Research note · September 2026 · about 6 minutes · this is the thinking behind what Cutting Hedge builds
When a tradesperson looks at an AI tool, “does it work?” is only half the question. The other half is: will it make me look bad in front of a customer?
The short version
- Among small businesses already using AI, the number-one problem is accuracy (46%).
- Trust isn’t a feeling you add afterwards. It comes from limits: what the system can do, what it can’t, and who’s responsible.
- The more a mistake would cost, the less the system should do on its own.
Both sides are asking the same things
You’re wondering whether it’ll get things wrong, whether customers will notice, whether their details are safe, and whether the business will still feel like yours.
Your customers are wondering whether it understood the problem, whether they can get a real person, whether the price is fair, and what happens if it’s urgent.
What the evidence shows
The US Federal Reserve’s 2026 report on small employer firms (a 2025 survey of 6,525 businesses) asked AI users what was hardest about it.
46%
of small businesses using AI said accuracy was a challenge — the most common answer. 43% struggled to make tools fit how their business works.
Among businesses planning to start, 54% worried about finding tools that fit their needs, and 37% about the time it would take to set up and learn.
Irish figures point the same way. ICT Skillnet and AI Ireland’s 2025 AI Readiness Pulse found data privacy and security fears held back 18% of respondents, resistance to change 12%, and trust issues 9%. Confidence was low overall, and people wanted practical tools and real examples, not theory.
In construction, the Royal Institution of Chartered Surveyors’ 2025 report frames the goal as AI that supports “trusted and safe practice” and serves the public good. Its own figures show how rarely that has been achieved at scale so far: fewer than 1% of organisations have AI embedded across the business.
| Trust question | What you — or your customer — want to know |
|---|---|
| Accuracy | Will it give a correct, useful answer? |
| Relevance | Does it understand this trade, this area, this job? |
| Control | Can a person check it, change it, stop it, or overrule it? |
| Honesty | Is it clear when something is automated? |
| Privacy | What happens to names, addresses, photos and job details? |
| Responsibility | Who answers for it if it’s wrong? |
| Recovery | Can mistakes be put right and learned from? |
Match the trust to the stakes
Drafting a Facebook caption is not the same as responding to a gas leak. Summarising a job note is not the same as signing off a structural calculation. How much you let a system do should depend on what a mistake would cost.
| Task | How much it can do alone | Why |
|---|---|---|
| Drafting a marketing post | A lot, with a quick look | A small wording slip rarely does harm |
| Summarising an enquiry | Some, with a check | Addresses, urgency and job type get misread |
| Replying to a missed call | Some, within clear limits | It can acknowledge the call — not diagnose or promise |
| Preparing a draft quote | Some, with your approval | Missing work or wrong quantities cost money and cause disputes |
| Safety or compliance advice | Very little without a qualified check | Bad advice can cause real harm |
| Final price or contract | None — your decision | The commitment is yours |
A trustworthy system isn’t the one that automates the most. It’s the one that automates the right bit and is upfront about its limits.
Five things that build trust
1. Be clear about what it is
Nobody should think an automated message came from you personally. Say what it does and what happens next, and don’t suggest anyone has looked at the job when nobody has.
2. Always leave a way to reach a person
If it’s urgent, complicated or upsetting, the customer needs a route to a human. For a small business that’s a reliable callback, not a 24-hour call centre. Automation should never be a dead end.
3. Limit what it’s allowed to decide
It can gather details, draft a reply or set a reminder. It shouldn’t decide something is safe, promise a price, guarantee a date, or agree to work.
4. Look after people’s information
You handle addresses, phone numbers, photos of people’s homes, key-safe codes, invoices, and sometimes personal circumstances. Before using any AI service, find out what gets sent, where it’s stored, who can see it, how long it’s kept, and whether it’s used to train someone else’s model.
5. Keep score, and fix what goes wrong
Note the errors, complaints, missed escalations and the times you had to step in. The US National Institute of Standards and Technology’s voluntary AI risk framework comes down to the same cycle: be clear on the purpose, spot the risks, test what comes out, decide who’s responsible, watch the results, and adjust.
What the customer should experience
Customers don’t need to know how it works. They need it to feel clear and respectful: their message acknowledged, only relevant questions asked, a clear next step, and an easy way to reach you.
A bad experience is the opposite: it hides that it’s automated, asks the same thing twice, sounds overconfident, misses that something’s urgent, or makes the customer repeat everything once they finally reach a person.
Something like this works because it doesn’t promise more than it can deliver:
“We use an automated assistant to take the basic details while we’re out on jobs. One of us will look at your request before giving any advice or a price.”
Trust inside the business counts too
Nobody trusts a system that’s dropped on them from outside. It goes better when you can test it on real but low-stakes examples, see how mistakes get fixed, and have a say in the wording and the rules. So instead of a talk about AI, run it through a handful of situations and ask one question: what must this never say?
| Test it with | What you find out |
|---|---|
| A normal enquiry | Whether it captures the right details |
| A half-finished enquiry | Whether it asks sensible follow-up questions |
| Someone outside your area | Whether it gives a clear answer without wasting anyone’s time |
| An urgent or safety-related call | Whether it gets it to a person fast |
| A draft that’s wrong | Whether you can easily fix or bin it |
Trust is what wins the work
Sooner or later, everyone will have access to much the same AI tools. What will set a business apart isn’t the tool — it’s how carefully it’s used. Getting back to people promptly, being straight about the process, looking after their details and keeping a person accountable is what makes a customer pick you.
What this research can’t tell you
The Federal Reserve survey covers US businesses with employees, isn’t specific to trades, and uses a convenience sample rather than a random one. The ICT Skillnet survey doesn’t publish its sample size. None of these studies asked Irish customers how they feel about automated replies from tradespeople — and I haven’t found a trustworthy figure for how many customers would rather deal with a person. So I haven’t used one.
Tell me what matters to you
I’m running a short survey asking local tradespeople which admin problems matter most to them and how comfortable they are with digital tools. It takes 6–8 minutes, and leaving your contact details is optional.
Keep reading
- 10 practical ways AI could help Irish trades businesses
- Can AI qualify customer enquiries without losing the human touch?
- Can AI reduce the burden of construction documentation?
- Could AI support apprentices without replacing experienced tradespeople?
Sources
- Federal Reserve Banks — 2026 Report on Employer Firms (Small Business Credit Survey, fielded September–November 2025)
- ICT Skillnet and AI Ireland — AI Readiness Pulse: what Irish businesses are really saying about AI in 2025
- RICS — Artificial intelligence in construction report (2025)
- NIST — Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1, July 2024)