The AI consulting industry has a fraud problem and nobody wants to name it.

I got scammed by an AI development firm two years ago.

Not the kind of scam where the money disappears and the emails stop. The more common kind. The kind where you pay a firm real money, they send you status updates that sound professional, they demo components that appear to work in controlled settings, and then when you actually try to use the product, it becomes clear that most of it doesn't exist. The codebase is half-vibe-coded and half-empty. The suppliers aren't being paid. The "team" of four developers is actually one offshore contractor and a project manager who learned the domain the week you signed the contract.

I rebuilt the product myself in three weeks. That's not a brag. It's how I know the firm I hired could have done it too, if they'd actually been trying.

This happens constantly in AI consulting. The category has a fraud problem, and the people running legitimate firms are reluctant to name it because naming it sounds like self-promotion. It isn't. It's a public service. Every client who gets scammed ends up convinced that AI consulting as a category is a scam. That's bad for the clients, bad for the firms doing real work, and bad for the buyers who now hesitate to try again.

Here's what the fraud looks like from the outside, so you can recognize it.

The first signal is scope inflation during sales. The firm's capability deck shows full-stack engineering, machine learning, cloud architecture, mobile, design. The sales call confirms they do all of it with their in-house team. The proposal lists team members you'll never meet. After signing, the project gets assigned to a subset of that team who handle your account specifically. That subset is smaller than the capability deck suggested. Usually much smaller.

The second signal is the "discovery call." A real discovery call is them learning your firm. A fake one is them learning your industry. If the firm is asking you to explain what your business does, what an RFP is, or what a Gantt chart is worth, you are paying them to go to school. That's fine if you accept it upfront. Most firms don't say that's what's happening.

The third signal is the GPT wrapper. The product they're building is a custom AI system, they said. The engineering is proprietary, they said. What you get is a chat interface connected to GPT-4 with your company's name at the top and a system prompt they'll tell you is "carefully tuned." The actual tuning is three paragraphs of instructions that anyone could have written in thirty minutes. The model is a foundation model they don't control. The "data pipeline" is copy-paste. You are paying custom-build prices for a skin on someone else's API.

The fourth signal is the contractor network masquerading as a team. The firm's website shows headshots of senior engineers. The LinkedIn profiles connect to the firm. The actual work is done by a rotating cast of contractors, some of whom are also working on three other projects simultaneously, some of whom have been with the firm for six weeks, none of whom were part of the sales conversation. When you ask specific questions about trade-offs, you get vague answers because the people answering are not the ones making the decisions.

The fifth signal is the status report that describes process, not progress. "This week the team made significant progress on the data ingestion pipeline and aligned on key architectural decisions." Translated: they had meetings. No specific features shipped. No milestones hit. No functional demos. When you push for a demo, you get a walkthrough of a mockup, not a working system.

The sixth signal, the fatal one, is the deadline slip without root cause analysis. "We're going to need another two weeks due to technical complexity we discovered during implementation." There's no specific cause named. No affected component identified. No change in scope justifying the delay. Just vague complexity. This is the signal that the project is in trouble and the firm is managing you, not the work.

If you see three of these signals, you're probably being scammed. If you see five, you definitely are.

The reason this industry can support this kind of fraud is that buyers typically can't inspect the work. AI systems are opaque. Custom builds are inspectable only if you have in-house technical talent, which most mid-market firms don't. The firms doing fraud rely on the fact that by the time the client figures it out, they've sunk enough into the project that they'd rather pay extra to finish than start over with someone else. That sunk cost is the core of the business model.

The only defense is skepticism during sales, specificity during scoping, and a willingness to walk away the moment the signals start appearing. The cheapest scam to survive is the one you identify in month one, before the sunk cost traps you.

I know this because I didn't do any of those things. I hired the wrong firm, trusted them for too long, and then had to rebuild from scratch. That experience is why BottleneckLabs.ai operates the way it does. No contractor network. Full-time employees only. Proposals detailed enough to be shared publicly. Methodology visible. Pricing tied to specific outcomes. Every signal the scam firms hide, we show.

The fraud problem in AI consulting isn't going to self-correct. The money is too good and the clients are too trusting. The only thing that helps is naming the patterns out loud so buyers know what to watch for.

That's why this is in writing.