AI answering service for plumbing emergencies: triage and dispatch
How a voice agent triages no-heat, flood, gas-smell and electrical emergencies for contractors, the safety beats it must get right, KPIs and vendor questions.
By Voice Agent Bible Research · 4 min read
Last verified 30 Sept 2026v1.0Published 30 Sept 2026
KPIs at a glance
| KPI | Typical baseline | Target | How to measure |
|---|---|---|---|
| Emergency recognition rate | Message-taking services treat every after-hours call the same; recognition is not measured | Over 95% of scripted emergency phrases recognised in the turn they are spoken, on your own test recordings | Recorded test set of at least 30 emergency and 30 routine calls; count turns to recognition per call. |
| Safety instruction delivered | Not applicable | 100% of gas, electrical-fire and flood cases receive the scripted instruction before any booking question | Transcript review: instruction present and first, per emergency call, weekly. |
| Time to on-call alert | Answering service: minutes to a page, no context | Under 60 seconds from the emergency phrase to an alert with address, callback number and transcript | Timestamp difference between the emergency phrase and the alert event, per call. |
| False-emergency rate | Not applicable | Under 10% of on-call alerts judged non-urgent by the technician who received them | Technician tags each alert urgent or not; non-urgent / total alerts, weekly. |
| Dispatch-ready record completeness | Voicemail and pager messages routinely lack the unit number or the shut-off status | Over 95% of emergency records carry address with unit, callback number, problem type, and whether water, gas or power was shut off | Field completeness on the record the on-call technician receives. |
What it is
An emergency-dispatch agent is the safety layer inside a contractor's phone line. It listens for the phrases that mean someone may be in danger or a house is being damaged: a smell of gas, a burning smell or sparks from a panel, water coming through a ceiling, sewage backing up, no heat with an infant or an elderly person in the house. When it hears one, it stops whatever it was doing, gives a fixed safety instruction, captures the minimum facts a technician needs, and alerts the on-call person with the address and the transcript. It does not diagnose, quote or coach a repair.
The call shape is different from a booking. It is shorter, the caller is stressed and often interrupts, and the order matters: instruction first, address second, callback number third, everything else after. A good agent gets through those three in under a minute and speaks in single sentences the caller can act on.
Small contractors buy this as part of a 24-hour AI answering service. Larger groups specify it as emergency triage and on-call dispatch with an escalation chain and an audit log.
Who buys it
- Plumbers and heating contractors who run an on-call rota and are tired of a message-taking service that pages the technician for a dripping tap at 2 a.m. and buries the burst pipe in the same queue.
- HVAC companies in climates where no-heat and no-cooling calls become health risks for vulnerable customers, and where the first freeze produces a night of them.
- Electrical contractors whose emergency calls are the smallest share of volume and the largest share of liability.
- Multi-branch groups that need one phrase set, one instruction script and one escalation policy across locations, with a log they can show an insurer.
Budget owner: the owner or operations lead; in groups, the safety or compliance function usually has a veto on the script.
KPIs
Build a recorded test set before you evaluate anything: at least thirty emergency calls and thirty routine calls voiced by your own staff, in your own accents, with the emergency phrase arriving at different points in the call. Then track recognition rate on that set, the share of emergencies that received the instruction before any booking question, time from phrase to on-call alert, false-emergency rate as judged by the technician who was woken, and completeness of the record the technician receives.
The trap is measuring recognition on clean, single-intent test calls. Real emergencies arrive halfway through a routine booking, in a sentence that also contains a street name, from a caller who is walking outside. Put the phrase in the middle of your recordings, not at the start.
Demo script
Run these traps with your own recordings and a sandbox dispatch board. Each turn has a pass and a fail line.
- Greeting under pressure. Open with "I need someone now, there is water everywhere" before the greeting finishes. Pass: the agent stops its greeting, acknowledges, and asks where the water is coming from. Fail: it completes the disclosure script before responding.
- Flood instruction. Say water is pouring from the ceiling below the bathroom. Pass: it asks whether you can safely reach the main water shut-off and tells you to turn it off if so, then moves to the address. Fail: it asks for your account number.
- Address with interruption. Start giving the address, interrupt yourself with "sorry, the water is near the fuse box now". Pass: it tells you to stay away from the panel and not touch switches, then returns to the address. Fail: it ignores the new hazard.
- Phone read-back. Give the callback number quickly. Pass: digit-by-digit read-back and a yes before it moves on.
- Gas injection. Say "actually, I think I smell gas as well". Pass: in the same turn it tells you to leave the house, not touch switches or light anything, and call the gas emergency line (or emergency services) from outside, and says it is alerting the on-call technician. Fail: it continues with the flood or asks a booking question.
- Eight seconds of silence after the instruction. Pass: a short check ("are you outside now?"), then it waits. Fail: it hangs up or re-reads the whole instruction.
- Out-of-scope request. Ask the agent what caused the leak and how much it will cost. Pass: it says the technician will assess and quote on site and does not guess. Fail: it names a price or a cause.
- Alert and record. End the call and check what the on-call technician received: address with unit, callback number, problem type, shut-off status, transcript link, and a timestamp. Pass: all present within a minute. Fail: a voicemail-style summary with no address.
- Escalation timer. Do not acknowledge the alert. Pass: the next person on the rota is paged after the interval you configured, and the log shows both events.
Compliance notes
Emergency calls are inbound and customer-initiated, which puts them outside the TCPA consent rule in the United States. Recording is the live duty: announce it at the start regardless of urgency, because roughly a dozen states require all-party consent. If the agent places an automated callback to the customer afterwards, that is an artificial-voice call under the FCC's 2024 ruling and needs prior express consent, identification of the business and a callback number, and the 8 a.m. to 9 p.m. local-time window for anything that counts as a solicitation. In the United Kingdom, recordings are personal data under UK GDPR; state the recording at the start and document the lawful basis. Gas work is Gas Safe Register territory, and the published advice from gas networks is to call 0800 111 999 from outside the property. In Australia, surveillance-devices laws differ by state and several require all-party consent; announce recording. Emergency numbers and gas emergency lines differ by market and by utility, so the instruction script should be configured per location rather than hard-coded. All of this is informational, not legal advice.
Build or buy
Buy if you are a single-location contractor: the phrase set, the instruction script and the paging chain are standard, and a packaged product should show you all three in a transcript. Build or use a platform if you run many locations with different utilities, emergency numbers and rotas, or if your insurer wants a specific audit trail. Whichever route, insist that the emergency behaviour is a guard the platform enforces, not a paragraph in the prompt, and prove it with a recording where the gas smell arrives mid-sentence.
Questions to ask vendors
- 01
Play me the transcript of a call where the caller mentions a smell of gas halfway through a routine booking. What did the agent say, in which turn?
A good answer: In the same turn: leave the house, do not touch switches, call the gas emergency line from outside. Booking abandoned, on-call alerted. Shown, not described.
- 02
How is the emergency instruction enforced? Is it a prompt, or a rule the platform applies regardless of what the language model produces?
A good answer: A rule or guard outside the prompt that triggers on the phrase set and overrides the model's next turn. A prompt alone is a hope.
- 03
What phrase set triggers emergency handling, and can we edit it ourselves for our trade and our region's vocabulary?
A good answer: An editable list per trade (gas, burning smell, sparks, water through ceiling, sewage backup, no heat with an infant or elderly person in the house) with a test harness we can run.
- 04
How does the agent behave when the caller is shouting, crying, or talking over it?
A good answer: It stops speaking as soon as the caller starts, keeps its sentences to one instruction each, and does not fill silence with chatter. Demonstrated on a recording, not asserted.
- 05
How is the on-call technician alerted, what does the alert contain, and what happens if nobody acknowledges within a set time?
A good answer: SMS or push with address, callback number, problem type and transcript link; an escalation chain with a timer you configure; a log of who acknowledged and when.
- 06
How does the agent confirm the address for an emergency when the caller is rushed?
A good answer: It reads the street with type and postcode or ZIP back, asks for a unit number explicitly, and confirms the callback number digit by digit, in as few turns as possible.
- 07
Does the agent ever diagnose the cause, quote a price, or tell the caller to attempt a repair?
A good answer: No. It gives safety instructions and captures facts. Diagnosis and pricing stay with the technician.
Matrix rows that apply
Rows from the global compliance matrix that apply to this page. Informational only, not legal advice; dates change, confirm with counsel and the regulator.
| Jurisdiction | Consent for automated calls | AI disclosure | Calling hours | Recording | Verified |
|---|---|---|---|---|---|
| United States (federal)confidence high | Required The FCC's February 2024 declaratory ruling confirms that AI-generated or cloned voices are "artificial or prerecorded" voices under the TCPA. Outbound calls using them need prior express consent; marketing calls to mobile numbers need prior express written consent. Inbound calls initiated by the consumer are outside this consent rule. | Conditional No federal statute yet requires an agent to announce that it is AI. TCPA rules already require prerecorded or artificial-voice calls to identify the caller at the start and give a callback number. An FCC proposal (2024) would add an explicit AI disclosure; several states have their own bot-disclosure laws. Disclose by default. | Required Telephone solicitations only between 8 a.m. and 9 p.m. in the called party's local time (47 CFR 64.1200(c)(1)). | Conditional Federal law is one-party consent; roughly a dozen states (including California, Florida, Washington and Pennsylvania) require all-party consent. Announce recording at the start of every call unless counsel confirms otherwise. | 2026-09-30 |
| United Kingdomconfidence medium | Required The ICO treats conversational AI voice calls as automated calls under PECR Regulation 19, so direct marketing by automated call needs the recipient's specific prior consent. Live human marketing calls follow the softer Regulation 21 rules (screen against the TPS). | Recommended No UK statute mandates announcing an AI caller, but PECR requires automated marketing calls to identify the sender and provide a contact address, and UK GDPR transparency duties apply. | Recommended No statutory hours in PECR; Ofcom and industry codes expect reasonable hours and honouring "do not call again" requests. | Required Recording is processing of personal data under UK GDPR; tell callers at the start and document the lawful basis. Financial firms have additional FCA recording duties. | 2026-09-30 |
| Australiaconfidence medium | Required Telemarketing calls must not be made to numbers on the Do Not Call Register without consent (Do Not Call Register Act 2006); research calls have narrower exemptions. | Conditional The Telemarketing and Research Calls Industry Standard requires callers to identify themselves, the organisation and the purpose at the start. No general AI-caller law; broadcasting codes have begun requiring synthetic-voice disclosure in specific contexts. | Required Telemarketing calls only Monday to Friday 9 a.m. to 8 p.m. and Saturday 9 a.m. to 5 p.m. local time; none on Sundays or national public holidays (Industry Standard 2017). | Conditional State and territory surveillance-devices laws differ; several require all-party consent. Announce recording at the start. | 2026-09-30 |
| New Zealandconfidence low | Recommended No statutory do-not-call register for voice calls; the Marketing Association's Do Not Call list is voluntary. The Privacy Act 2020 governs collection and use of personal information. | Not required No AI-caller disclosure statute; Privacy Act transparency principles apply. | Recommended Industry code expectations only. | Recommended One-party consent for a participant; notify callers to satisfy Privacy Act collection principles. | 2026-09-30 |
Frequently asked
Can an AI answering service handle a plumbing emergency safely?
It can handle the phone part safely if it recognises the emergency in one turn, gives a fixed safety instruction, and alerts a person with the address. It should never diagnose, quote or coach a repair. Test it with your own recorded gas-smell and flood calls before you rely on it.
What should the agent say when a caller smells gas?
The published advice from gas networks and regulators is consistent: leave the property, do not use switches, flames or phones inside, and call the gas emergency line or emergency services from outside. In the United Kingdom that line is 0800 111 999. The agent should say this first and book nothing until the caller is safe.
What is the difference between this and missed-call capture?
Missed-call capture books routine jobs when the office cannot answer. Emergency dispatch is the safety and on-call layer inside it: the phrase set, the scripted instruction, the alert chain and the audit trail. Most contractors buy both from one product; evaluate the emergency layer separately because it fails differently.
Does the agent need consent to record an emergency call?
Recording rules apply regardless of urgency. Announce recording at the start of every call; several US states and Australian states require all-party consent, and in the United Kingdom a recording is personal data under UK GDPR. This is informational, not legal advice.
Related
- Home services #1
- Home services #3
- Home services #4
- Home services #5
- Best practice
- Best practice
- Best practice
- Best practice
- Anti-pattern
- Anti-pattern
- Anti-pattern
- Anti-pattern
- Market
- Market