Insights · 6 min read
How to choose an AI development company: 8 questions to ask
By Gnaneswar Kalinga, founder, MarketRun · Updated
The short answer
Choose by evidence, not by pitch. Ask to see products the company built and still runs, a written scope and price before you commit, how the AI is kept in check, who owns the code and data, how accuracy is measured, who you will actually work with, what happens after launch, and what it would refuse to build. A company that answers all eight plainly is usually safe for a small, paid first step.
Hiring someone to build AI or an app is hard because every demo looks good. These eight questions work on any company, including us. Ask them in writing and keep the answers.
| Question | A good answer | A red flag |
|---|---|---|
| 1. What have you built that is still running? | Named products or systems, with their real status and what went wrong along the way | Only slides, mock-ups or logos with no detail |
| 2. What will it cost, and when is that fixed? | A published starting price, then a fixed quote after a defined scoping step | A vague range, or a quote with no scoping behind it |
| 3. How do you stop the AI doing something wrong? | A person approves anything customers see, every action is recorded, and the AI can only do what it is allowed to | "It runs fully on its own" with no safeguards |
| 4. How do you measure accuracy? | A set of real test cases, scored before launch and after changes | "Our model is very accurate" with no numbers or method |
| 5. Who owns the code, data and accounts? | You do, in writing, with accounts set up in your name | The company keeps the code or holds your data in its own accounts |
| 6. Who will I actually work with? | Named people, a weekly rhythm, one place to talk | A sales contact who disappears after signing |
| 7. What happens after launch? | Fixes included for a stated period, then a clear care plan or a clean handover | Silence, or a forced long contract |
| 8. What would you refuse to build or advise against? | Real examples, including telling clients to wait or use a ready-made tool | "We can build anything" |
Why evidence beats a pitch
AI projects fail quietly: the demo works, the real data is messier, and nobody measured. Companies that run their own products have already met those problems and can tell you what they did about them. Founders on Indian startup forums give the same advice: ask for a demo of apps the team has built.
Start small and paid
A short, paid first step, such as a two-week scoping sprint, shows how a team really works before you commit a build budget. If they will not scope first, treat any fixed price as a guess.
How MarketRun answers
- Products we built and run, with honest status, are on our products page.
- Starting prices are published. Websites and automation are quoted after a 30-minute call; apps and software start with the two-week sprint from ₹50,000, which ends in a fixed quote.
- AI never sends your customers anything unless a person approved it or it comes from answers your team approved.
- Code is yours on full payment, and accounts are set up in your name.
- Our how-we-work page lists the weekly rhythm.
Questions people ask
What should I ask an app company before building an MVP?
The same eight questions, plus: what is in the first version and what is not, who will be on the build, and what support costs after launch. Ask for the scope on one page.
Should I pick the cheapest quote?
Treat a very low quote as a question, not a bargain. It usually means a template, a copied app, or a scope that grows later. Compare written scopes, not totals.