integration is usually the largest part of a real project, so a vague answer here is a warning sign.
10. Who maintains the system after launch, and what does that cost?
AI systems need monitoring and improvement. A complete proposal includes the running costs and the support model, not only the build. See our budgeting guide.
Ownership and independence
11. Who owns the code, prompts, data and documentation?
You should. Make sure the contract states that everything built for you belongs to you, and that you receive the documentation needed for another team to take over.
12. How locked in are we to a specific AI provider?
A good architecture lets you switch AI models as better or cheaper ones appear. Ask how the solution would change if you wanted to move to a different provider in a year.
Signs of a strong partner
Taken together, the answers should show a partner who:
- Starts from your business problem, not from a technology.
- Has production experience and is open about past problems.
- Treats testing, security and integration as core work.
- Prices both the build and the run.
- Leaves you in control of what was built.
A practical next step
Before committing to a large project, consider a short, fixed-scope engagement: an assessment of one process, or a single use case taken to production. It shows you how the partner works, what they deliver and how they communicate, with limited risk. Our 90-day roadmap describes what that first project can look like.
We’re happy to answer all twelve questions about our own work. Book a call and ask us.


