Key takeaways
- AI phone answering is good at the repetitive part of calls: greeting, taking details, answering standard questions, booking routine visits and passing urgent calls on.
- It should not diagnose faults, price unseen repairs or handle a gas smell beyond reading an approved safety message and escalating.
- Construction is a late adopter: 81% of UK construction businesses used no AI at all in autumn 2026[1], down from 91.9% in 2023[2].
- Check its work. 84% of UK businesses using AI reported at least some human checking of outputs[3]; listen to recordings weekly.
- Pilot it on overflow and evening calls first, with mystery calls, before it answers your main line.
What AI phone answering means now
A few years ago, 'automated call answering' meant a menu: press 1, press 2, leave a message. AI phone answering is different. The caller speaks normally, the system understands the request, asks follow-up questions, and either completes a task such as booking a service or hands the call to a person. For a plumbing and heating firm the attraction is plain: calls arrive while everyone is on a job, and an unanswered call to a plumber is often a lost one.
The technology is useful, but it is not magic, and the honest answer to 'should I use it?' depends on which calls you want it to take. This guide separates what it handles well from what still needs a person.
Where the trade stands
Construction businesses, which include plumbing and heating firms in the official statistics, have been slow to adopt AI. In autumn 2026, 81% of UK construction businesses said they used no AI technology at all[1], down from 91.9% in September 2023[2]. Only 6.7% used large language model text generation, against 19% of all businesses[4]. Among construction firms that had not used AI, the most-cited barrier was lack of expertise, ahead of cost[5].
Source: Office for National Statistics, 2023[2]
Show the data
| Item | Value |
|---|---|
| September 2023 | 91.9% |
| Autumn 2026 | 81% |
Customers are further ahead. Six in ten UK adults reported using AI chatbots in the previous three months in 2024[6]. Attitudes are mixed, though: around four in ten expected AI to have a positive impact on them personally, while three in ten expected a negative one[7]. That split matters on the phone. Some callers will be perfectly happy speaking to an AI that books them in; others will want a person, and the system must make that easy.
What it does well
- Answering every call promptly, including when the line is busy, at lunchtime and in the evening.
- Greeting callers in your firm's name and taking structured details: name, number, postcode, fault, boiler make and any fault code.
- Reading details back to confirm them, which reduces wrong numbers and misheard postcodes.
- Answering standard questions from your own information: areas covered, opening hours, whether you do landlord certificates, your published call-out charge if you choose to share it.
- Booking routine work such as annual services and quote visits into slots you have made available.
- Recognising urgent calls from your rules and passing them to the engineer on call.
- Sending the office a written summary, with the recording, so nothing relies on memory.
This is the part of the job that is repetitive for people and easy to get wrong when an engineer answers with one hand. Done well, it means the office starts the day with a list of complete, booked or triaged jobs rather than a voicemail backlog.
What it cannot or should not do
- Diagnose faults. Talking a caller through boiler resets or valve adjustments is a job for a qualified engineer, if anyone.
- Price unseen repairs. A confident AI guess becomes a complaint when the real price arrives.
- Handle disputes, complaints or upset customers beyond taking details and arranging a call from a named person.
- Make judgement calls your rules do not cover, such as squeezing a job in for a long-standing customer.
- Take card numbers by voice, unless the system is specifically built and assessed for payments. Payment links are safer.
- Pretend to be a person. If a caller asks, it should say it is an automated assistant answering for your firm.
When a person must take over
| Situation | Why a person is needed | What the AI should do |
|---|---|---|
| Caller asks for a person | Their choice, and refusing loses trust | Transfer or promise a call-back with a time |
| Suspected gas leak or carbon monoxide | Life safety | Read your approved safety message and the emergency number, flag urgent, alert on-call |
| Vulnerable caller with no heat or water | Priority and care | Mark urgent and route to the on-call engineer |
| Complaint about previous work | Needs empathy and authority | Take details and book a call from the owner or manager |
| Commercial or agent contract terms | Needs the account owner | Take details and pass to the right person |
| Anything the AI is unsure about | Better a call-back than a wrong answer | Say it does not know and arrange a call-back |
Safety-critical calls
Gas smells, suspected carbon monoxide and serious flooding are where any call handling, human or AI, must follow a fixed script. For gas, the script is short: leave the property, ventilate if safe, do not use switches or flames, and call the gas emergency number. The AI should give no technical advice beyond that, mark the call urgent and alert your on-call engineer. Ava, for example, reads the firm's approved safety message and the gas emergency number, gives no technical advice and flags the call as urgent. Whatever system you use, write the safety script yourself, test it with mystery calls and listen to every real safety call afterwards.
Check its work every week
AI makes mistakes: a misheard postcode, a booking in the wrong slot, a call it treats as routine that a person would have spotted as urgent. Businesses that use AI generally keep people in the loop; 84% of UK businesses using AI in 2025 reported at least some human checking of AI outputs, and 67% reported significant checking[3]. For phone answering, that means:
- Listening to a sample of recordings each week, including every call marked urgent.
- Comparing bookings against the diary rules you set.
- Reviewing any call where the caller asked for a person or sounded unhappy.
- Updating the AI's information when your hours, area or services change.
Questions to ask any provider
- Which tasks are live today, and which are on the roadmap? Ask for a demonstration of each live task with your own scenarios.
- How does it decide a call is urgent, and can we write those rules?
- How does a caller reach a person, and how fast?
- Where are recordings and transcripts stored, for how long, and will you sign a UK GDPR processor contract?
- What happens when it does not know the answer?
- Can we see every call, with outcome, in one place?
Be wary of any provider that claims everything is already possible. With Ava, for instance, answering calls, taking jobs, following the on-call rota and booking routine visits are live, while text confirmations and connections to job-management software are still listed as planned. A clear line between live and planned is a good sign.
Prepare the information it will use
An AI answering service is only as good as what you tell it. Most disappointing pilots come from thin set-up information, not from the technology. Before switching it on, write down, in plain language:
- The postcodes or towns you cover, and any you deliberately avoid.
- The jobs you do and do not take, in the words customers use, such as 'boiler not firing' or 'blocked toilet'.
- Your opening hours, bank holiday cover and what happens out of hours.
- Your urgency rules, including vulnerable occupants, no heating in cold weather and active leaks.
- Who is on call, in what order, and what to tell the caller if nobody answers.
- What it may say about charges, and the exact words, or that it should say a person will confirm.
- Answers to the ten questions customers ask most, taken from your own calls.
Review this information every season. Winter rules, new service areas and staff changes all need updating, and a system working from last year's rota is worse than none.
A four-week pilot
Week 1: set up and rehearse
Load your areas, services, hours and urgency rules. Make at least ten mystery calls covering routine, urgent, landlord and safety scenarios.
Week 2: overflow only
Send only busy and unanswered daytime calls to the AI. Listen to every recording.
Week 3: add evenings
Switch evening calls to the AI with your on-call rota connected. Check every urgent alert reached the right phone.
Week 4: review
Count calls answered that would have been missed, bookings made, hand-overs and errors. Decide whether to widen, adjust or stop.
Example: a pilot on evening calls
Example: a fictional four-engineer heating firm pilots AI answering on evening and overflow calls only. In the first week, mystery calls reveal that the system treats 'no hot water' as routine even when the caller mentions a newborn baby, so the owner adds a vulnerability rule. By week four the office starts each morning with a list of complete job details instead of half-heard voicemails, and the on-call engineer receives fewer but better-described urgent calls. The firm keeps a person on the main line during the day and reviews recordings every Friday. For how this fits with ring groups and overflow, see routing calls across several engineers.
Frequently asked questions
Can AI answer the phone for a plumbing business?
Yes. It can greet callers in your firm's name, take job details, answer standard questions, book routine visits and pass urgent calls to your on-call engineer. It should not diagnose faults or price unseen repairs.
Is AI phone answering safe for gas leak calls?
Only with a fixed, approved safety message: leave the property, ventilate if safe and call the gas emergency number, with no technical advice and an urgent alert to you.
Will customers mind talking to an AI?
Some will, some will not. Make it easy to reach a person, have the AI say it is automated if asked, and review calls where callers asked for a human.
How should I trial AI phone answering?
Start with overflow and evening calls, run mystery calls covering routine, urgent and safety scenarios, listen to recordings weekly and widen its use only when the results are good.
References
Office for National Statistics (2026). Business insights and impact on the UK economy: wave 165 dataset (AI tables) (opens in a new tab) ↑
Office for National Statistics (2023). Business insights and impact on the UK economy: wave 165 dataset (AI tables) (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2025). AI adoption survey report 2025 (opens in a new tab) ↑
Office for National Statistics (2026). Business insights and impact on the UK economy: wave 165 dataset (AI tables) (opens in a new tab) ↑
Office for National Statistics (2026). Business insights and impact on the UK economy: wave 165 dataset (AI Prevent or Delayed) (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑