Choose an AI consulting firm by asking who will actually build the work, what problem they'll solve first and how success is measured, how it fits your existing systems, where your data goes, what you own at the end, and how you could leave. A firm that answers plainly, in writing, is worth a closer look.
Why does choosing the right AI consulting firm matter so much?
Because many AI projects fail, and the most common cause starts before the build. RAND's 2024 study of why AI projects fail, based on interviews with 65 data scientists and engineers, noted that, by some estimates, more than 80% of AI projects fail, and found the most common root cause was misunderstanding the problem the AI was meant to solve.1 Gartner predicted in 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs or unclear business value.2
Those are problems a good partner should help you avoid. They are also hard to see in a sales pitch, which is why the questions below focus on how a firm works.
Be careful with labels, too. Gartner estimated in 2025 that only about 130 of the thousands of vendors marketing agentic AI were offering real agentic capabilities.3
What are the 8 questions to ask an AI consulting firm?
Who builds it, what you'll solve first and how you'll measure it, whether it fits your systems, where your data goes, what you own, how you'd leave, how risk is managed, and whether you can talk to a similar client. Ask every firm on your shortlist the same eight questions, ideally in writing, and compare the answers side by side.
| Question | A good answer sounds like | Red flag |
|---|---|---|
| 1. Who will actually build it? | Named engineers who scope, build and support the work | Senior people sell it, then hand it to someone else |
| 2. What will you solve first, and how will we measure it? | One workflow, with a baseline and a target | "We'll explore use cases" with no measure of success |
| 3. Will it work inside our existing systems? | Builds on the tools your team already uses | Requires replacing core systems to get started |
| 4. Where will our data go? | A clear map of where data is processed, stored and logged | Vague assurances, or "it's all in the cloud" |
| 5. What do we own at the end? | You own the code, the data and the configurations | Your workflows only run on the firm's platform |
| 6. What happens if we want to leave? | Documentation, data return and a transition plan | No exit terms, or long lock-in |
| 7. How do you manage AI risk? | Human approval for consequential steps, logging, testing | "The model is very accurate" |
| 8. Can we talk to a client like us? | A reference in a similar industry and size | Only logos, or case studies with no specifics |
Why ask who will actually build it?
Because the people in the sales meeting are not always the people who do the work. Ask for the names and roles of the team, how many other clients each engineer is working with, and who is on call when something breaks after launch.
Also ask what happens after go-live. AI workflows need monitoring and adjustment as documents, rules and models change. If a firm plans to hand off at launch, ask who will monitor and adjust the work after that. We compared the two models in AI-native vs. traditional consulting.
How should a firm define success before building?
With one workflow, a baseline and a target. "Cut intake time from three days to one" is a goal. "Explore AI opportunities" is not.
This is where RAND found projects most often go wrong: stakeholders misunderstand or miscommunicate what problem the AI is meant to solve.1 A good firm will spend real time with the people doing the work before it proposes anything, and it will tell you when a simpler tool or no AI at all is the better answer.
What should you ask about data, ownership and exit?
Ask where your data goes, who owns what gets built, and how you would leave. Get the answers in the contract.
Bank regulators give a useful checklist here, and it works for any industry. The 2023 interagency guidance on third-party relationships from the Federal Reserve, FDIC and OCC tells banks to look at whether data the vendor generates becomes the bank's property, whether source code is held in escrow, how the bank's data is returned or destroyed at the end, and how long the vendor must support an orderly transition.4 If you are a bank, these are terms the guidance asks you to consider; see our work with banks and credit unions.
- Data path: where prompts, documents, outputs and logs are processed and stored, and who can access them. For banks, we covered how to keep customer data in-region.
- Ownership: code, configurations, prompts, data and documentation belong to you.
- Exit: data return or destruction, a transition period, and enough documentation that another team could take over.
How should an AI firm manage risk?
With a clear process you can inspect. Ask how the firm decides what the AI may do on its own, where people approve the output, how it tests before launch, and what it logs.
If you want a shared vocabulary, NIST's AI Risk Management Framework organizes the work into four functions: Govern, Map, Measure and Manage.5 A firm doesn't need to follow it to the letter, but it should be able to explain how its process covers each one.
How would Soren AI answer these questions?
Ask us the same eight questions. Here are short versions of five of our answers. Ask us the rest, including references, on a call.
- Who builds it: the whole engineering team, typically 10 or more engineers, not one person assigned to your account.
- First steps: in the first two weeks we interview the people doing the work and agree on a clear scope and goal. The first workflow is chosen for clear, attributable results and is typically live within two to three weeks of getting access.
- Your systems and data: we build into the systems you already use and have deployed in the cloud, on premises and hybrid.
- Ownership: you own everything we build.
- After launch: our engineers stay to monitor, support and improve it. Engagements usually run a year or more.
More detail is on how we work.
Put us through the eight questions
Start with a 30-minute call with our team. Bring your questions and one workflow you'd like to fix, and we'll tell you honestly whether we're the right fit.
Frequently asked questions
What does an AI consulting firm actually do?
It depends on the firm. Some deliver strategy, analysis and a roadmap, and someone else builds it. Others embed engineers who build working AI into your systems and support it after launch. Ask which one you're buying, because the deliverable at the end, a plan or a working system, is very different.
Should we hire an AI consultant or build an in-house team?
An in-house team gives you long-term ownership but takes time to hire and ramp up, and hiring experienced AI engineers takes time. A firm can get the first workflows live while you build internal capability. If you hire a firm, make sure your team owns what gets built and learns how it works.
What should be in an AI consulting contract?
At minimum: a defined scope and success measures, who owns the code, data and configurations, where data is processed and stored, security and access controls, support after launch, and exit terms covering data return or destruction and a transition period. Regulated businesses should also map the contract to their third-party risk requirements.
How long should an AI consulting engagement take before we see results?
You should see a working first workflow within weeks if the firm builds rather than only advises. At Soren, the first workflow is typically live within two to three weeks of getting access to the systems it needs. Broader programs take longer, but each step should show measurable results.
Sources
- The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND, August 2024
- Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025, Gartner, July 2024
- Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, Gartner, June 2025
- Interagency Guidance on Third-Party Relationships: Risk Management, Federal Reserve, FDIC and OCC, June 2023
- Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST, January 2023
