What I ask on every first call
A look inside the 20-minute discovery call: what I listen for, the one question that screens about a third of leads, and when I say no.
I’ve done dozens of these calls. They run about 20 minutes. By the end I know whether I can build something useful, whether the economics make sense, and whether this is going to be a good deployment.
Here’s how every one goes.
The first five minutes are just listening
I open with: “Walk me through how you handle [calls / inquiries / bookings] today.”
Not “do you have a process” — I want them to narrate it from the beginning, unprompted. What they mention first tells me what’s actually painful. What they skip over tells me where they’ve gone numb to friction.
If they describe their process in precise detail — “a lead comes in, I call back within the hour, if I miss them I send a text template from my phone” — that’s a good client signal. They know their operation. Agents built on top of known operations work. Agents built on top of “we handle it somehow” do not.
The question that screens about a third of leads
Somewhere in the first ten minutes I ask: “Are you comfortable with the agent occasionally getting something wrong or incomplete?”
The right answer is some version of “yes, as long as I can catch it” or “for routine stuff, yeah, that’s fine.”
If someone says “no — it has to be right every time, no exceptions,” I end the call politely. Not because they’re wrong to want accuracy. But AI is a capable junior assistant, not a guaranteed-correct system. If the first bad answer creates a crisis — a patient gets the wrong pre-op instruction, a client gets a confidential detail mixed up — we shouldn’t deploy one.
This one question saves both of us from a deployment that fails three months in.
What I’m mapping by the midpoint
By the ten-minute mark I’m building a picture of four things:
Volume. Is there enough daily friction to justify the cost? Five customer inquiries a week doesn’t support a $4,000 Telegram agent — the math doesn’t work. Fifty a week is a different conversation.
Ownership. Who handles the exception cases? Every agent escalates things it can’t resolve. Someone has to be the human-in-the-loop — the person who gets the flag and makes the call. If there’s no one, the deployment decays within six months. Not because the agent broke, but because nobody was watching.
Stack. What tools are already in place? Google Calendar, Calendly, HubSpot, a booking system, a property management portal — the integrations change the deployment scope and the price. I need to know what’s there before I can quote anything.
Risk profile. How bad is a wrong answer? A realtor missing a follow-up text is annoying. An attorney with a confidentiality breach is a crisis. Same underlying AI, very different guardrail requirements, very different deployment shape.
When I say no
Three situations I don’t take on:
No clear use case. “Can it do everything?” is not a use case. I need the specific job — intake, scheduling, after-hours coverage, dispatch, follow-up. Broad scope produces a broad deployment that helps no one and frustrates everyone.
No defined owner. If nobody’s going to review the agent’s escalations and spot-check its logs once a week, the deployment drifts. Prompts go stale. The agent starts giving answers that made sense four months ago but not now. Someone gets a bad answer and blames the tool — correctly.
Budget misalignment. If someone’s expecting to spend $200/month on what’s a $4,000 build, we’re not solving the same problem. It’s not a fit issue, it’s arithmetic. Better to surface that on the call than in the proposal.
After the call
If it’s a fit: I send a short written proposal within 24 hours — what I’m building, what it costs, what the handoff looks like. The handoff matters as much as the build; you walk away owning the deployment, not renting access to it.
If it’s not a fit: I say so during the call. No “let me think about it and follow up.” You walked into this call with a real problem — you deserve a clear answer.
Most people who’ve already worked through the five pre-deployment questions before we talk arrive knowing most of this already. The call is shorter. The proposal is sharper. That post exists because I kept covering the same ground in the first five minutes of every call.
