AI consultant red flags

Why red flags matter more with AI

AI projects are easy to demo and hard to run. A proposal-drafting tool can look impressive in a 20-minute meeting and still fail in practice because it pulls the wrong rate card, nobody reviews its output, or staff go back to the old template after two weeks. Because demos are so convincing, the warning signs tend to show up in how a consultant talks about process, measurement and responsibility rather than in the technology itself.

None of the signs below proves someone is dishonest. Several are just habits of inexperienced providers. But each one raises the risk that you pay for something that does not change your numbers.

It also helps to remember that you are the expert on your business. A consultant knows the tools, but you know which clients are sensitive, which staff are stretched and which mistakes would cost you a relationship. If a proposal does not reflect that knowledge, it was probably written before anyone listened.

Red flags in the sales conversation

One more sign deserves attention: a consultant who cannot explain their recommendation in plain words to your office manager. If the people doing the work cannot follow the plan, they will not adopt the result, however clever it is.

• Promised results. Phrases like you will save 20 hours a week, offered before they have seen your workflow. Ask instead how they would measure time today and what range they would consider a success.
• Tool first, problem second. They recommend a specific platform or agent product in the first meeting. Ask what they would need to learn before choosing.
• Every answer is a build. They never mention settings in software you already own, such as automations inside HubSpot, Clio or ServiceTitan.
• Pressure to sign quickly, such as a discount that expires this week.
• Vague credentials. Lots of logos, but no named systems they have configured or migrated.

Red flags in the proposal

A good proposal is short and specific. For a medical practice automating appointment reminders, it would name the scheduling system, the reminder channels, the no-show rate today, the target range, the opt-out handling, and who checks message wording before launch.

Watch the language as well. A proposal full of terms like autonomous agents, end-to-end AI or digital workforce, with few mentions of your actual systems and staff, is usually describing the consultant's product rather than your problem.

• No baseline. Nothing in the document says how the current process performs today.
• No acceptance criteria. Success is described as the system is live rather than specific tests with real examples.
• Open-ended scope. Hours are billed with no deliverables or end date.
• Running costs missing. The build price is clear but software subscriptions, AI usage fees and maintenance are not.
• No human review step. Client-facing output goes out automatically with no named approver.

Red flags around data and ownership

These matter most because they are hardest to fix later. Be wary if the consultant wants to build in their own Zapier, Make or OpenAI account and give you access, rather than building in accounts your company owns. When the relationship ends, your automations may go with them.

Be equally wary of vague answers about where your data goes. Ask which AI services will see client information, whether those services train on it, and what agreements are in place. A practice handling patient data needs to know whether a HIPAA business associate agreement covers each vendor. A law firm needs to consider confidentiality and ABA Formal Opinion 512. If the consultant brushes these off as paperwork, keep looking.

Also ask how prompts, instructions and configuration will be documented. In AI projects, much of the value sits in carefully written instructions and test examples. If those live in the consultant's private notes, you will struggle to maintain or improve the system once they leave, and you may end up paying again to rebuild work you already funded.

Red flags after launch

Adoption is where many projects quietly fail. If a consultant's plan ends on launch day, the old process will often creep back within weeks.

Ask before signing what support looks like after launch: who you call when something breaks, how quickly they respond, and what it costs. An automation that fails silently at a busy time of year can do more harm than having no automation.

• No training plan for the people who do the work.
• No internal owner named for each tool or automation.
• No usage review at 30 or 60 days.
• Documentation lives only in the consultant's head.
• Problems are fixed quietly instead of logged and explained.

How to check a consultant's claims

You do not need technical knowledge to test what a consultant tells you. A few simple checks catch most exaggeration.

Notice how the consultant reacts to these questions. Experienced providers expect them and answer plainly. Defensiveness, changes of subject or answers that rely on jargon are themselves useful information. You are about to trust this person with client data and staff time, so a little scrutiny is reasonable.

• Ask to see a live example, not a recording, of something similar they built, and ask what broke in the first month.
• Ask the reference whether the promised result was measured, and how.
• Ask what a specific tool costs per month at your volume, then check the vendor's own pricing page.
• Ask which of their recommendations you could do yourself with settings you already own.
• Ask what they would do if the pilot did not meet its targets.
• Ask who else will have access to your data, including subcontractors.

What good looks like instead

The positive version of each red flag is simple. A good consultant asks about your process before naming tools, measures before building, writes acceptance criteria you can test, builds in your accounts, explains data handling plainly, includes a review step wherever output reaches a client, and plans training and a usage check after launch. They will also tell you when the right answer is a settings change, a cancelled subscription, or nothing at all.

InTheory tries to work this way, with a written scope and baseline before kickoff, but the checklist applies to anyone you are considering. Print the red flags, take them into each conversation, and note which ones come up.