The $118,000 Voicemail: AI After-Hours Intake for an Indianapolis HVAC Company

Reading time: 10 minutes. What you'll see: a different problem than our manufacturing engagement — not slow quotes, but leads that die at night — and how the same method finds, prices, and fixes it.

The company

Redbud Heating & Air is a residential HVAC contractor on the northeast side of Indianapolis — Castleton up through Fishers and Noblesville. Eighteen employees, around $4M revenue, six trucks, a two-person office. Solid Google reviews, steady referral base, busy season chaos.

Their problem isn't finding customers. It's that customers can't find a human after 6 PM. An AC dies on a July evening in Fishers; the homeowner calls three companies from the map results. Two go to voicemail. One answers. Guess who gets the $6,000 replacement.

Redbud is the voicemail two-thirds of the time — nights, weekends, and whenever both office staff are on other calls. In home services, that's not a communication problem. It's a revenue leak with a timestamp on every drip.

Phase 1 — Observe (week 1)

The process map for a service business centers on one artery: the inbound request. We pulled the phone system's own logs (every modern system has them; almost nobody reads them) and mapped where requests enter and where they die:

The logs put a number on the leak: 31% of inbound calls were going unanswered — concentrated exactly when emergency-repair intent is highest. Industry estimates put the home-services average in the 25–35% range, so Redbud wasn't unusually bad. That's the uncomfortable part: this is normal. Normal is the opportunity.

Phase 2 — Analyze & Quantify (week 2)

Same discipline as every engagement — every input labeled, then the model runs across the assumptions:

That yields roughly 320 lost bookable jobs a year. At $450 average: about $145,000 in lost revenue; ~$118,000 of it attributable to the after-hours and busy-line windows an intake system can actually cover. At 45% gross margin, the recoverable profit pool is ~$53,000/year — before counting a single replacement sale.

And the honest grid, negative row included:

Year-1 cost in the model: $23,300 ($9,500 build + $1,150/month operation). Payback at the confirmed baseline: under nine months. And the first row is the audit doing its real job — if Redbud's logs had shown a 15% miss rate, the written verdict would have been a $95/month answering service, not our system, and we'd have put that in writing.

Phase 3 — Build (weeks 3–8)

The same two house rules as every build: AI only where it earns its place (the conversation is AI; the emergency triage is deterministic rules, because "should we page the on-call tech" is not a question to leave to a model's mood) and the human stays in the loop — the office reviews every overnight booking each morning, and the system's job is to make that review take four minutes instead of reconstructing voicemail fragments.

One thing we deliberately did not build: outbound anything. No AI sales calls, no automated follow-up sequences. Intake was the leak; intake is the fix. Scope discipline is why the build is $9,500 and not $30,000.

Phase 4 — Measure (from night one)

• Answer rate: target >95% of all inbound, all hours (baseline: 69%)
• Booked-by-AI jobs: counted per week, valued at actual invoice totals — the model's recovery-rate assumption, now a measured number
• Emergency escalations: paged correctly vs. missed — the safety metric, reviewed weekly
• Exception rate: conversations the AI couldn't handle and handed off — the honesty metric
• The 90-day question: is the +$26,400 row coming true? The dashboard answers it; the sensitivity table already says which assumption to check if it isn't.

What this maps to on our published pricing

The exit that costs nothing, same as always: a don't build verdict means you keep the analysis and the phone-log findings, and you owe nothing past the audit.

---

Questions Indianapolis service businesses ask us about this

Does an AI intake system replace my office staff?

No — it covers the hours and overflow they physically can't, and hands them organized mornings instead of voicemail archaeology. Every business we've modeled keeps its people; the system keeps their nights.

What does AI answering actually cost for an HVAC company?

Off-the-shelf AI receptionists run roughly $25–$300/month on published vendor pricing and are worth trying first if your only problem is basic message-taking. A system integrated with your scheduler and dispatch rules — the version that books real jobs — is a build: ours start at $7,500, and the audit tells you which category you actually need before you spend either amount.

We're not in HVAC. Does this apply?

The vertical is scenery. Plumbers, electricians, dental offices, vet clinics, property managers — anywhere inquiries arrive around the clock and a human can't, the same model applies and the same audit prices it for your numbers.

---

Fictional client, real method. To run it on your actual phone logs: book the free scan. Our measurement methodology is published in full, and the companion engagements cover a manufacturer's four-day quotes and a freight brokerage's inbox problem.