The conversation that happens on almost every discovery call
Operator: We need AI. Me: What’s broken right now? Operator: Our intake form forwards leads to the wrong person half the time. Me: That’s not an AI problem.
This exchange happens on almost every discovery call I run. Some version of it. The operator arrived convinced that AI is the answer. They read three “AI for SMB” posts last week and decided the tool is what their business needs.
They haven’t asked the prior question.
The prior question is: where is the business actually bleeding, and would AI move the needle on that specific bleed?
Most of the time, when an operator can articulate the real bleed (and most of them can, once asked), the fix isn’t AI. It’s a routing rule, a captured field, a documented process, a hire. The AI conversation, if it ever happens, comes after those fixes are in place.
But the operator didn’t arrive ready to have the prior-question conversation. They arrived to talk about AI. That’s the work I have to do that I shouldn’t have to.
What the market trained operators to skip
The reason operators arrive at discovery calls with the tool already picked is straightforward: the market trained them to.
Every consultant on the AI-for-SMB feed is selling AI agents. The pitch shape is consistent: add AI to your business and unlock [outcome]. Add AI to your sales process. Add AI to your operations. Add AI to your reporting.
Notice what’s missing from those pitches. There’s no operational diagnosis. There’s no question about what’s broken. The tool is offered before the problem is named. The pitch puts AI ahead of pain.
Operator: I want to add AI to our sales process. Me: What about your sales process is broken right now? Operator: I’m not sure. But everyone’s adding AI.
That second exchange happens nearly as often as the first. The operator can feel the pressure to add AI even without being able to name what AI would fix. That’s a market problem, not an operator problem. The market made AI feel like a prerequisite for staying relevant. The discomfort of not adding AI is being sold as the urgency.
The work of an honest first conversation is to slow that pressure down. Not to talk anyone out of AI. To put AI back in its correct position in the sequence: after the question about what’s actually broken.
Three fixes that aren’t AI
Three operational problems I see across 5-50 person businesses, all of which get pitched AI solutions, none of which are AI problems.
The intake form that misroutes.
A 15-person business has a lead intake form. Half the time it forwards leads to the wrong person. The owner has been pitched an AI lead-router. The pitch is: AI reads the lead, classifies the intent, sends it to the right team member.
The fix is the form. The form has six dropdown fields, half of which are stale, and two free-text fields where prospects describe what they need. The misrouting happens because the form was set up four years ago and the team has changed since. The fix is twenty minutes of form rework and one updated routing rule.
Layer an AI lead-router on top of that form and the AI now classifies leads against stale routing logic. Leads still go to the wrong person. They just get there with more confidence.
The manufacturing shop without a CRM.
A small manufacturing shop tracks customer interactions in three places: the owner’s email, the shop manager’s head, and a printed log on the wall. The owner has been pitched an AI chatbot to handle customer inquiries.
The fix is a CRM. Customer history needs to live in one structured system before any tool, AI or otherwise, can sit on top and answer questions about it. Build the chatbot first and it answers customer questions with three-source disagreement and confident hallucination.
The fix family here is documentation, structure, single source of truth. Boring, infrastructural, unmistakably the right first move. The AI chatbot can come after, if it still makes sense once the documentation is in place. Often it doesn’t.
The service business with seven undocumented handoffs.
A service business has seven distinct handoffs in a typical job lifecycle: estimate, schedule, dispatch, on-site work, completion check, invoice, follow-up. None of them are documented in a system. The team coordinates by text message and tribal knowledge.
The owner has been pitched AI to automate the handoffs.
The fix is documentation. Until the handoffs are named, captured, and structured, automation has nothing to operate on. AI on top of undocumented handoffs accelerates the chaos. The wrong information moves faster. The unprocessed exceptions pile up at machine speed.
The actual fix is unglamorous: write down what happens at each handoff, structure it as a workflow, make sure the team uses it. Then evaluate whether automation makes sense. Often it does. Sometimes the automation is AI. Often it isn’t.
AI sequences in fine. It just sequences second.
The pattern in those three cases isn’t anti-AI. It’s about sequence.
AI works. It works extremely well when sequenced correctly. The lead-router built on a clean form catches misclassifications a human would miss. The chatbot fed by a structured CRM answers customer questions accurately. The handoff automation built on documented workflows actually saves the team time.
The catch is the word “when.” AI works when the operational foundation is in place. AI built on a broken foundation amplifies whatever’s wrong with the foundation, including chaos, contradiction, and bad data. The output looks correct. It isn’t.
So the question on every discovery call isn’t “should we add AI?” It’s “what’s the operational state, and would AI fix this specific bleed, or would a cheaper, faster intervention do it better?”
When the answer is AI: build it. AI is right for certain shapes of problem: high-volume classification on text data the human team can’t keep up with, conversational lookup over a knowledge base that’s actually organized, pattern recognition across data the team can’t manually scan. Those are real wins.
When the answer isn’t AI: don’t build it. Most of the time in 5-50 person businesses, the answer isn’t AI. It’s a process fix that takes a week, costs a few thousand dollars, and produces immediate, traceable ROI.
Process fix first. Automation second. Sometimes the automation is AI. Often it isn’t.
The exchange I want to have
The exchange I want operators to arrive at the discovery call ready to have looks different from the one that opens this post.
Me: What’s broken right now? Operator: Three things. We’re losing about a quarter of inbound leads at the form. Customer history is in three places nobody can search. And we don’t know which jobs are profitable. Me: Okay. Let’s start there.
That conversation produces ROI in weeks. The first one usually produces a six-month AI build the operator eventually pulls the plug on, blaming AI.
If you’re an operator considering AI for your business, the highest-leverage thing you can do before the next consultant call is to name three places your business is bleeding. Not what tool you want to add. What is breaking that costs you money or time. Whatever you write down, that’s your real list. The AI conversation, if it makes sense at all, lives inside that list. Probably attached to one or two items at most. Often to none.
AI joins the conversation when it earns the seat.