How to choose an AI consultant. Ask before you sign.
Before hiring an AI consultant, ask how they will measure a baseline and report ROI, who builds the work, who owns the code and accounts, how they handle your data, what happens after launch and how they bring your staff along. Modern Thesis, an AI strategy and implementation firm for non-tech companies, wrote this guide to make that comparison easier.
Why does the choice of AI consultant matter so much?
The choice matters because most of the risk in AI work sits in execution, not in the idea. In RSM's 2025 middle market survey, 62% of companies said generative AI had been harder to implement than expected, and 70% said they needed outside help to get the most from it.1 MIT's 2025 study found that projects built with outside partners reached deployment about 67% of the time, against about 33% for internal builds.2 A good partner improves those odds. A poor one costs money and goodwill with your team.
What should you ask an AI consultant before you hire them?
Ask questions that show whether a firm only advises or also builds, measures and stays. Here are eight, with what a strong answer sounds like.
1. How will you measure a baseline and report ROI?
A strong answer names the workflows, measures their time, cost and error rates before anything is built, sets a target and reports results after launch in numbers you already track. A weak answer promises "efficiency gains" or counts automations, with no before number to compare against. See our guide to measuring AI ROI for the method.
2. Who will actually build it?
A strong answer is specific. You meet the people who will design and build the system before you sign, and they take part in the strategy work rather than receiving a document afterward. A weak answer is vague, or the people selling the work disappear once the contract is signed.
3. Who owns the code, the accounts and the documentation?
You should. A strong answer puts software accounts and AI provider accounts in your company's name, hands over code and documentation, and lets you move the work to someone else if you choose. Be careful with arrangements where the system runs on the provider's own platform and stops if you stop paying.
4. How will you handle our data?
A strong answer is in writing before anyone touches your systems. It limits access to what the work needs, says plainly that your data won't be used to train AI models, names the AI providers involved and their terms, and explains how access is removed when the engagement ends. If the answer is "don't worry about it," worry.
5. What happens after launch?
AI systems need looking after: AI providers retire models, connected software changes and accuracy drifts. A strong answer describes who monitors the system, how often results are reviewed and what upkeep costs. A weak answer ends at handover. Our page on AI stewardship covers what post-launch care should include.
6. How will you bring our staff along?
BCG's 2024 survey traced about 70% of AI problems to people and processes, and only about 10% to the AI algorithms themselves.3 A strong answer includes a plan for communication, training and incentives, and speaks directly to the fear that AI puts jobs at risk. A weak answer is a training video and a login.
7. Can you show us working software on our industry's data?
A strong answer is a demonstration on public data from your industry, or on a sample of one of your own workflows, built for the conversation. A weak answer is a polished demo of a general chatbot that never touches work like yours.
8. Where would you tell us not to use AI?
A strong answer comes easily, with examples of work where AI won't pay for itself or where a person's judgment should stay in charge. A firm that sees AI everywhere has stopped looking at the business.
What are the red flags when hiring an AI consultant?
- Generic output. Recommendations that could apply to any company, with no mention of your systems, your numbers or your people. Some of this is chatbot output forwarded under a consultant's name.
- Demos that never touch real work. Impressive on a sample, then nobody can show it running on your documents, your data or your volume.
- No plan for people. A rollout that assumes staff will adopt the system because it exists.
- Results promised before measurement. A savings percentage quoted before anyone has looked at your workflows.
- Lock-in. Accounts in the provider's name, code you can't see, or a system that only runs on their platform.
- Long strategy, no build. Months of workshops before anything runs. A first working system should arrive while momentum is high.
- Disrespect for what you already use. Good partners start with the tools you pay for and build custom only where those fall short.
Does the type of firm matter?
The type of firm matters less than the answers above, but each category has a natural fit. Large consulting firms suit enterprise programs across many countries. Automation agencies suit one clearly defined task. Fractional CTOs suit companies that need part-time leadership across all of their technology. Strategy and implementation firms suit companies that need to decide what to build, build it, bring the team along and keep it running. Pick the category that matches the problem, then use the questions to pick the firm.
How Modern Thesis handles this
We hold ourselves to the same questions. Every engagement starts with a strategic assessment that measures a baseline on key workflows. You talk to the people who will build the work: our AI specialists and Forward Deployed Engineers, led by Jim Coleman on strategy. You own your data, accounts, code and documentation, and we never use client data to train AI models. Our data policy sets out the rest. Change management is part of every engagement, and stewardship has no end date. If AI won't pay for itself somewhere in your business, we'll tell you.
Questions
How many AI consultants should we talk to before choosing one?
Talk to two or three, and ask each one the same questions in the same order. That makes the answers comparable and shows quickly who has thought about measurement, ownership and life after launch. Give each the same short description of one workflow you want to improve. How they respond to a real problem, with questions or with a pitch, tells you more than any slide deck.
Should we hire an AI consultant who specializes in our industry?
Industry knowledge helps, but it matters less than whether the firm will learn how your company actually works. Every business in an industry runs its approvals, data and customers a little differently. Ask how they will map your real workflows before proposing anything, and ask to see working software on data from your industry. A firm that does both can earn the knowledge quickly. A label alone proves little.
What is the difference between an AI consultant and an AI implementation partner?
An AI consultant typically advises: what to do, in what order and with which tools. An AI implementation partner also builds the systems, brings the team along and maintains the work after launch. Many firms use both words. Ask who will write and connect the software, and who will look after it once it is live. The answer tells you which kind of firm you are talking to.
Is it a red flag if a consultant won't quote a price on the first call?
No. Price depends on scope, and a careful firm needs to understand your workflows, systems and data before it can price the work well. It is a red flag if you still have no clear price, scope and deliverables before work begins, or if the quote leaves out upkeep after launch. A fixed-fee first step with defined outputs is a reasonable thing to ask for.
Do we need a technical person on our side to work with an AI consultant?
No, but you need an owner. A good outside team handles the technical work and explains it in plain terms. What it can't supply is a leader inside your company who has authority to make decisions, a few hours of time from the people who do the work, and access to the systems involved. Without an internal owner, even good work tends to stall.