Before hiring AI consulting: the questions worth asking
The cheapest stage of an AI project is the one that happens before signing. Questions asked on time cost minutes; missing answers are paid for over months. AI consulting is better evaluated with a prepared questionnaire than with a good hunch, and experienced firms know it: they answer with data, examples, and documents, and they welcome a client who asks. This guide gathers the questions worth asking before hiring AI consulting, grouped by topic, with the signals that reveal solid answers.
Real experience separates itself from declared experience with three direct questions.
First: which projects have you taken to production, and when? Many firms count pilots; few count systems still operating months later. A system in production faced real data, real users, and real maintenance, and that track record can be verified.
Second: can I speak with two past clients? Live references reveal what the written case study omits: how the firm responds when something gets complicated, how it handles scope changes, and how it treats the internal team.
Third: what experience do you have in my industry? Every industry has its data, its regulations, and its rhythms. A firm that knows the sector arrives with better questions and avoids lessons charged at client prices.
Context makes these questions even more important. The McKinsey State of AI 2025 survey reports that 78% of organizations already use artificial intelligence in some function, and the supply of services grew at the same pace. Quality varies widely across firms, and verifying past cases is the main filter.
Method separates firms with craft from those that improvise. The central question: what does the project look like week by week? A solid answer describes phases with deliverables, review points, and criteria for moving forward or adjusting.
It is also worth asking how they handle the pilot. Serious firms propose starting with a scoped case, with metrics defined before building, and deciding expansion with results in hand. A Gartner press release from June 2025 estimates that over 40% of agentic AI projects will be canceled by 2027, citing unclear value among the causes. Asking "how will the value of this project be measured from week one?" attacks that cause head on.
Another useful question: how do you document technical decisions? The choice of model, architecture, and tools should exist in writing with their reasons. That documentation allows auditing, switching providers if needed, and understanding the system two years from now.
Finally: what do you need from my team, and when? Healthy projects declare the internal hours they require from the start: interviews, access, validations, and training. A proposal that promises to advance with zero load on the internal team usually ends with systems disconnected from real operations.
Data is the project's raw material, and its treatment deserves explicit questions before any access.
Where will company data be processed? The answer should name regions, cloud providers, and concrete tools. Who will have access during the project, and how is it revoked at the end? Which data leaves company systems and which stays inside?
It is also worth asking about the models: which AI models will you use, under what terms, and do those providers train on the data they receive? The enterprise agreements of major providers typically exclude training on customer data, and the consulting firm should know those terms in detail.
Compliance questions depend on the sector. In healthcare, finance, or regulated services: how do you handle traceability, retention, and human review? Asking for examples of past projects with similar requirements shows whether the answer comes from experience or from theory.
The proposal is signed by leadership; the work is done by a concrete team. Asking who those people will be, with which roles and which dedication, changes the conversation.
Who will be the firm's project owner, and across how many simultaneous projects? Which technical profiles will participate, and in which phases? Is the team that made the sale the same one that will execute?
Seniority deserves a direct question: what share of the work will be done by profiles with production experience? Mixed teams are normal and healthy, and transparency about the mix allows judging the price fairly.
A question that reveals work culture: how do you transfer knowledge to the internal team? Firms that train their client secure future operations; firms that concentrate all knowledge create dependency. Requesting documented transfer sessions as a deliverable turns that difference into a commitment.
Price is understood through its breakdown: how much goes to diagnosis, to construction, and to operation? What assumptions support the figure, and what happens if the data arrives in worse condition than expected or an integration proves more complex?
Timelines deserve the same precision: how long is each phase estimated to take, and which factors could extend it? Estimates with ranges and reasons signal experience; exact promises with zero margin signal commercial optimism.
It also helps to ask what is included and what sits outside: tool licenses, cloud costs, model consumption, additional training, and post-launch adjustments. Well-declared companion costs prevent surprises on the second invoice. The full detail of deliverables and price components is developed in what you get and what drives the price, the companion piece to this guide.
A very useful closing question: what happens if, at the end of the diagnostic, the recommendation is to wait? Firms that earn trust accept that possibility in writing; firms that need to sell the big project avoid it.
Launch is the midpoint of the project, and post-launch operation decides the return. Who will maintain the system when something changes: a model updates, a process shifts, an unexpected behavior appears? Under which service-level agreements and at what cost?
Monitoring deserves its own question: which indicators will watch the system's quality in production, and who receives the alerts? AI systems shift behavior with new data, and continuous watch keeps the quality the pilot demonstrated.
Adoption is also worth asking about: how do you support the team during the first weeks of use? Launch training covers the start; ongoing support turns the system into a habit.
And the long-term question: what does the exit look like? If two years from now the company decides to operate with another provider or with its internal team, do the documentation, access, and knowledge stay on the company's side? The answer defines whether the investment builds internal capability or rents external capability. To understand what these technologies are before evaluating proposals, the guide to AI agents for business offers the base concepts.
Beyond the answers, a firm's behavior during the sale anticipates the project. Three positive signals repeat: specific questions about processes and data before quoting, proposals with deliverables and dates, and transparency about what the project will demand from the internal team.
In the opposite direction, watch for excessive speed: quotes on the first call, scopes that fit in one sentence, and calendars promising production in days. Commercial haste is paid later with technical delays.
Treatment during negotiation also matters. A firm that hears an objection and adjusts the proposal with judgment will show the same attitude when the project gets complicated. One that answers with pressure or lightning discounts is selling something else.
A simple proof: ask for the first phase to be a small, closed diagnostic. Whoever accepts gladly demonstrates confidence in their method; whoever pushes the big contract from day one reveals their priorities.
One more question belongs to the buyer's own calendar: when should this project start for the company to absorb it well? A firm that asks about internal availability, peak seasons, and decision cycles is planning for adoption. A project launched against the company's busiest quarter competes with everything else and usually loses. The honest conversation about timing belongs in the same meetings as price and scope, and the answers often reshape the calendar in ways that save months.
The questionnaire works best sent in writing before the first meeting. Written answers allow comparing firms on equal terms, and the quality of the answers already says a lot: precision against generalities, examples against promises, documents against adjectives.
A final recommendation: involve the people who will operate the system in the interviews. Departments spot questions leadership is unaware of, and their reading of the answers anticipates future adoption.
One additional habit multiplies the questionnaire's value: keep each firm's answers and review them at the end of the project. What was promised against what was delivered is the most honest measure of a provider, and that comparison feeds the next buying decision.
Hiring AI consulting with good questions turns selection into an evidence-based process. The right firm is recognizable because its answers can be verified.
Sources
Was this article helpful?