How to choose an AI consulting firm: 10 questions to ask before you hire

To choose an AI consulting firm, look for proof that it can build working software on data like yours, test that with a small proof of concept, and agree on ownership, success measures and post-launch support before you sign. The ten questions below cover each of those, in the order most buyers need them.

AI consulting is crowded. Large strategy firms, IT services companies, software agencies, specialist AI firms and independent consultants all offer AI services, and their websites say much the same things. The differences only show when you ask specific questions and look for specific evidence.

We are an AI consulting firm ourselves, so take this guide as the list we would want a careful buyer to use on us. Every question here is one we expect clients to ask.

Start with what you need

Before you talk to any firm, write down what you want to change. Not "use AI", but a specific problem: bids that take too long to prepare, invoices that nobody has time to check, leads that go cold, forecasts that are always off. Note who does that work today, which systems and data it involves, and how you would know it had improved.

A one-page brief like this does two things. It lets you compare firms on the same problem, and it shows you quickly which firms listen. A good firm will ask more questions about your brief than it answers about itself.

If you do not yet know where AI would help most, that is fine too. Then you are hiring for discovery first, and the questions about how a firm finds and sizes opportunities matter most. Our AI implementation roadmap explains what that first phase should produce.

10 questions to ask an AI consulting firm

Use the scorecard below as you go. Rate each firm on what it showed you, then compare. Must-have criteria count three times.

1. Can you show us working software you have built?

This is the most important question, because it separates firms that build from firms that advise. Ask for a live demo of a system the firm built, not a slide or a video of a prototype. Ask what it does, who uses it, how long it has been running and what broke along the way. A firm that has shipped real systems will answer in detail. A firm that mostly produces strategy will steer back to frameworks.

2. Have you solved a problem like ours before?

Look for experience with your type of problem more than your exact industry. Reading documents, extracting data from them and drafting responses is similar work whether the documents are bids, contracts or invoices. Forecasting demand is similar across retail and manufacturing. Ask for case studies with the problem, what was built and what changed, and notice whether the firm can explain the hard parts.

3. Who will actually do the work?

Many firms sell with senior people and deliver with junior ones. Ask to meet the people who will design and build your system, and ask how much of their time your project will get. Find out whether any of the work is subcontracted, and to whom. The people in the room during the sale should be the people in the room during the build, or you should know exactly who replaces them.

4. How will you decide what to build first?

A good firm does not start by building. It starts by understanding the workflow, then ranks opportunities by the value at stake and how feasible they are, and proposes a narrow first step. Ask how the firm sizes value, what data it needs to do that, and how it decides when an idea is not worth pursuing. Be wary of a firm that agrees with every idea you bring. Part of its job is to tell you which ones will not pay off.

5. Will you prove it on our data before we commit?

A proof of concept built on your own data, in a few weeks, is the best test of whether a firm can deliver. It shows whether the approach works on your real documents and records, which are always messier than demo data, and it shows you how the firm works day to day. Ask what the proof of concept will include, how success will be judged, what it costs and what happens if it does not work.

6. How will you handle our data, and who owns what you build?

Ask where your data will be stored and processed, which AI model providers will see it, and whether any of it will be used to train someone else's models. Ask how access is controlled and logged. Then settle ownership in writing: the code, the trained models, the prompts, the evaluation data and your own data. You should be able to keep using everything built for you if the relationship ends.

7. How will we measure success?

Agree on the measures before the build starts, tied to the problem in your brief: hours saved on a process you already track, revenue from new leads, errors caught, time to respond. Ask how the firm will measure the baseline, how it will test the system's accuracy before launch, and how it will report results after. A firm that cannot say how it will measure success is unlikely to deliver it. This is also the most common reason AI pilots never reach production.

8. What happens after launch?

AI systems need ongoing care. AI models are retired and replaced, data drifts, and new cases appear as the business changes. Ask who monitors the system, how quality problems are caught, how changes are handled and what support your team gets. If you plan to run the system yourselves, ask what the handover includes: documentation, tests, training and time working side by side.

9. How do you price, and can you price it differently?

Ask the firm to separate the build, ongoing maintenance and usage costs such as AI model fees, and to estimate the total over three years. Then ask whether it would structure the price another way: a build fee alone, a build fee plus maintenance, or lower fees plus a share of results. A firm confident in its work will often take part of its fee from the outcome. Our guide to AI consulting pricing models compares the options.

10. Can we speak to a client?

Ask for a reference whose project resembles yours, and ask that client what the firm got wrong and how it responded, whether the system is still in use, and whether they would hire the firm again. A reference that only praises the firm's people, and says little about the system, is worth less than one that describes what was built and how it is used.

Red flags when hiring an AI consultant

Some warning signs show up early. Any one of these deserves a direct follow-up question:

• Only slides and demos on sample data. No working system you can see, and no client still using one.
• A large upfront commitment before any proof. A long contract or a big build fee with no proof of concept on your data first.
• Promises without conditions. Guaranteed accuracy or savings before the firm has seen your data.
• Vague answers on ownership. Code, models or prompts that stay with the firm, or a contract that is unclear about it.
• A tool looking for a problem. A firm that leads with one platform or vendor and fits every problem to it.
• No plan after launch. No monitoring, no maintenance, and no handover beyond a final meeting.
• A different team after the sale. The senior people disappear once the contract is signed.

Large consultancy, specialist firm or freelancer?

Each type of AI consulting firm suits different work. The right choice depends on the size of the change and how much building it involves:

• Large consultancies suit enterprise-wide programs: AI strategy across many business units, operating model changes and large transformations that need many people at once. They bring breadth and senior access. They also bring higher rates, longer timelines and, often, a gap between the strategy team and the people who build.
• Specialist AI firms suit building specific systems: an agent that handles a workflow, a model trained on your data, a tool built into the software your team already uses. They are usually faster and more hands-on, with the same people advising and building. Check that they can support the system after launch and that their experience matches your problem.
• Freelancers and independent consultants suit small, well-defined pieces of work: a prototype, an integration, a review of an existing system. They are flexible and cost less. The risk is continuity: a single person may not be available to maintain the system, and may not cover everything from data to design to deployment.

Some companies also ask about AI agencies, which usually means a software or marketing agency that has added AI services. Judge them the same way: by working systems they have built, and by who will do the work.

Before hiring anyone, check whether the problem is already solved by a product. If a subscription does the job, it will usually be cheaper and faster than a custom build. Our guide to custom AI versus off-the-shelf tools explains how to tell.

How to run the selection

1. Write the brief: the problem, who does the work today, the systems and data involved, and how you will measure improvement.
2. Shortlist three firms whose case studies match your type of problem.
3. Hold one working session with each, on your brief, with the people who would do the work. Score each firm on the ten questions.
4. Call references for the top two, focusing on what was built and whether it is still used.
5. Start with a proof of concept on your data, with success measures and a decision point agreed in advance.
6. Then commit to the build, with ownership, support and pricing settled in the contract.

How DeepVelocity works with clients

DeepVelocity is an AI product development and consulting company based in New York. We build and run our own AI products, including VelocityBids and QuarterHour, so we know what it takes to keep an AI system working after launch. Our founders came from McKinsey & Company, and the same people who scope the work build it.

Every engagement follows the same path: discovery, sizing the value at stake and choosing a starting point, a proof of concept on your own data, a production build, and then maintenance and change management so the system keeps up with your business. You can see examples in our case studies, such as industrial bid intelligence and an AI-assisted quote workflow, or read more about our custom AI solutions and who we are.