Technology AI

Choosing an AI phone tool: a checklist for trades

A checklist for heating firms choosing an AI phone tool: booking, urgent call triage, hand-off to the team, data protection, trials and contracts.

Ava research teamPublished 6 min read

On this page
  1. Start with the outcomes
  2. What makes firms hesitate
  3. 1. Booking
  4. 2. Triage and urgent calls
  5. 3. Hand-off to your team
  6. 4. Data, privacy and security
  7. 5. Trial, contract and exit
  8. Score every option the same way
  9. Running the evaluation
  10. Red flags
  11. Worked example: scoring two tools
  12. Frequently asked questions
  13. References

Key takeaways

  • Write down the three outcomes you want before looking at any product.
  • Test booking, urgent calls and the hand-off with your own scripted calls, not the supplier's demo.
  • Check the data protection basics: processor contract, where data goes, retention and an impact assessment where risk is high.
  • Insist on a trial on live calls that includes a busy period, and an easy exit.
  • Score every option on the same sheet so the decision is not made on the nicest voice.

Start with the outcomes

AI phone tools are sold on how human they sound. That matters less than three practical outcomes, and you should write yours down before any demo. For example: no urgent breakdown call waits more than a few minutes to reach the on-call engineer; routine services and safety checks are booked without a call back; and the office receives a summary it can act on without ringing the customer.

With those written down, every question below has a purpose, and every product can be judged against the same yardstick.

Add the limits as well: what the tool must never do. A typical list for a heating firm is no technical or safety advice beyond the approved script, no prices beyond those you have approved, and no promises of arrival times your rota cannot keep.

What makes firms hesitate

Official figures suggest the hesitation is about fit and know-how more than money. Among construction businesses in autumn 2026, the most-cited barrier to AI was lack of expertise at 8.2%, ahead of cost at 5.9%[1]. Asked what would help, 19% named ease of integration and use, against 10.2% for affordability[1]. A checklist addresses both: it gives you the questions an expert would ask, and it puts integration at the centre.

Factors and barriers in AI use, UK construction businesses, autumn 2026
Factors and barriers in AI use, UK construction businesses, autumn 2026Bar chart. Ease of integration and use would help: 19%; Affordability would help: 10.2%; Barrier: lack of AI expertise: 8.2%; Barrier: cost: 5.9%.Ease of integration and use would helpEase of integration and use would help: 19%19%Affordability would helpAffordability would help: 10.2%10.2%Barrier: lack of AI expertiseBarrier: lack of AI expertise: 8.2%8.2%Barrier: costBarrier: cost: 5.9%5.9%Factors and barriers in AI use, UK construction businesses, autumn 2026Bar chart. Ease of integration and use would help: 19%; Affordability would help: 10.2%; Barrier: lack of AI expertise: 8.2%; Barrier: cost: 5.9%.Ease of integration and use would helpEase of integration and use would help: 19%19%Affordability would helpAffordability would help: 10.2%10.2%Barrier: lack of AI expertiseBarrier: lack of AI expertise: 8.2%8.2%Barrier: costBarrier: cost: 5.9%5.9%

Source: Office for National Statistics, 2026[1]

Show the data
ItemValue
Ease of integration and use would help19%
Affordability would help10.2%
Barrier: lack of AI expertise8.2%
Barrier: cost5.9%

1. Booking

  • Which job types can it book, and which does it leave as requests for the office?
  • Does it book into your real diary, or a separate one you have to copy across?
  • Can you limit slots by job type, engineer, area, day and season?
  • Does it take the details the engineer needs: appliance, location in the property, access, parking, tenant or agent contact?
  • How does it handle a caller who wants a time you do not offer?
  • Can it move or cancel an existing booking found by phone number?

2. Triage and urgent calls

  • Does it apply your definition of urgent, written by you, rather than its own?
  • Can it follow an on-call rota, with a backup and a defined step if nobody answers?
  • What exactly does it say to a caller who smells gas? It should read your approved safety message and the gas emergency number, give no technical advice and flag the call urgent.
  • Does it recognise vulnerable callers, such as an elderly person with no heating, and treat them by your rules?
  • Does it ever promise an arrival time? It should only give times your rules allow.
  • What does it tell a caller when it does not know the answer?

Run the gas smell test call yourself, more than once, with different wording: “I can smell gas”, “there's a funny smell near the boiler”, “my carbon monoxide alarm is going off”. Every version must trigger the safety script.

3. Hand-off to your team

  • What does the summary contain, and could an engineer go to the job from it?
  • Is the recording or transcript attached?
  • Does it reach the right person by the right channel: on-call engineer, office inbox, job system?
  • Are urgent jobs clearly marked?
  • How do you correct a wrong summary and change the rule that caused it?
  • Does it link to your job management software today, or is that only planned? Ask for a live demonstration, not a logo.

Most businesses that use AI keep checking its work: 84% reported at least some human input or checking of AI outputs in 2025[2]. Plan who reads the summaries, and how often, before you start.

4. Data, privacy and security

  • A supplier that handles call data for you is a processor and must work under a contract that includes the terms data protection law requires, such as acting only on your documented instructions and deleting or returning data at the end[3]. Ask for it.
  • Where are recordings and transcripts stored, and who at the supplier can access them?
  • How long are they kept, and can you set that period?
  • Is caller data used to train the supplier's systems, and can you opt out?
  • Does the tool tell callers they are speaking to an automated system and that calls are recorded?
  • Have you assessed the risk? Where processing using new technologies is likely to result in a high risk to people's rights, an impact assessment is required before you start[4]; even where it is not required, writing one is good practice.
  • Are accounts protected with two-step sign-in? Cyber attacks are common, with 43% of UK businesses reporting a breach or attack in the year before the 2025 government survey[5].

5. Trial, contract and exit

  • Can you trial it on live calls for long enough to include a cold snap or a busy Monday?
  • Who sets it up, how long does it take, and how quickly can you change scripts and rules yourself?
  • Is the notice period short, and are there set-up or exit charges?
  • Can you export recordings, transcripts and booking data when you leave?
  • What support do you get when something goes wrong at the weekend?

Agree before the trial what success looks like, using your written outcomes, so the decision at the end is about evidence rather than impressions.

Score every option the same way

Example scoring sheet for comparing AI phone tools
AreaMust-haveWeight
Gas smell and urgent callsCorrect safety script every time; rota followedHigh
BookingBooks into your real diary by your rulesHigh
Hand-offSummary an engineer can act on, with recordingHigh
Data protectionProcessor contract, retention control, clear disclosure to callersHigh
Ease of changeYou can edit rules and scripts quicklyMedium
IntegrationWorks with your diary and inbox todayMedium
Trial and exitLive trial, short notice, data exportMedium
Caller experienceClear, short, offers a person when neededMedium

Any option that fails a high-weight must-have is out, however well it scores elsewhere. That rule stops a polished demo outweighing a wrong answer on a gas call.

Score each area from one to five after the test calls and again after the live trial. Ask a second person, ideally whoever runs the diary, to score independently; where your scores differ, listen to the recordings together before deciding.

Running the evaluation

  1. Write your outcomes and rules

    Urgency definitions, on-call rota, areas covered, job types to book, approved answers.

  2. Shortlist two or three tools

    Only those that claim to do what your outcomes need.

  3. Run scripted test calls

    The same calls on each: breakdown, gas smell, landlord booking, out-of-area, complaint, awkward question.

  4. Check the paperwork

    Processor contract, privacy information, retention settings and exit terms.

  5. Trial the best on live calls

    Read every summary in the first week and a sample after that.

  6. Score and decide

    Use the sheet, note the reasons, and set a review date.

Red flags

  • The supplier will not let you test with your own scripts before signing.
  • Features shown in the demo turn out to be on a roadmap rather than live.
  • Nobody can show you the processor contract or explain data retention.
  • The tool gives technical or safety advice beyond your approved script.
  • Pricing for your busiest month is unclear.

If you are considering Ava, ask it exactly the same questions and run the same test calls; a fair comparison is the only useful one. See also AI employee vs a basic AI receptionist and AI call handling and UK GDPR.

Worked example: scoring two tools

Example: a fictional heating firm scores two AI phone tools. Tool A sounds very natural and books well, but on the third gas test call, worded as a funny smell near the boiler, it offers to book a service. That fails a must-have, so it is out. Tool B is less polished but triggers the safety script on every variant, follows the rota and sends a usable summary with the recording. Its processor contract is clear and data retention can be set. The firm trials Tool B for a month, reads every summary in week one, and keeps it with a review date in three months.

Frequently asked questions

What should I ask an AI phone answering provider?

Ask how it books, how it handles urgent and gas smell calls, what your team receives after each call, how data is protected and retained, and how you can trial it and leave.

How do I test an AI receptionist?

Write scripted calls, including a breakdown, a gas smell in different wordings, a landlord booking and an awkward question, and run the same calls on every option.

Do I need a contract for an AI phone tool?

Yes. A supplier handling call data for you acts as a processor and must work under a written contract with the terms UK GDPR requires.

How long should a trial be?

Long enough to include busy periods, ideally a few weeks that include a cold spell, with every summary read in the first week.

References

  1. Office for National Statistics (2026). Business insights and impact on the UK economy: wave 165 dataset (AI Use Factors) (opens in a new tab) ↑

  2. Department for Science, Innovation and Technology (GOV.UK) (2025). AI adoption survey report 2025 (opens in a new tab) ↑

  3. legislation.gov.uk (The National Archives) (2016). UK GDPR Article 28: Processor (opens in a new tab) ↑

  4. legislation.gov.uk (The National Archives) (2016). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑

  5. Department for Science, Innovation and Technology (GOV.UK) (2025). Cyber security breaches survey 2025 (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo