Key takeaways
- Most construction firms are not using AI yet: 81% used none in autumn 2026, against 64.1% of all businesses[1][2].
- Administration is one of the two most common areas for business AI use, cited by 72% of users and planners[3].
- Good first jobs: drafting letters and reports, summarising notes, tidying data and answering routine calls, always with a person checking.
- 84% of businesses using AI check its outputs at least partly[4]; in a gas firm, anything touching safety or price must be checked.
- Treat customer data carefully: use business accounts, know where data goes and include AI tools in your privacy notice.
Where small trade firms are with AI
AI is being talked about everywhere, but the trade is mostly not using it yet. In the ONS business survey for late September to early October 2026, 81% of UK construction businesses said they did not use any artificial intelligence technology, compared with 64.1% of all businesses[1][2]. Among the smallest businesses, with up to nine employees, 65.7% used none and 18.8% used text generation tools[5].
That gap is an opportunity for firms willing to start carefully. Of construction businesses in the same survey, 31.5% had not tried to use AI at all; among the rest, the most cited barrier was lack of AI expertise, ahead of cost[6]. The jobs worth handing off first are the repetitive office tasks that eat evenings, not anything that touches gas safety.
Admin tasks AI handles well
| Task | What the tool does | What a person checks |
|---|---|---|
| Customer letters and emails | Drafts a clear reply from your notes: quote follow-ups, appointment changes, complaint acknowledgements | Facts, tone, prices and promises |
| Job reports | Turns engineer notes and photos into a tidy report for the customer or landlord | Technical accuracy and safety wording |
| Quote write-ups | Formats the engineer's survey notes into your quote template | Every price, item and exclusion |
| Summaries | Summarises long email chains with agents, landlords or main contractors | That nothing important was dropped |
| Data tidying | Cleans customer lists, splits addresses, finds duplicates | A sample of records before import |
| Policies and templates | Drafts a first version of a handbook section or form | Legal accuracy, ideally with advice |
| Call answering | Answers routine calls, takes job details and passes them to the team | Call notes, triage rules and escalations |
These match how businesses already use AI. Marketing and administration were the most common areas for AI use or plans in 2025, each cited by 72%[3], and the main reason given for adopting AI was to increase efficiency or productivity, cited by 65% of current and prospective users[7].
What to keep human
- Any decision about whether an appliance is safe, and any safety advice beyond your written triage card.
- Final prices, discounts and credit decisions.
- Disciplinary decisions, dismissals and anything sensitive about staff.
- Legal letters, debt claims and responses to formal complaints, unless reviewed by someone qualified.
- Anything you would not be comfortable explaining to the customer if they asked how it was produced.
AI tools can write confident nonsense. A drafted landlord report that invents a reading, or a quote that leaves out a flue extension, can cost far more than the time saved. Keep a person responsible for every document that leaves the office.
The human check is the system
Most businesses already work this way: 84% of those using AI reported at least some human checking of AI outputs, and 67% reported significant checking[4]. Build the check into the process rather than leaving it to good intentions.
Give the tool good input
Paste the engineer's actual notes and readings, not a vague description.
Ask for a specific output
For example: 'Write a plain-English summary for a landlord of this boiler service, under 150 words, without adding any readings or recommendations not in the notes.'
Check against the source
Compare every number, date and recommendation with the original notes.
Edit for your voice
Make sure it sounds like your firm and says nothing you would not say.
Record who approved it
Initial the job note or system record so you know who checked it.
Writing instructions that work
The quality of the output depends on the instruction. Vague requests produce vague, generic text that sounds nothing like a local heating firm. Good instructions say who the reader is, what the output is for, what to include, what to leave out and how long it should be. Save the instructions that work in a shared document so everyone in the office uses the same ones.
- Quote follow-up: 'Write a short, friendly follow-up email to a homeowner who received a boiler quote five days ago. Mention the quote number, offer to answer questions and do not offer a discount.'
- Job summary: 'Turn these engineer notes into three bullet points a letting agent can forward to a landlord. Keep all readings exactly as written and add nothing.'
- Complaint acknowledgement: 'Draft an acknowledgement that thanks the customer, says we are looking into it and gives a date for a full reply. Do not admit fault or explain the cause.'
- Data tidy: 'Split this list into separate columns for name, first line of address, town and postcode. Flag any row you are unsure about rather than guessing.'
Notice what each instruction forbids. Telling the tool what not to do, such as inventing readings, offering discounts or admitting fault, is as important as telling it what to write.
Customer data and security
Customer names, addresses, phone numbers and the fact that someone is elderly and living alone are personal data. Before using any AI tool with them, check where the data goes, whether it is used to train the provider's models, and whether you can switch that off. Use business accounts rather than personal ones, limit access to the staff who need it, and add the tool to your privacy notice and records.
Security matters as much as privacy. In 2025, 43% of UK businesses reported a cyber security breach or attack in the previous 12 months[8]. New tools mean new logins and new places where data sits. Turn on two-factor authentication, remove access when staff leave and never paste passwords or bank details into a chat tool.
A 30-day start
- Week 1: list the five admin tasks that take the most office time. Pick one low-risk task, such as quote follow-up emails.
- Week 2: write a short instruction and a checking rule for that task. Try it on ten real examples and compare with what you would have written.
- Week 3: if it saves time and passes the checks, use it for all of that task. Note any errors and adjust the instruction.
- Week 4: measure the time saved and decide whether to add a second task.
- Throughout: tell staff what the tool is for, what it must not be used for and who checks the output.
Bringing the team with you
Staff often worry that AI is a step towards replacing them. Be clear about the purpose: taking the repetitive typing off their plate so they can spend more time with customers, engineers and the diary. Involve the office team in choosing which task to try first, and let them set the checking rules. People who help design the process are far more likely to make it work, and far more likely to tell you when the tool gets something wrong.
Write a one-page policy covering approved tools, what may and may not be pasted into them, who checks outputs and what to do if something goes wrong. Review it every six months as the tools change.
Phone calls: a special case
Missed calls are often the most expensive admin gap, because a caller who reaches voicemail may ring the next firm. AI call answering is one of the more developed tools for trades. Ava, for example, answers calls, takes the job details, follows your rules for gas-smell calls and sends the job to your team; creating jobs directly in job management software is still planned rather than live. Whatever you use, write the triage rules yourself and review the call notes regularly, as you would for a new member of staff. Our guide to training office staff on the phones applies to an automated answerer as much as a person.
Example: one evening back each week
Example (fictional): the owner of a three-engineer firm in Gloucester spends Tuesday evenings writing up quotes and chasing unanswered ones. She starts using a text tool to turn her survey notes into the firm's quote template and to draft follow-up emails at day three and day ten. She checks every price and item herself. After a month, quotes go out the day after the survey instead of three days later, follow-ups never get forgotten, and her Tuesday evenings are free.
Frequently asked questions
Is it safe to use AI tools with customer data?
It can be, if you use a business account, know where the data is stored and whether it is used for training, limit access and mention the tool in your privacy notice.
What admin should a small firm hand to AI first?
Low-risk, repetitive writing: quote follow-ups, appointment emails, job report drafts and summaries, with a person checking every output.
Can AI give customers safety advice?
Only your written, approved safety wording, such as the steps for a gas smell. Decisions about whether an appliance is safe must stay with qualified engineers.
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 (2026). 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) ↑
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) (2025). AI adoption survey report 2025 (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2025). Cyber security breaches survey 2025 (opens in a new tab) ↑