Most AI consulting engagements disappoint, and almost never because the technology failed. They fail for a handful of predictable, human reasons, all of which you can spot before you sign. Here's what to watch for in any partner you talk to, and what good looks like instead.
They sell the technology, not the outcome
The tell is a first conversation that's all about models, platforms, and an impressive demo, and almost nothing about your business. A demo on someone else's data proves the tech can work somewhere. It says nothing about whether it pays off in your operation. The right first conversation is mostly questions about where your time and money actually go, and only then a view on whether AI helps.
If a partner is more excited about the tool than about your problem, you will end up with an impressive tool and an unsolved problem.
They start too big
The transformation deck is seductive and dangerous. A twelve-month, company-wide AI program is a large bet placed before anyone has proof it will pay off in your specific business. When it wobbles, and big programs wobble, you're too far in to stop cleanly.
Real value is almost always proven small first. One high-value use case, built on your real data, measured honestly, before anyone commits to a bigger build. Starting small isn't a lack of ambition. It's how you earn the right to be ambitious without betting the company.
They keep the keys
This is the one that costs the most over time. The work gets built on the consultant's stack, hosted in their environment, in code you never see, with an open-ended retainer to keep it running. It works, but the day you stop paying, it's gone. You were never buying a capability. You were renting one.
Before you sign, ask three plain questions:
- Do we own the code, the models, and the documentation outright?
- Can our own team, or another vendor, run and change this without you?
- If we walk away in a year, what do we keep?
If the answers are vague, the lock-in is the product. A good partner hands everything over and is happy to make themselves replaceable, because they'd rather you stay by choice than because you're trapped.
They oversell and under-measure
“90% automation” on a first deployment is a fantasy, and a revealing one. The same partners who project huge savings up front are usually the ones who never establish a baseline or measure the result, so the savings can never be checked. The two failures travel together: oversell the projection, skip the measurement, and hope nobody does the math.
What good looks like is the opposite: conservative projections, a baseline captured before the work starts, and results reported in real numbers afterward, including the honest cases where the number came in lower than hoped. You want a partner who is more careful with your money than you are.
They ignore the unglamorous part: adoption
A surprising number of AI projects technically work and still fail, because the tool got built and nobody changed how they work. The pilot impresses in a demo, then quietly dies because it didn't fit the real workflow, nobody was trained, and the team didn't trust it.
Adoption is where value actually lands, and it's the least glamorous, most-skipped part of the job. A partner who only talks about building, and never about training, change, and getting people to actually use the thing, is selling you half the project.
They won't tell you no
When AI is the only thing you sell, every problem looks like an AI problem. That's the deepest conflict of interest in the field. A partner who recommends an AI project in every situation, including the ones where the honest answer is “fix your data first” or “this isn't worth it yet,” is optimizing for their invoice, not your outcome.
What good looks like instead
None of the fixes are technical. They're about how the work is structured and who carries the risk. When you're evaluating a partner, look for:
- A bounded first step. A small, fixed-scope, fixed-price engagement that proves value on your real data before any big commitment.
- Honest measurement. A baseline before, real numbers after, conservative projections throughout, and the truth when something doesn't work.
- You own what is built for you. Once it is paid for, you keep the code, the documentation, and a trained team. No lock-in, nothing held hostage.
- Adoption built in. Training and change treated as part of the job, not an afterthought.
- The willingness to say no. A partner who'll talk you out of a bad idea is worth ten who'll cheerfully sell you one.
That's the whole philosophy behind how we work, and it's also just a sensible checklist for hiring anyone to help with AI. Use it on us, and on everyone else you talk to.