Key takeaways
- Construction firms are slower AI adopters than business overall: 17.1% used AI in a business function in the latest Census survey, against 24.2% of all businesses[1].
- Write the rules before you switch anything on: service area, what counts as an emergency, intake fields, rate wording, escalation path and the no-answer step.
- Some decisions stay with people: refusing someone in danger, quoting repair prices, waiving fees and judging gas or electrical safety.
- Run at least ten scripted test calls, including the awkward ones, and review real recordings every morning for the first month.
What after-hours AI answering should do
The job at night is narrow and important: answer every call, find out what is wrong and where, decide whether a person needs to be woken up, and leave the office a clean record in the morning. A good setup does those four things consistently. It does not need to diagnose a water heater or close a sale.
Use of AI in business is growing but still uneven. In the Census Bureau's Business Trends and Outlook Survey for the two weeks to October 18, 2026, 24.2% of all US businesses said they used AI in a business function, and 27.6% expected to in the next six months[2]. Construction lagged, at 17.1% using AI now[1]. Among small businesses more broadly, 58% said they used generative AI in 2025, according to the U.S. Chamber of Commerce[3]. A plumbing shop that sets up AI answering now is early, and the rules it writes will matter more than which tool it picks.
Source: U.S. Census Bureau, 2026[1]
Show the data
| Item | Value |
|---|---|
| All US businesses | 24.2% |
| Construction businesses | 17.1% |
Decide the scope first
List the calls you get after hours and decide, for each, what the AI handles alone and what it passes on. Start narrow. You can widen the scope once you trust the results.
| Call type | AI handles alone | Passes to a person |
|---|---|---|
| Active leak or burst pipe | Shutoff advice, full intake, read-back | Always, via the on-call path |
| Sewage backup | Stop-water-use advice, intake | When still rising or no usable toilet |
| No hot water, no leak | Intake, morning booking offer | When vulnerable occupants or relief valve discharge |
| Gas smell | Tell caller to leave and call the utility or 911 | Notify on-call manager after the call |
| Routine booking request | Collect details, offer the next open slot | No |
| Billing or complaint | Take a message | Morning office, flagged |
| Sales and spam calls | End politely | No |
A smaller shop can start narrower still. In the example of a one-van plumber whose calls are answered while he is on the job, only the calls he cannot pick up are forwarded, routine jobs go into windows he opens, and urgent calls reach his cell under rules he writes.
Write the rules
Service area
A list of ZIP codes you serve at night, which may be smaller than your daytime area. Calls outside it get a polite refer-out script.
Emergency definitions
Plain-language triggers such as water actively leaking, sewage rising, no water to the home, or no usable toilet with people at home.
Intake fields
Name, address with ZIP code, callback number, problem in the caller's words, what has been shut off, who owns the property and access notes. Every field is read back.
Rate wording
The exact sentence about your after-hours fee and any card authorization, approved by you. The AI reads it as written and does not negotiate.
Escalation path
Who gets the call or text for an urgent job, in what order, and how long to wait before trying the next person.
No-answer step
What the caller is told if nobody on the path responds, including safety advice and a firm morning time.
Where the record goes
A summary, recording and transcript for every call, sent to the inbox or system the morning team actually checks.
Keep the rules in one document that you own, not scattered across a vendor's settings screens. When you change the night service area, a rate or the on-call rotation, update the document first and then the system, and note the date. That record is what lets you work out later why a call was handled a particular way, and it makes switching tools far easier if you ever need to.
Write the greeting
The first ten seconds decide whether a panicked caller stays on the line. Keep the greeting short, name the company, say that the call is recorded and get to the problem. Avoid long menus and avoid pretending to be a named human employee; callers who later realize they spoke to an AI and were not told tend to feel misled.
Greeting to adapt: βThanks for calling [company]. This is our after-hours line, and the call is recorded. I can take down your problem and reach our on-call plumber if itβs urgent. Is water leaking or is anyone in danger right now?β
Ending with the safety question means the most urgent callers are sorted in the first exchange, and everyone else hears straight away that they have reached the right place.
Decisions AI should never make alone
- Turning away someone who describes danger: gas, water near electrics, sewage with a vulnerable person present. These always reach a person or emergency services.
- Quoting the price of a repair. It can read your approved after-hours fee, but repair pricing needs a tech on site.
- Waiving or discounting fees, or promising a free visit.
- Promising an arrival time that the on-call tech has not confirmed.
- Judging whether a gas appliance or electrical panel is safe to use.
- Deciding a complaint or a refund.
Write these into the instructions as hard limits, and include the exact words the AI should use instead, such as βI canβt give a repair price, but the plumber will confirm it with you before starting any work.β
Set up the escalation path
For urgent calls, the AI should reach the on-call tech first, then a backup, then a manager, with a fixed wait at each step. An SI Employee such as Ava follows the path you approve: on-call technician first, then your backup, then a clear no-answer step. Whatever tool you use, the caller should be told what is happening, such as βIβm reaching our on-call plumber now and will stay with you,β rather than left on hold. The detailed rules are in escalation rules for AI call handling.
- Caller
- SI Employee
- Your team
- System
01Caller
Caller rings at 1:10 a.m.
Water is spraying from a pipe in the basement ceiling.
02SI Employee
Greets and asks about danger
Names the company, says the call is recorded, asks if water is near electrics.
03SI Employee
Guides caller to shutoff
Walks them to the main valve, then takes address, ZIP code and callback number.
04SI Employee
Reads back, marks urgent
Confirms every field and reads the approved after-hours fee exactly as written.
05Your team
On-call tech takes it
Takes the transferred call and gives the caller an arrival time.
06System
Record ready for 8 a.m.
Summary, recording and transcript land in the inbox the office checks.
Example: a fictional shop. The order of people and the wording come from the rules you write, not from the tool.
Give the on-call tech a short, structured alert: urgency, address, problem, what has been shut off and the callback number. A tech woken at 2 a.m. should be able to decide in thirty seconds whether to roll.
Recording, consent and outbound calls
Recordings are what make AI answering reviewable. Federal law allows recording when one party consents, but some states require all parties to agree[4], so the greeting should say the call is recorded. If the system also makes outbound calls, such as callbacks to missed callers, note that the FCC ruled in February 2024 that AI-generated voices count as artificial under the TCPA[5]. Returning a call the customer just made is different from marketing, but check the consent rules with your adviser before switching on any automated outbound calling.
Run the test calls
Before going live, have two or three people call the line with scripted scenarios at night, from their own phones. Score each call on whether the intake was complete, the read-back was correct, the escalation fired when it should and the summary arrived.
- Burst pipe with the caller unsure where the main shutoff is.
- Sewage backup in a rental, landlord unreachable.
- No hot water, no leak, caller with a newborn.
- Gas smell reported almost in passing.
- Address outside the night service area.
- Caller who refuses to give a callback number.
- Caller speaking Spanish, if you serve Spanish-speaking customers.
- Routine request for a faucet repair next week.
- Angry existing customer about an invoice.
- Caller who asks for the repair price and pushes for a number.
Then test the failure modes: the on-call tech's phone off, the backup not answering, and two urgent calls arriving within five minutes.
Example: a first week
Example: a fictional four-tech shop switches on AI answering for nights only. In the first week it takes nineteen calls. Six are urgent and all reach the on-call tech within a few minutes; five are booked into morning slots; the rest are routine requests, messages and spam. The morning review finds two problems: the AI was reading the after-hours fee before asking about safety, and one caller's apartment number was missed. The owner moves the fee sentence later in the flow and adds unit number as a required field.
| Input | Example |
|---|---|
| Night calls a week | 19 |
| Calls from new customers (%) | 40% |
| New callers who book (%) | 50% |
| Average job value | $500 |
| After-hours calls a month | 82 |
| From new customers | 33 |
| Bookings if answered | 16 |
| Monthly after-hours value | $8,227 |
How it is worked out: After-hours calls a week Γ 4.33 weeks Γ new-customer share Γ booking rate Γ average value.
Assumptions
- Nineteen calls a week, matching the first week above. Spam and wrong numbers are part of that count, which is why the new-customer share is set low at 40%.
- Half of new callers booking is set for this example; use the booking rate from your own call log.
- Average job value of $500, from the median homeowner spend on plumbing fixture projects in 2021 to 2023.[6]
Example only. This is value already arriving at night, not new revenue; the setup decides how much of it your shop keeps.
Review every morning for a month
- Listen to every urgent call and two or three routine ones.
- Check each summary against the recording for address, problem and promise.
- Track how many urgent calls reached a person and how long it took.
- Note any question the AI could not answer and decide whether to add an answer or route it to a person.
- Ask the on-call techs whether the alerts gave them what they needed.
Callers rarely pick up unknown numbers, so a live answer at night is worth more than a callback[7]. After a month, the review can drop to urgent calls and a weekly sample. Before you buy, read what to test in an AI receptionist for plumbing, and if you want to hear a working setup first, have the demo SI Employee call your phone.
Frequently asked questions
Can AI answer plumbing emergency calls at night?
Yes, if it is set up with clear emergency definitions, a full intake with read-back, and an escalation path that reaches your on-call tech for urgent calls. It should not make safety or pricing decisions alone.
What should an AI never decide on a plumbing call?
Turning away someone in danger, quoting repair prices, waiving fees, promising unconfirmed arrival times, judging gas or electrical safety, and settling complaints.
Do I need to tell callers the call is recorded?
It is the safest approach. Federal law allows one-party consent, but some states require every party to agree, so announce recording in the greeting.
How do I test an after-hours AI answering setup?
Run scripted test calls covering emergencies, edge cases and failure modes, such as the on-call tech not answering, and check the intake, escalation and summary for each one.
How long does after-hours AI answering take to set up?
Switching it on is quick; the work is in the rules. Plan an evening to write the scope, emergency triggers, fee wording and on-call path, a night of scripted test calls, and a month of short morning reviews.
References
U.S. Census Bureau (2026). Business Trends and Outlook Survey (BTOS), response estimates by sector (opens in a new tab) β
U.S. Census Bureau (2026). Business Trends and Outlook Survey (BTOS), national response estimates (opens in a new tab) β
U.S. Chamber of Commerce Technology Engagement Center (2025). Empowering Small Business: The Impact of Technology on U.S. Small Business (2025) (opens in a new tab) β
Legal Information Institute, Cornell Law School (U.S. Code) (2026). 18 U.S. Code Β§ 2511 - Interception and disclosure of wire, oral, or electronic communications prohibited (opens in a new tab) β
Federal Communications Commission (2024). FCC Makes AI-Generated Voices in Robocalls Illegal (news release) (opens in a new tab) β
U.S. Census Bureau and U.S. Department of Housing and Urban Development (2021-2023). Home Improvements (2023 American Housing Survey infographic) (opens in a new tab) β
Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) β