A restaurant owner in Asheville paid an "AI agency" $8,000 to build a customer feedback system. What he got was a Google Form connected to ChatGPT with a $20/month API key. The "custom AI solution" was a prompt template that said "summarize this restaurant review."
He could have built the same thing in an afternoon with a YouTube tutorial. He didn't know that. The agency did.
This story plays out constantly. The AI agency market has exploded — thousands of new agencies in the last year alone — and most of what's in it is smoke.
The Gold Rush
Start an AI agency. That's the advice in every business influencer's playbook right now. The pitch goes like this: AI tools are cheap, businesses don't understand them, and you can charge thousands for what amounts to connecting a few off-the-shelf services together.
It's not wrong, technically. AI tools are cheap. Businesses don't understand them. And you can charge a lot for connecting things together.
The problem is that "connecting things together" is not the same as solving a business problem. And the gap between those two things is where businesses get burned.
What Most "AI Agencies" Actually Do
Here's the standard playbook:
Step 1: Buy a ChatGPT API subscription ($20-$200/month).
Step 2: Build a simple wrapper — a chatbot, a form processor, a content generator — using a no-code tool or basic API integration.
Step 3: Add a nice interface and some branding.
Step 4: Sell it as a "custom AI solution" for $5,000-$30,000.
The deliverable costs about $50/month to run. The setup takes a few days. The markup is astronomical.
Is this always a scam? No. Sometimes the wrapper solves a real problem and the business owner genuinely couldn't have built it themselves. The expertise in understanding the problem, choosing the right approach, and making it reliable has real value.
But that's the minority case. Far more common is the agency that sells the technology without understanding the business problem — or worse, without caring whether the technology actually solves it.
The Five Warning Signs
After seeing this pattern repeat across dozens of conversations with business owners, here are the red flags:
1. They Lead with Technology, Not Problems
"We use GPT-4 with RAG and vector embeddings" means nothing to a business owner. And it shouldn't. What matters is: what problem does it solve, and how do you know it works?
A good consultant leads with your problem. They ask questions about your business before they mention any technology. They want to understand the workflow, the pain points, the volume of work, and the cost of doing it the current way.
A bad one leads with buzzwords and hopes you're too impressed to ask what they mean. We wrote a full breakdown of how to evaluate AI vendors if you want a detailed checklist for these conversations.
2. They Can't Explain What Happens When It Breaks
Every AI system breaks sometimes. The model returns something weird. The data has an edge case nobody anticipated. The API goes down.
Ask your potential AI agency: "What happens when the system gives a wrong answer?" If they say "it won't" or "AI is very accurate now" — run. If they say "here's the fallback process, here's how you'll know, and here's how we fix it" — that's a better sign.
The willingness to talk about failure is one of the best indicators of competence. Real production systems need error handling, retry logic, and fallback behaviors. If your vendor hasn't thought about what happens when things break, they haven't thought about production at all.
3. They Won't Show You What's Under the Hood
"Proprietary technology" is the favorite hiding place of thin solutions. If an agency built you a system, you should be able to understand — at least at a high level — what's in it.
That doesn't mean you need to read code. It means they should be willing to explain: what model are they using, what data goes in, what processing happens, and what comes out. If they refuse to explain the architecture, ask yourself what they're hiding.
Often the answer is: there's not much architecture to hide.
4. They Have No Industry-Specific Knowledge
AI for a restaurant is different from AI for a plumbing company is different from AI for a law firm. The technology might overlap, but the business context is completely different.
An agency that claims to serve "all industries" is probably not deeply knowledgeable about any of them. That doesn't automatically make them bad — generalists exist — but it means you should probe harder on whether they understand your specific workflows, regulations, and customer expectations.
5. The Pricing Doesn't Match the Complexity
Here's a rough reality check:
- Connecting a chatbot to your existing FAQ: $500-$2,000. If someone charges $15,000 for this, they're overcharging.
- Automating a multi-step workflow with business rules, data integration, and error handling: $5,000-$20,000. This is real work.
- Building a custom AI system that integrates with your operations, handles edge cases, and includes monitoring and maintenance: $15,000-$50,000+. This is genuine engineering.
If the price is high but the deliverable is simple, you're paying for the word "AI," not for the work.
What Good AI Consulting Actually Looks Like
Real AI consulting starts with understanding your business, not selling you technology. Here's what a good engagement looks like:
Discovery first. The consultant spends time learning your operation. They watch how work flows through your business. They talk to the people doing the work. They identify the specific bottlenecks where automation could help.
Honest assessment. Sometimes the answer is "you don't need AI for this." A good consultant will tell you that. A spreadsheet formula, a Zapier workflow, or even just a better-organized Google Drive might solve your problem at a fraction of the cost.
Clear deliverables. You know exactly what you're getting, how it works, and what it costs to maintain. No mystery boxes. No proprietary black boxes you can't inspect.
Maintenance plan. AI systems need ongoing care. Models change, your business changes, edge cases appear. If the agency delivers and disappears, you'll be stuck with a system that degrades over time with nobody to fix it.
Knowledge transfer. At the end of the engagement, your team should understand the system well enough to use it confidently and identify when something's wrong. If the agency is the only one who can operate the thing they built, you're dependent on them forever. That's a feature for them, not for you. (This is why we wrote about packaging your business knowledge — when you own the knowledge layer, you're never locked in.)
The Uncomfortable Question
If you've already hired an AI agency, ask yourself this: could you describe, in one paragraph, what the system they built actually does and how it works?
If you can, great. You probably got real value.
If you can't — if it's a black box that you pay for monthly and hope it keeps working — it might be worth asking for a plain-language explanation. What you learn might surprise you.
How We Think About It
We're a software consultancy. We build things for non-technical businesses. Some of those things involve AI. A lot of them don't.
Our approach is simple: we start with the problem, figure out the simplest thing that solves it, and build that. If AI is the right tool, we use it. If a script and a spreadsheet would do the same job, we recommend that instead — even though it means a smaller invoice.
That's not altruism. It's strategy. A business owner who pays $2,000 for a solution that works will come back when they need something bigger. A business owner who pays $15,000 for a solution that doesn't work won't come back at all.
The best way to tell a good consultant from a bad one is the same way you tell a good mechanic from a bad one: the good one tells you what you don't need replaced.
Want a second opinion on an AI solution you've been pitched? Our team will give you an honest assessment of whether it's worth the investment — no charge for the conversation. Get in touch.
Related Posts
Stay Connected
Follow us for practical insights on using technology to grow your business.