Week 0 · baseline
Measure current live-answer rate, median speed-to-lead, and booked rate.
Field Product Manager · New York
“Discovery is easy. Coordination is hard. Commitments are harder still.”
The problem you are building aroundThis is not a note about an engineering seat. It is one page showing how I find the real problem, size the number that moves, and prove it cheaply. That is the muscle a field pilot actually runs on.
A field-discovery teardown
The vertical is a deliberate composite, drawn from a real small-business pattern. The point is not HVAC. It is the method, and every number below is a hypothesis to be tested in the pilot, not a claim.
Owner-operated. Roughly forty inbound leads a week from calls, web forms, and Google Business Profile. Peak demand lands on the hottest and coldest days, exactly when the office cannot pick up every call.
“I want an AI receptionist that answers the phone.”
On peak days and after hours, a large share of inbound leads hit voicemail, and a homeowner with no heat calls the next company on the list. The bottleneck is speed-to-lead, not headcount. An AI receptionist is the request. Recovered revenue from unanswered demand is the job.
I would not build off one owner's complaint. I would call three more shops in the same trade and ask two questions: what percent of inbound never gets answered live, and what is a booked job worth. If the answer clusters across shops, it is a market pattern worth a product bet. If it is one disorganized owner, it is a services fix, not a product.
An agentic intake agent that answers every inbound lead in under a minute, around the clock.
Three weeks. Low cost. Reversible.
Measure current live-answer rate, median speed-to-lead, and booked rate.
Route after-hours and overflow only to the agent.
Incremental booked appointments and revenue attributed to recovered leads.
Escalation accuracy and homeowner satisfaction, so we do not book junk.
Incremental booked revenue does not clear cost, or the owner spends time cleaning up mistakes.
Net-positive and trusted enough to hand over all hours. Then add the follow-up agent and the next trade that showed the pattern.
The owner never sees a model. He sees three more booked jobs this month and a nightly text that reconciles to his bank deposits. That is the only proof that matters to a small business, and it is what turns a pilot into a paying, sticky customer.
The part that generalizes
Swap the vertical and the one-number math changes. The method does not.
The stated request is a symptom. The job is the thing underneath it that actually moves money.
Name the single metric the business already tracks, and make the whole pilot rise or fall on it.
The smallest reversible experiment that could show the effect, with an honest kill line drawn before starting.
Results the owner can reconcile against real bank deposits. That is what earns trust.
Point me at a vertical Blue cares about and I will run this on it. That is the field work, and it is the part I actually enjoy.