A plumber in Phoenix tracked his inbound calls for one quarter. He had been refusing to install call tracking for years on the grounds that “the office handles it.” The office did not handle it. Of 412 inbound calls that quarter, 88 went to voicemail. Of the 88 voicemails, 31 were returned within four hours. The other 57 were either returned the next day, returned never, or returned after the prospect had already booked with someone else.
Plumber’s average ticket: $580. Close rate on inbound calls his office actually answered: 64%. Close rate on calls that went to voicemail and got a callback the next day: 18%.
Do the math forward. He lost 57 prospects at a 46-point delta in close rate. At $580 a ticket, that is $15,200 in one quarter. Annualized, $60,800.
The plumber thought he was losing five or six thousand a year to missed calls. He was off by an order of magnitude.
The pattern is industry-wide. Hiya’s 2026 State of the Call report found that 86% of calls from unknown numbers go unanswered, half of consumers say spam call volume increased in the trailing year, and the average financial loss per consumer to phone scams hit $800. The buyer’s filter on inbound has hardened across every vertical. An owner whose office sends inbound straight to voicemail during peak hours is competing for attention against that filter.
Why the math is always worse than the owner thinks.
Three things stack against the owner that they never see clearly until they pull the data.
The first is that missed calls cluster. Nobody loses calls evenly across the week. They lose them in the four-to-seven evening window when the office has gone home and the techs are still on jobs. They lose them on Saturday morning when nobody is staffed. They lose them during peak season when the office is already running ragged. The annualized number above is averaged. The peak weeks are worse.
The second is that the lost calls are disproportionately the high-intent ones. A homeowner with a hot water heater leaking onto the basement floor at 7 PM does not leave a voicemail and wait. They call the next plumber on the list. The calls that go to voicemail are not a random sample of inbound. They are the calls where the buyer was already half-decided and the next-best option was one tap away.
The third is that the close-rate delta widens the longer the callback takes. A four-hour callback on an emergency lead still books at maybe 25%. A next-day callback is closer to 10%. A two-day callback is essentially zero. We have seen multi-location operators with three-day callback queues during peak season effectively burn every emergency lead that comes in after 5 PM, and nobody on the dispatch side flags it because the office is technically “calling everyone back.”
What we actually deploy.
The fix is a voice agent. Not the kind that reads a script and routes you to a menu. The kind that takes the call live, qualifies the job, confirms the address, books the appointment in your scheduling system, and texts the homeowner a confirmation before the call ends.
We build them on top of the dispatch software you already use. ServiceTitan, Housecall Pro, Jobber, Workiz, the ones that have real APIs and the ones we have built adapters for. The agent answers in the voice of the business. It knows your service area, your pricing tiers, your trip-charge policy, and your “we don’t do mobile-home pier replacements” rule. If the caller asks something the agent doesn’t know, it warm-transfers to the office during business hours or takes a callback request after hours, with a real transcript and a real summary attached.
The agent does not pretend to be a human, and it does not try. When asked, it identifies itself as a virtual assistant. The verbiage and pacing are tuned to the trade. Plumbing intake sounds different from law-firm intake sounds different from medical-practice intake. We document the trade-by-trade patterns under our AI automation service and the specific industry plays under plumbing, HVAC, and the rest of our vertical catalog.
Voice agents in this category are no longer exotic. McKinsey’s State of AI 2025 found that 62% of surveyed organizations are at least experimenting with AI agents, with meaningful adoption now broadly distributed across industries rather than concentrated in tech. Home-service operations sit squarely inside that adoption curve, and the dollar-per-call math above is what makes the engagement quote itself.
What gets qualified, what gets booked, what gets escalated.
The agent runs a three-bucket qualification on every call.
Bucket one: book it. Routine service calls that match your standard pricing tiers and service area. Drain cleaning, water heater service, fixture replacement, scheduled maintenance. The agent confirms the address, checks the calendar, books the slot, and sends an SMS confirmation. The dispatcher sees a fully-formed appointment in the scheduler the moment the call ends.
Bucket two: escalate. True emergencies (active leak, gas smell, no heat in winter), complex commercial work, or anything the agent flags as outside its confidence range. These warm-transfer to on-call during business hours and route to a callback queue with full conversation context after hours. The on-call tech gets the address, the symptom, and the urgency level on their phone before they call back.
Bucket three: nurture. The caller wanted a price quote, ballpark estimate, or general information. The agent gives the standard range, captures the caller’s intent and contact info, and drops them into a follow-up sequence that texts them with a real quote opportunity inside business hours. The follow-up runs across SMS and email until the lead either books, declines, or opts out.
That three-bucket flow is the same architecture we deploy for law firm intake and for medical and dental practice booking, with the qualification logic tuned to each vertical.
The numbers we look for.
Three things tell us whether a voice agent is going to pay back inside the first month.
Inbound call volume above 200 a month. Below that, the math gets thin. The agent still works, but the payback period stretches.
Voicemail rate above 15%. Most businesses underestimate this by half. We pull the actual carrier-side data before we quote.
Average ticket above $200. Lower than that, the close-rate delta still matters, but the absolute dollars don’t justify the engagement unless volume is very high.
If your inbound is below 100 calls a month, the math is usually better solved by working on the visibility side first. The Phoenix plumber’s call volume was high because his local SEO was already strong. The agent paid back because it caught the demand that the visibility work was already creating.
What it costs.
Pricing is set against your actual call data, not a published list. The qualifying math is the payback: on a home-service operation with the volume and ticket-size thresholds above, the engagement breaks even inside the first month. The breakeven on the Phoenix plumber’s calls above was eleven days. We quote after we have pulled the carrier-side voicemail data and modeled the recovered close-rate against your average ticket.
We will not tell you the agent is going to replace your office staff. It is not. It is going to make sure your office staff stops getting interrupted twelve times an hour by callers who only want a quote on a hot water heater install. Office handles the work the office is good at: complex jobs, repeat customers, the regulars who want to talk to the same human. Agent handles the rest. Inbound stops leaking.
Frequently Asked Questions.
How natural does the voice agent sound?
Indistinguishable from a human on routine calls in 2026. The voice models have crossed the line where a caller in the first thirty seconds of a service intake call cannot tell. We disclose when asked, because that is the right thing to do and because trying to deceive callers is bad business. Most callers do not ask.
What happens if the agent gets a call it cannot handle?
It warm-transfers to a live person during business hours and takes a structured callback request after hours, with a full transcript attached. We tune the escalation logic against your actual call transcripts in the first month.
Can the agent handle multiple languages?
English and Spanish out of the box. Other major languages available with a short tuning engagement. Mid-conversation language switching works in 2026 voice models.
Does it work with our existing scheduler?
Yes for ServiceTitan, Housecall Pro, Jobber, Workiz, FieldEdge, and the major dispatch platforms. Custom integrations are built during discovery for stack we have not adapted before, handled by our AI consultancy team when the API surface is unusual.
How long does it take to deploy?
Two weeks to first live pilot on one channel (usually inbound voice). Four to six weeks for full multi-channel coverage (voice, web chat, SMS, follow-up sequences).
If you want to see the agent live before quoting, book a strategy call. We’ll show you the demo on a real account.
