What should I ask before hiring someone to build AI?

Use these questions as a conversation guide, not a script. The note under each one describes what a good answer sounds like. They are also the questions we expect to answer in our own AI integration proposals.

What problem do you think we're solving?

They restate it in their own words, including who does the work today and where it goes wrong. If they describe a product instead, they have not listened yet.

Is AI the right tool here, or would a simpler automation do?

A straight comparison. Sometimes a rule-based automation, a clearer form or a process change is enough, and they say so.

What would you build first?

One narrow workflow that can be tested on real examples, not a full platform on day one.

Which of our existing systems would it connect to?

They name your actual tools, such as your CRM, email, calendar or shared drive, and explain how the connection would work, or say what they still need to find out.

What data will the system need, and why?

Only what the task requires, with a reason for each item. Broad access to everything is a warning sign.

Where will our data be stored and processed, and which outside services will see it?

A specific answer covering the AI model provider, hosting and any other services involved, and whether their terms allow your data to be used for training.

Who on your side can access our systems, and how is that removed later?

Named people, individual accounts rather than shared logins, and a plan to remove access when the work ends.

For more on this area, see how secure AI automation is for business data.

What happens when the AI gets something wrong?

They expect it to happen and describe how mistakes are caught, logged and corrected.

Which actions will need a person's approval?

Anything that contacts customers, commits money or changes records permanently, at least until the system has proven reliable.

How will you test it before it goes live?

On real examples from your business, including awkward ones, with you reviewing the results.

What happens if a connected service is down?

The system stops safely, alerts someone and does not lose work in progress.

For example, a bookkeeping firm asks a provider to sort incoming client emails and draft replies. A good answer to the first question would be: drafts are never sent automatically at first, a staff member approves each one, and anything the system cannot classify goes to a shared inbox with a note explaining why. A weak answer is "the model is very accurate."

What drives the build cost, and what will it cost to run?

A breakdown of build, model usage, subscriptions, hosting and maintenance, with an explanation of what makes each go up or down.

Who owns the code, accounts and data?

Set out in writing before work starts, with accounts in your business's name wherever possible.

What documentation will we receive?

Enough for another developer to understand, run and change the system without starting over.

How will we know whether it was worth it?

They suggest agreeing a baseline and what to measure before building. Our guide to measuring ROI from AI automation explains how.

Who do we contact when something breaks?

A named person and a response expectation in writing, not a general inbox.

How will the system be monitored?

Logs of what it did, alerts when something fails, and a regular review of a sample of its work.

What happens when an AI model or connected tool changes?

They explain how updates are tested before they reach your live system.

Can we extend it later without rebuilding?

A clear view of how the first workflow could grow, and what would need to change.

It depends on how much ongoing work there is. A full-time hire makes sense when AI and automation will be continuous work across several teams, and you have enough for that person to do and someone who can manage the work. For a first project or a few defined workflows, an outside partner is usually the more practical start, because you get design, build and integration skills without committing to a new role.

Either way, the questions above apply, whether you are interviewing a candidate or comparing providers. For the wider decision, see how to choose an AI integration company.