TL;DR: AI answering for a plumbing shop costs $49 to $249 per month at the SMB tier and roughly $1,000-plus per month at the ServiceTitan-native tier, verified and labeled below. The tools answer after-hours calls, screen spam, and book jobs. None of them publishes an accuracy number for the only call that matters: the 2 a.m. burst pipe. If the bot misroutes that one, the month's subscription bought you a lost $485 emergency ticket and a one-star review. Configure the escalation keywords first, test with 50 real calls, then judge the price.
Search "AI answering service for plumbers" and nearly every result is a vendor ranking itself first. The roundups recycle the same unverifiable claims, and at least one circulating list recommends Numa to home-services shops — a company whose own homepage sells exclusively to car dealerships. That is the quality bar this page is competing with. So here is what we actually did: fetched the pricing pages, labeled every number as vendor-verified, third-party, or unverified, and put the failure mode the vendors skip — the misrouted emergency call — at the center instead of the bottom. We have not run our hands-on suite yet. Every claim below is labeled to match.
Best AI answering services for plumbers in 2026 (at-a-glance)
An AI answering service for plumbers picks up the calls your office misses — nights, weekends, ring-overflow — answers basic questions, screens spam, and books jobs into a calendar or field service platform. SMB plans run $49 to $249 per month, verified July 2026. The ServiceTitan-native tier is quote-only, reported around $1,000-plus per month.
| Tool | What it does | Price (verified July 2026) | The catch |
|---|---|---|---|
| Rosie | AI-only answering: booking, warm and waterfall transfers, mid-call texting, bilingual English/Spanish included. | $49/mo (250 min), $149/mo (1,000 min), $299/mo (2,000 min) — vendor page. Overage reportedly $0.25/min (third-party review; verify on heyrosie.com). | No plumbing-specific emergency-dispatch logic advertised. You build the escalation rules yourself. |
| Goodcall | AI agent with unlimited minutes, billed per unique monthly customer. | $79 / $129 / $249 per agent/mo (100 / 250 / 500 unique customers, then $0.50 each) — vendor page. Annual −15%. | No emergency-dispatch feature named on the pricing page. A Trustpilot reviewer reported reaching an actual human was hard. |
| Smith.ai AI Receptionist | AI answers first, escalates to 500+ North America-based live agents. Per-call billing. | Free $0 (25 calls, then $3.00/call); $150/mo (75 calls); $270/mo (150 calls); $500/mo (300 calls) — extracted from the pricing page markup, July 2026. The page itself is now gated behind a contact form. | User reports of auto-transfers to live agents without consent inflating per-call bills — individual claims, unverified. |
| Ruby | Human receptionists with AI assist, billed per minute. Hold time counts. | $250/mo (50 min), $395/mo (100 min), $720/mo (200 min), $1,725/mo (500 min) — vendor page. Overage rate not published. | Roughly 5x the cost of AI-only per minute. Capterra reviewers report finding competitors at half the price. |
| Avoca AI | ServiceTitan-native AI CSR: answers, qualifies, books jobs directly into ST. | Quote-only. Third-party reports: ~$1,000–$3,000/mo mid-market, $2,500–$3,500 enterprise, per-minute billing — UNVERIFIED against vendor. | Targets $3M+ revenue shops with 5+ CSRs on ServiceTitan or Housecall Pro. Wrong size for a two-truck operation. |
| ServiceTitan Contact Center Pro | ServiceTitan's own AI voice agents for overflow and after-hours booking. | Not published. Third-party report: $400–$600+/mo plus $1,000–$5,000 onboarding, 12-month minimum — UNVERIFIED. | Only meaningful if you already pay for ServiceTitan. Pricing requires a sales call. |
| Retell AI (build-your-own) | The voice-agent platform many white-label "AI receptionists" are built on. | $0.07–$0.31/min all-in, usage-based; phone number $2/mo — vendor page. | You assemble the agent, the booking flow, and the emergency escalation yourself. No safety net. |
Prices checked July 12, 2026, against heyrosie.com/pricing , goodcall.com/pricing , smith.ai/pricing/ai-receptionist (page markup; the visible page is form-gated), ruby.com/plans-and-pricing , and retellai.com/pricing . Avoca and ServiceTitan Contact Center Pro published no price on that date; their figures are third-party reports and labeled as such.
Disclosure: we have no affiliate or business ties to any vendor named here as of publication. If that changes, this paragraph will say so. Our funding model is in our editorial policy .
The 2 a.m. burst pipe is the whole product
Every vendor demo shows the easy call: "What are your hours?" "Do you service Maple Grove?" Those calls were never the problem. The call that justifies the subscription is the one at 2 a.m. from a homeowner standing in two inches of water. That call has three properties that break voice AI: the caller is panicked and talking fast, the background is loud, and the cost of a miss is total. This is the section vendor pricing pages do not have.
The failure mode is documented. A third-party troubleshooting writeup describes an AI agent hearing "burst pipe in the basement" as "first hype in the basement" and responding with service-area information instead of dispatching. The same writeup cites a CTA study putting misrecognition as the top complaint for 68% of voice-AI users — we could not trace that study to a primary source, so treat the 68% as unverified. The anecdote does not need the statistic. Put that error on a call where the wrong response means the caller hangs up and dials the next plumber on Google.
The stakes are quantifiable, with labels. A third-party guide aimed at plumbers puts the average emergency ticket around $485 with a ~62% booking rate — vendor-adjacent marketing math, use it as an order of magnitude, not gospel. Another vendor blog claims AI books 85–95% of after-hours calls versus 40–55% for humans; no methodology is stated, so we treat it as marketing. What is not marketing: a missed emergency call is revenue that goes to a competitor the same night, and neither Rosie nor Goodcall names an emergency-dispatch feature on its pricing page. The capability may exist in configuration; it is not a first-class product promise, and that silence is informative.
The mitigation pattern from third-party implementation guides is the right starting point, and it is configuration work, not a checkbox. First, keyword-triggered escalation: "burst," "leak," "flooding," "sewage," "no water," "no hot water" — any of these transfers immediately to the on-call tech's cell, no qualifying questions, no booking flow. Second, a dead-man default: if confidence is low or the caller says "human" or "emergency," transfer, don't retry. Third, 50 to 100 scripted test calls before go-live — stressed voices, speakerphone, background noise, accents — logging every misroute. If a vendor will not sit through a demo where you play the panicked 2 a.m. caller, you have learned what you needed to know. We have not run this suite against these tools yet ourselves; when we do, the results land here with dated recordings and a changelog entry.
What after-hours answering actually costs: the per-call math
The pricing models in the table do not compare cleanly on purpose, so here is the arithmetic on a common denominator: an average call of roughly three minutes.
Per-minute (Rosie, Ruby, Retell). Rosie 's $49 plan works out to about $0.20 per included minute, or roughly $0.59 per three-minute call; its $299 plan drops that to about $0.45 per call. Ruby's human-staffed $250 plan is $5.00 per minute — about $15 per call, and vendor documentation confirms billing starts when the receptionist takes the call, hold time included. That is the price of a human voice, and it is roughly 25 times Rosie 's rate.
Per-call (Smith.ai). The AI tiers work out to $1.67–$3.00 per call depending on volume. Smith.ai argues per-call beats per-minute because minute-billed vendors have an incentive to let calls run long. Fair point — and the mirror image is that Smith.ai 's own escalation to live agents is where user reports (G2/Trustpilot, summarized by third-party roundups, individual claims unverified) describe bills inflating through auto-transfers the customer did not ask for. Every billing model has an incentive seam. Find the seam before you sign.
Per-customer (Goodcall). Unlimited minutes, billed on unique monthly callers: 100 for $79, then $0.50 per additional unique customer. For a plumbing shop with repeat maintenance clients, this is genuinely cheap. For a shop whose after-hours line draws one-off price-shoppers, every new caller is a billable unit whether or not they book. Model your actual call log against it before assuming it is the budget option.
The cost floor (Retell). Retell publishes what the underlying technology costs because it sells to the people who build these products: $0.07–$0.31 per minute all-in — voice infrastructure at $0.055/min, transcription, telephony at $0.015/min, and the language model at $0.025–$0.16/min depending on which model runs the conversation (the same per-token economics we track in our LLM API pricing data ). A three-minute call costs a reseller roughly $0.21 to $0.93 raw. A vendor charging $2–$3 per call is marking that up five- to tenfold. The markup is not a scandal — integrations, tuned prompts, and support cost money. But make vendors itemize what the margin buys, and note that "tested emergency routing" is the only item on that list that protects the 2 a.m. call.
"Integrates with ServiceTitan": what that phrase actually means
Every tool on this page will claim it "integrates with your systems." For a plumbing or HVAC shop, that phrase spans a canyon, and the price difference across it is 10x.
On one side: booking into a calendar. Rosie and Goodcall book appointments into calendar slots. That is real and useful — for a shop that dispatches off Google Calendar. It is not a ServiceTitan job. Nobody checked the capacity board, the service zone, the membership status, or which tech is on call. If your dispatcher spends the first hour of every morning re-keying the AI's "appointments" into ServiceTitan as actual jobs, the AI did not remove work. It moved the work to 7 a.m. and added a transcription-error step in the middle.
On the other side: booking a job. Avoca is a ServiceTitan-marketplace-listed native integration that books jobs directly into ST, and it is priced like it: quote-only, with third-party reports of $1,000–$3,000 per month for a mid-market shop and $2,500–$3,500 at the top — unverified against the vendor, who publishes nothing. Avoca 's stated target is shops at $3M+ revenue with five or more CSRs on ServiceTitan or Housecall Pro, and the company raised $125M at a $1B valuation in 2026 to chase exactly that segment. ServiceTitan's answer is in-house: Contact Center Pro's AI voice agents handle overflow and after-hours booking natively, with no published price; one third-party report puts it at $400–$600+ per month plus $1,000–$5,000 onboarding on a 12-month minimum, which we could not verify.
The buying question is therefore not "does it integrate" but "what record exists in my FSM when the call ends, created by whom." Get the vendor to demo an after-hours call that ends as a ServiceTitan job with the correct business unit, job type, and arrival window — or accept, with open eyes, that you are buying a message-taking service with a calendar attached. Both are legitimate purchases at their respective prices. Paying Avoca money for calendar-tier behavior is the mistake.
A note on SERP hygiene, because it shows who you are taking advice from: Numa, which circulates in home-services listicles, sells exclusively to car dealerships per its own homepage. Any roundup recommending it to plumbers copied a list without checking. Verify everything, including us.
Will your 2 a.m. caller even talk to a bot?
The uncomfortable data point: a 2026 OnePoll survey of 6,000 US, UK, and Canadian consumers found 31% would hang up if connected to AI — up from 29% six months earlier — and 85% prefer a real person. Label attached: the survey was commissioned by AnswerConnect, a human answering service with an obvious interest in that result. Discount accordingly. But do not discount to zero; the pattern of consumers shouting "human" and "agent" at phone trees is reported widely enough to be familiar to anyone who has called an airline.
For a plumber the calculus is specific. Your after-hours caller with a flooded basement is the least patient caller in the economy. The realistic comparison, though, is not "bot versus friendly human" — it is "bot versus voicemail," because voicemail is what most small shops actually run at 2 a.m., and against voicemail an AI that answers on the second ring and either books or escalates wins some fraction of callers who would otherwise have dialed the next result. The design rules that follow from the hang-up data: the agent identifies itself immediately, the word "human" always transfers without argument, mid-call SMS ( Rosie and others support this) confirms the booking in writing, and Spanish support matters in most US metros — included with Rosie , worth confirming with every other vendor.
One honesty note: our research pass found no first-person plumber or HVAC practitioner reviews of these specific tools, and we do not fabricate quotes. The user feedback cited on this page comes from review-platform summaries and is labeled where used. When we collect first-party practitioner reports, this section gets them, with links.
Callbacks, texts, and recordings: the legal box around your robot
Three legal surfaces touch an AI phone agent, and the vendor's terms of service will not carry the liability for any of them. Not legal advice — confirm all of this with your own counsel.
Recording and transcription. These tools record and transcribe essentially every call; that is how they book jobs and generate summaries. Roughly a dozen states (the count varies with how Michigan, Oregon and Connecticut are treated) — including California, Florida, Illinois, Pennsylvania, Washington, Maryland, and Massachusetts — require all-party consent to record (per a third-party state-law guide for AI agents). California Penal Code §632 requires all-party consent, and §637.2 attaches $5,000 statutory damages per violation; compliance blogs document an active CIPA class-action wave aimed specifically at AI call analytics. The operational rule: the agent discloses recording at the top of the call, before any data capture, and for cross-state calls you follow the stricter state. Confirm the disclosure line is actually in your vendor's default script — do not assume.
Outbound calls and texts. The FCC's declaratory ruling of February 8, 2024 ( FCC 24-17 ) classifies AI-generated voices as "artificial or prerecorded" under the TCPA, which means outbound AI-voice calls require prior express consent — written consent for marketing — with penalties of $500 to $1,500 per call, uncapped (per third-party legal summaries). The nuance that matters for a plumbing shop: answering inbound calls is not the regulated act. But the moment your "inbound" AI receptionist calls a missed lead back, or fires follow-up texts about a maintenance-plan promotion, you have crossed into TCPA territory. Several of these products offer exactly those features. Ask the vendor which of its features place outbound communications, and get the consent-capture flow in writing before enabling any of them.
AI disclosure. Compliance blogs cite a California requirement that AI voices disclose themselves upfront; we could not verify the cited statute text and flag it as unverified. Disclosure is cheap and defuses both the legal question and the hang-up problem, so the practical answer is: have the agent say it is an automated assistant regardless of what the statute compels.
Where these tools fall short
The pattern-level gaps first. No vendor on this page publishes an independent accuracy benchmark, a misroute rate on emergency calls, or a refund policy for a blown dispatch. Every quantitative claim in this market is vendor-authored or anecdotal. And pricing goes opaque — Avoca , ServiceTitan Contact Center Pro, Smith.ai 's form-gated page — exactly where the checks get biggest; third parties rank on Google by simply publishing the numbers the vendors hide.
Tool by tool, on record:
- Rosie publishes no overage rate, and nothing plumbing-specific: the emergency escalation that makes or breaks the product is configuration you do and test yourself.
- Goodcall names no emergency-dispatch feature on its pricing page, its unique-customer billing punishes one-off-inquiry call patterns, and a Trustpilot reviewer reported difficulty reaching a human — the one failure a home-services shop cannot tolerate.
- Smith.ai carries user reports of unrequested AI-to-human transfers inflating per-call bills (individual claims, unverified; overall Trustpilot ~4.3/5 across 336+ reviews as of January 2026), and its pricing page went behind a contact form — the wrong direction for a market whose core disease is pricing opacity.
- Ruby is the most expensive per minute here by a wide margin, with recurring cost complaints on Capterra, and publishes no overage rate.
- Avoca and ServiceTitan Contact Center Pro publish no pricing at all; every dollar figure attached to them in this article is a third-party report we could not verify. Both are sales-call purchases with onboarding costs and, reportedly, contract minimums.
- Retell is not a product for a plumbing shop; it is the parts bin. Building on it means you own the prompt, the booking flow, the consent script, and every 2 a.m. misroute.
And the one we cannot yet report: how these tools actually perform on stressed, noisy, panicked emergency calls. Nobody measures it publicly, we have not run our suite yet, and until someone does, every "handles emergencies" claim on a vendor page is a hypothesis you are testing with your own customers.
All guides in this topic
- AI Receptionist — the category hub: how AI phone answering works, the full vendor landscape across pricing models, the consent-law checklist, and how we test.
- AI Receptionist for Dental Offices — the HIPAA angle: why an AI phone agent is a business associate, the BAA and subcontractor-chain questions, and PMS integration reality for Dentrix, Eaglesoft, and Open Dental.
- AI Receptionist for Law Firms — legal intake without a privilege leak: ABA Formal Opinion 512, bot-disclosure rules, conflict-check limits, and Smith.ai versus Ruby versus DIY intake bots.
No comments yet