What to Ask an AI Consulting Partner Before You Sign

What am I buying, exactly?

Start here, because the rest of the diligence depends on it. Most AI engagements fall into one of four shapes, and confusing them is the most common way a budget gets wasted.

Engagement shape	What you get	Best when	Fails when
Strategy and roadmap	Analysis, prioritization, a plan	Leadership genuinely disagrees about direction	The plan needs someone to execute it and nobody is assigned
Build	A working system, integration, code	The bottleneck is a missing capability	The tools you own are already unused
Enablement and training	Skills, standards, curriculum	People lack the ability, and the work is otherwise ready	Adoption is blocked by process or permission
Operating change	Redesigned workflows, new decision rights, measurement	The tools work and the return is missing	Leadership will not change how work is assigned

If a proposal spans all four, ask which one it will start with and what the first thirty days produce. A firm that cannot answer that is selling capacity rather than an outcome.

It is worth noting how early most organizations still are. In the US Census Bureau's Business Trends and Outlook Survey, overall AI use by businesses sat between 17% and 20% from December 2025 to May 2026, rising to 37% among firms with at least 250 employees. Anyone claiming a settled playbook for your industry is describing a shorter history than they are implying.

The questions that reveal how a partner works

Question	Why it matters	A weak answer	A strong answer
Who does the work day to day?	Senior people often sell and juniors deliver	Names appear only on the org chart	The delivery lead is in the room and will stay
What does the deliverable look like? Show me a redacted one.	Separates real output from a proposal template	"We tailor everything to the client"	A real artifact, sanitized, handed over
What do you need from us, in hours and in whose diary?	Most engagements fail on client-side capacity	"Minimal involvement"	Named roles, hours per week, in writing
How will we know in ninety days if this is not working?	Creates an exit before you need one	Milestones with no failure condition	A stated check, a date, and what happens if it is missed
What would you do if we did not hire you?	Tests whether they have a view or a product	"You would struggle"	A real first move you could take alone
Which part of this could our own people do?	An honest partner shrinks their own scope	"All of it needs us"	A clear line between what to buy and what to build internally
What happens to your model if the AI tooling changes in six months?	Tool-specific engagements age fast	Deep dependence on one vendor's roadmap	The method survives a tool swap
Who owns the IP, the prompts, the data, and the documentation?	Determines whether you keep anything	Vague, or "industry standard"	Specific, in the contract, favorable to you
What measurement will exist after you leave?	Distinguishes a project from a change	A final report	Instrumentation your team runs
What have you declined to do for a client?	Tests whether they have a scope at all	"We take on most things"	A specific example, with the reasoning

The two questions people skip are the ones about client-side hours and about what happens after the engagement. Both are where the money leaks. An engagement that requires eight hours a week from a director who does not have eight hours will underdeliver for reasons that have nothing to do with the firm's competence, and a change with no measurement behind it reverts within two quarters.

What does good evidence look like from a young firm?

This is where buyers get stuck. Every firm in the category is newer at AI than they sound, and the well-known case studies are mostly from organizations that do not resemble yours.

Evidence worth weighting: work product you can read, a method described in enough detail that you could critique it, a named person who will be accountable, references you selected rather than references you were handed, and a willingness to say what did not work on a previous engagement. Evidence worth discounting: logos without a description of what was done, aggregate percentage claims with no methodology attached, and certifications from the vendors whose tools they will recommend.

Ask for one reference from an engagement that went badly. The reaction to the question tells you as much as the answer does.

When should you not hire an outside partner at all?

Three situations make external help a poor use of money, and a partner who names them unprompted is worth more than one who does not.

If leadership has not agreed on what the business needs AI to do, a consultant cannot manufacture that agreement. The engagement becomes an expensive facilitation exercise and the disagreement returns the week after the final readout.

If the constraint is a decision only an executive can make, buying analysis postpones it. Deciding where recovered capacity goes, or which function owns a workflow, is not work that can be outsourced.

If nobody internally has time to own the change, the engagement produces a document. This one is common and rarely admitted during the sales process, because the buyer is embarrassed to say it and the seller has no incentive to raise it.

The BCG 2026 workforce survey is a useful sanity check on where the leverage sits: clear strategy lifted measurable business impact by around 25 percentage points, while better tools without that strategy and redesign moved it by roughly 5. If a proposal is mostly about tools, the expected return is on the wrong side of that comparison.

The questions that do not flatter us

We would rather you asked these than found out later.

Is a small firm the right risk? A small team gives you the senior people on the work itself and a method that is not diluted through five layers. It also means less bench, less redundancy if someone is unavailable, and no global footprint if you need simultaneous work across regions. If your program needs two hundred consultants in eleven countries, that is a real requirement and we are not it.

Do you need engineering or do you need alignment? Average Robot works on the alignment side: direction, skill, reinvestment, the decisions that determine whether the tools you already pay for produce a return. If your constraint is a data platform that needs rebuilding or a product that needs shipping, a build partner will serve you better, and the two are not substitutes.

What if the answer is that you do not need us for long? Our engagements are bounded on purpose. A ninety-day AI Profit Sprint ends with a plan your team runs. That is a worse commercial model than an open-ended retainer and a better one for a buyer, and it is worth asking any firm where their revenue comes from after the first engagement ends.

Before the first call

Write down the answer to one question: what would have to be true in twelve months for this spend to have been obviously worth it? Write it as a result with a number attached. Six points of margin. A share of pipeline the team cannot reach today. A capability you would otherwise have to hire twelve people to build.

Take that sentence into every conversation and ask each firm how their engagement connects to it. The ones who can trace the line will do it in a minute. The ones who cannot will change the subject to their methodology.

If it helps to have a view on where your own gap sits before you brief anyone, the free AI Profit Readiness Assessment gives you a stage and a first move in a few minutes, and there is no call attached to it.