Red flags when hiring an AI consultant

We do this work, so a list like this is awkward to write and more useful because of it. Three of the flags below could be pointed at us on a bad day, and I have marked them.

They talk about the technology before they ask about your business

The most reliable signal there is. A consultant who opens by explaining what large language models can do has told you they have one solution and are looking for a place to apply it.

The competent version of this conversation is boring. It is a lot of questions about how work moves through your company, where things get stuck, who does what twice. The technology comes up late, if at all, in the first meeting.

No diagnosis, but a proposal anyway

If you receive a scoped proposal with deliverables after one call, they are selling a template. Nobody can know what your company should automate after forty minutes, and pretending otherwise means the recommendations were decided before you spoke.

Everything is a use case for the thing they already built

Watch for the shape where every problem you describe maps onto the same solution. Ask directly: what have you told a client not to automate? A consultant with no answer has never turned down work, and someone who never turns down work is not diagnosing anything.

They cannot explain how it fails

Every system in production fails in ways you should know about in advance. Wrong outputs, edge cases, the thing that happens when a customer says something nobody anticipated. If the answer to "what breaks" is a reassurance rather than a list, they have not run anything at scale.

No numbers attached to any recommendation

"This will save time" is not a finding. A recommendation should carry an estimate of what it is worth and what it costs to run, even a rough one, because the whole point of the exercise is deciding what to do first. Without that you have received an opinion, not a roadmap.

The deliverable is a slide deck and nothing else

Ask what you physically own at the end. A presentation is not a plan. What you want is something specific enough that a different vendor could build from it, which brings us to the next one.

You cannot take the work anywhere else

If the roadmap only functions if they build it, that is not a roadmap, it is a sales document with a diagnosis attached. The test is simple: could a competent third party execute this without them? If not, ask why not.

Vague about who does the work

Common and expensive. You meet a senior person, you sign, and the work is delivered by someone three years out of university. Ask who specifically will be in the room, and put the names in the agreement.

They will not talk about your data

Any real engagement touches your systems and your customer information. A consultant who has not raised where data goes, who processes it, what gets retained, and what happens at the end has either not thought about it or does not want you thinking about it. That conversation should come from them, unprompted.

Urgency that comes from them rather than from you

Pressure to sign this quarter, a discount that expires, a warning that competitors are moving. Your AI timeline has nothing to do with their sales cycle, and a firm with a real pipeline does not need to rush you.

Certainty about your results

Anyone promising a specific improvement before looking at your operation is guessing, and the confident version of a guess is a lie. The honest position before a diagnosis is a range and a set of assumptions.

The three that apply to us

We have a product. We build AI agents, so we have a genuine incentive to find work our agents can do. The mitigation is that the roadmap is yours to take elsewhere and we say so in writing, but you should still weight our recommendations knowing we sell a hammer. Ask us what we would recommend if we did not build anything.

We are small. That means senior attention on your project, which is real, and it also means less capacity and no bench if someone is unavailable. Ask us what happens if the engagement doubles in size.

Our track record is short. Rapled is new. We would rather say that than dress it up, and the correct response is to weight what we demonstrate over what we claim. Make us show you a working thing rather than a case study.

How to use this

Do not treat these as disqualifiers. Treat them as questions, and listen to whether the answers are specific.

The single best filter is asking a consultant to tell you about a project that went badly and what they changed afterward. People who have done the work have a real answer and tell it without much prompting. People who have not will describe a challenge that turned out fine.

If you want the positive version, the traits that predict a good engagement is the companion to this page, and where to actually look for one covers sourcing. What a consultant should be asking you is the more revealing list, and how to verify somebody is legitimate covers the checks you can run in an afternoon.

Our own version of the engagement is on the consulting page, including the part where the roadmap is yours whether we build it or not.