A counter-intuitive position from a studio whose business is building AI systems.
Most agencies in this category have one job: sell you AI. The pitch is dressed up in different words — automation, intelligence, augmentation, copilots, agents — but the conclusion is always the same shape: yes, you need this, and yes, we should build it for you.
We sell something different. We sell clarity. And clarity, more often than founders expect, ends with us telling them not to build the AI thing they came in asking for.
Here are three real conversations that ended in us recommending against AI.
Case one: the law firm that wanted a chatbot
A mid-sized regional law firm came to us wanting an "AI assistant" on their website to handle initial client inquiries. They'd seen competitors deploy them and felt behind. The pitch was compelling on the surface: 24/7 availability, faster intake, reduced load on the receptionist.
We spent two days with their team mapping the actual intake workflow. What we found: 70% of their inquiries were complex cross-border legal questions where a wrong or even slightly off answer could expose the firm to liability. The receptionist wasn't actually answering anything substantive — she was triaging to the right lawyer within minutes, which clients consistently rated as a high-trust experience.
The right answer wasn't a chatbot. It was a structured intake form that pre-qualified the matter type, captured the urgency, and routed instantly to the right practice area. We built that in three weeks. No AI.
Their intake-to-consultation time dropped 40%. Client satisfaction went up because real humans were still on the other end of the conversation.
Case two: the consultancy that wanted an "AI knowledge base"
A 25-person strategy consultancy wanted to feed all their past project memos into a RAG system so consultants could "ask their own institutional knowledge questions."
We looked at how their team actually worked. The issue wasn't search. The issue was that their senior partners had stopped writing meaningful project debriefs five years ago. The knowledge wasn't lost — it was never recorded. Building a sophisticated retrieval system on top of an empty well would have been theater.
We recommended they reinstate a one-hour project closeout ritual with a structured template. No system to build. Just a process change and a Notion template. Six months later they had enough institutional knowledge captured that a retrieval system actually made sense — and we built it then.
Case three: the recruitment agency that wanted AI candidate screening
A recruitment agency wanted us to build AI-powered CV screening to handle the volume of applications they were receiving. Standard ask. We could have built it in four weeks and billed accordingly.
When we asked why they were receiving such high volume, the answer was: a job board partnership pushing low-quality candidates into their funnel. The CVs weren't a screening problem. They were a sourcing problem. We recommended renegotiating the partnership terms and tightening the application requirements. Volume dropped 60%. Quality went up. Their existing recruiters could handle it without any new system.
Why we operate this way
There's a self-interested version of this answer and an honest one. The self-interested version: founders trust us more after we've told them not to spend money, which makes them more likely to spend it with us later. That's true.
The honest version is simpler. We've watched too many businesses spend six figures on AI systems that solved the wrong problem, and the burned-and-resentful version of that client is worse for the category than no client at all. Every failed AI implementation makes the next buyer more skeptical. We have a stake in those projects being good ones.
There's also an operational point. AI is a multiplier. Multiply a working process by AI and you get a great system. Multiply a broken process by AI and you get a broken process that's now faster, more expensive, and harder to fix. Most service businesses we audit don't have an AI problem. They have a process problem dressed up as an AI problem because AI is in the headlines and processes aren't.
What this means for the audit
Our diagnostic engagement — what we call the AI Opportunity Audit — has a possible outcome that catches founders off guard. Sometimes the recommendation is don't build anything yet, and here's the three-page roadmap of what to fix first.
We charge for the audit either way. Founders pay for the clarity, not for the conclusion. About one in five audits ends in "don't build the AI thing." The other four end with a clear, ROI-scored recommendation for what to build — and usually that thing is smaller and more focused than the founder originally imagined.
If you've been told that AI is the answer and you're not entirely sure what the question was, we should probably talk.