Key takeaways
- Construction is more cautious about AI than most sectors[1], which is a chance to adopt it with controls from day one.
- The biggest practical risks are confident wrong answers, unsafe technical advice, customer data pasted into the wrong tool and AI-assisted scams.
- Phishing is already the most common cyber attack on UK businesses[2]; AI makes those messages more convincing.
- Keep a person in charge of anything involving safety, money or a legal record, and write that down in a one-page policy.
- Customers are split on AI[3], so be open about where you use it and make it easy to reach a person.
Why caution is not the same as risk control
Plumbing and heating firms are sceptical about AI, and the numbers show it. In 2025, 88% of UK construction businesses neither used nor planned to use AI, against 80% of businesses overall[1]. Around one in six UK businesses were using at least one AI technology[4].
Source: Department for Science, Innovation and Technology (GOV.UK), 2025[1]
Show the data
| Item | Value |
|---|---|
| Construction | 88% |
| All businesses | 80% |
Caution is reasonable, but it does not protect you on its own. Staff are already using free chatbots on their phones to write quotes, look up fault codes and draft emails. Customers are receiving scam messages written by AI. The question is not whether AI touches your business, but whether anyone has decided the rules.
Five ways AI goes wrong in a trade business
These are the failure types that matter for a firm working on gas, water and heating systems in people's homes.
- Confident wrong answers. Language models produce fluent text whether or not it is correct. A fault-code explanation, a regulation summary or a part compatibility answer can be wrong in a way that sounds right.
- Unsafe advice to customers. A chatbot that tells a caller how to reset a boiler that smells of gas, or how to bleed a system with water near electrics, creates a real hazard. Gas work in particular is restricted by law to approved persons[5].
- Data leaks. Customer names, addresses, door codes and vulnerability details pasted into a general tool may be stored or used in ways you did not intend.
- Scams that look genuine. AI makes phishing emails, fake supplier invoices and cloned voices more convincing.
- Lost trust. A customer who feels tricked into talking to a machine, or who receives an obviously generic message, may not come back.
A risk register you can copy
| Risk | Warning sign | Control | Owner |
|---|---|---|---|
| Wrong technical answer | Staff quoting chatbot answers on fault codes or regulations | Check against manufacturer manuals and official guidance before acting | Lead engineer |
| Unsafe customer advice | Any automated reply mentioning resets, gas or electrics | Fixed safety scripts; gas and CO calls always routed to a person and the emergency number | Owner |
| Customer data in public tools | Addresses or notes pasted into free chatbots | Business accounts only; strip personal data first | Office manager |
| Invented content in records | Readings or tests in a report that were never done | Engineer checks every AI draft against forms and photos | Each engineer |
| Phishing and fake invoices | Urgent payment requests, changed bank details | Phone the supplier on a known number before paying | Whoever pays bills |
| Customer distrust | Complaints about not reaching a person | Say when AI is used and offer a person at any point | Owner |
Data and security risks in numbers
Cyber attacks are not a big-company problem. In the 2025 government survey, 43% of UK businesses reported a breach or attack in the previous 12 months[6]. Phishing was the most common type, experienced by 85% of businesses that had a breach[2], and 35% of micro businesses identified phishing attacks[7]. Only 9% of micro businesses were aware of the Cyber Essentials scheme[8].
AI raises the quality of these attacks. The old clues, such as clumsy wording and odd greetings, are disappearing. The defence is process: no change to bank details without a phone call to a number you already hold, no password resets from links, and two-step login on email and accounting.
On personal data, UK GDPR requires that data be limited to what is necessary, kept accurate, and protected with appropriate security[9]. In practice: business accounts with clear data settings, no customer details in free tools, and a habit of removing names and addresses before using AI to draft anything.
Human checking is the main control
The most effective control is also the simplest: a named person reads every AI output that leaves the business. Most firms already do this. In 2025, 84% of UK businesses using AI reported at least some human checking, and 67% reported significant checking[10].
Decide in advance which outputs need which level of check. A social media caption needs a quick read. A quote, a job report or anything to do with gas safety needs the person responsible for the work to compare it with the facts.
If an AI output involves safety, money or a legal record, the person who would answer for it in a complaint is the person who approves it.
Being open with customers
Public opinion is mixed. Around four in ten UK adults expect AI to have a positive effect on them personally, and three in ten expect a negative one[3]. That split is a reason to be straightforward, not to hide AI use.
- If calls or chats are handled by AI, say so in the greeting and offer a person on request.
- Mention AI use in your privacy notice in plain words.
- Never let an automated system make promises about price, arrival time or safety that a person has not agreed.
- If a customer complains about an AI interaction, treat it like any complaint: apologise, fix it and record what changed.
A one-page AI policy
List approved tools
Name the tools staff may use and the account type for each. Anything else needs the owner's say-so.
List banned uses
No AI for gas safety records, readings, unsafe-situation decisions, or advice to customers on gas or carbon monoxide.
Set data rules
No customer names, addresses, access codes or health details in tools outside your business accounts.
Set the check
Who approves each type of output, from social posts to quotes and reports.
Report problems
Anyone who spots a wrong or unsafe output reports it the same day; the owner reviews the list monthly.
Training staff to spot AI mistakes
A policy only works if staff recognise a bad output when they see one. Spend twenty minutes at a team meeting going through real examples: a fault-code answer that was wrong, a draft reply that invented a detail, a scam email that looked like a supplier. Ask engineers to bring their own examples next time. The aim is a habit of asking, where did this come from, and how would I check it?
New starters need the same briefing in their first week. Apprentices and temporary engineers are often the heaviest users of chatbots for technical questions, and the least able to spot when an answer is out of date or wrong for the appliance in front of them.
Example: a near miss
Example (fictional): an office assistant at a four-person firm in Norwich used a free chatbot to answer a customer's late-night message about a boiler showing a fault code. The draft suggested pressing reset several times. The customer had also mentioned a faint smell near the boiler. Because the firm's policy said any message mentioning a smell goes to the engineer on call, the assistant forwarded it instead. The engineer phoned, gave the emergency number and safety advice, and the gas emergency service attended. The policy did its job because it was written down before anyone needed it.
Questions to ask before adding any AI tool
Most AI risk in a small firm comes from tools that nobody chose on purpose. Run every new tool past the same short list before anyone uses it with real customers or real jobs.
- What will it do, in one sentence, and what will it never do?
- What customer or staff data will it see, and where is that data stored?
- Can we turn it off quickly if something goes wrong, and who can do that?
- Who checks its output, and how often?
- What happens when it does not know the answer: does it say so, or guess?
- If a customer complains about it, who responds and what do they say?
If you cannot answer all six, the tool is not ready for live use. Run it on old jobs or internal tasks first, record what it gets wrong, and come back to the list.
Next steps
Start with low-risk uses such as drafting review replies and tidying job notes, where a person always checks the result. Then look at how customers react to AI on the phone before you put AI between you and your callers. Lack of expertise is the most common barrier construction firms report[11]; a short written policy is the cheapest way to build it.
Frequently asked questions
What is the biggest AI risk for a plumbing business?
Unsafe or wrong advice reaching a customer without a person checking it, especially on gas, carbon monoxide or electrics. Keep those calls with a person and the emergency number.
Do we need to tell customers we use AI?
Being open is the safer course. Say so in call greetings and your privacy notice, and always offer a person on request.
Is it safe for engineers to use free chatbots on site?
For general questions, with care. Never paste customer details, and check any technical answer against the manufacturer's manual or official guidance before acting on it.
How do we protect against AI-written scam emails?
Rely on process rather than spotting bad wording: verify bank detail changes by phone, use two-step login and never reset passwords from email links.
References
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) ↑
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) (2025). AI adoption survey report 2025 (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (1998). The Gas Safety (Installation and Use) Regulations 1998, regulation 3 (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2025). Cyber security breaches survey 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) ↑
Department for Science, Innovation and Technology (GOV.UK) (2025). Cyber security breaches survey 2025 (opens in a new tab) ↑
legislation.gov.uk (2016). Regulation (EU) 2016/679 (UK GDPR), Article 5: principles relating to processing of personal data (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 Prevent or Delayed) (opens in a new tab) ↑