What to do when AI gets it wrong in front of a customer

When AI makes a mistake in front of a customer, treat it as your own mistake: correct it quickly and in person, then put a human check where it slipped through.

Sooner or later it will happen. If you use AI to draft replies, answer questions on your website or write up quotes, at some point it will tell a real customer something confidently wrong: a price that was never yours, say, or a refund policy it made up because the question seemed to deserve one. How much harm that does depends less on the error itself than on what you do in the next hour, and on whether anyone was checking in the first place.

What follows is general information, not legal advice. If a mistake has cost a customer real money, talk to your solicitor or insurer early.

Why does AI get things wrong so confidently?

AI tools produce the most likely-sounding answer, and a likely-sounding answer is not always a true one. When the tool doesn’t have the fact it needs, it rarely says so. It fills the gap with something plausible, in the same calm tone it uses for everything else. That is what people mean by AI hallucination, and the business risk is simple: the wrong answer arrives looking exactly like a right one.

The errors that reach customers tend to take a few familiar shapes:

• Invented facts. A delivery date, a stock level or a discount that nobody gave it.
• Stale facts. Last year’s prices or hours, picked up from an old document it was handed.
• Invented policy. Asked something your instructions don’t cover, it guesses what a reasonable business would do and states it as yours.
• Wrong tone. A cheerful reply to a complaint, or a breezy one to someone whose order went missing a week before Christmas.

Who is responsible when AI gives a customer the wrong answer?

Assume you are. Whatever a customer reads from your business, they read as coming from you, whether a person or a tool wrote it.

There is a well-known case on exactly this. In 2024 a Canadian tribunal ruled against Air Canada after the chatbot on its website told a grieving passenger he could claim a bereavement fare after he had already travelled, which was not the airline’s actual policy. The airline argued, in effect, that the chatbot was responsible for its own words. The tribunal didn’t accept that, and the airline had to pay. As far as I know no Irish court has had to rule on a case like it, and I wouldn’t bet a business on one seeing it differently.

Customers forgive a mistake far more readily than they forgive a business that blames its software.

What to do when an AI error reaches a customer

Correct it yourself, quickly, then deal with whatever let it through. In order:

1. Contact them the same day. A short message or call from a named person. “The reply you got earlier had our cancellation terms wrong, and I’m sorry. Here is what actually applies.”
2. Don’t hide behind the tool. If they ask, you can say an automated reply got it wrong, but the apology comes from you.
3. Decide whether to honour it. If the customer acted on the wrong answer (booked, paid, turned down another supplier), honouring it is often the cheapest way out. If nobody relied on it, a correction is enough.
4. Look for the same answer elsewhere. If the tool made up a policy once, it may have done it ten times. Search the sent folder or the chat history.
5. Fix the source, then add the check. Update the information it works from, and put a review step where the mistake got out.

Say you run a guesthouse in Kinsale and the assistant on your website tells a guest that cancellation is free up to 24 hours before arrival, when your terms say seven days. The guest cancels two days out and expects a full refund. Most owners would refund it, write the seven-day rule into the assistant’s instructions in plain words, and tell it to pass any cancellation question to a person. The cost is one room night and an afternoon. Arguing about it would likely cost a review.

The person-in-the-loop rule: where to put the human check

Put a person between the AI and the customer wherever a wrong answer would cost money, break a promise or upset someone. Everywhere else, let it run and spot-check it.

Someone should read a message before it goes if it:

• states a price, a date or a deadline
• describes a refund, cancellation, warranty or any other term
• answers a complaint, or anyone who is clearly upset
• gives advice the customer might act on (health, legal, money, safety)
• is the first message a new customer gets from you

The rest (a booking confirmation built from your calendar, an acknowledgement that an enquiry arrived, an internal summary) can usually go without a check, as long as somebody reads a handful each week. If you have written AI rules for staff, this list belongs in them. The post on writing an AI policy for a small business has a one-page version you can copy.

How to review AI output without reading every word twice

Look hardest at the parts most likely to be wrong and most expensive if they are, and skim the rest. A quick way to review AI output before it goes to a customer:

• Check every number and date against the source. This is where invented facts hide.
• Ask whether you supplied each fact. If a sentence contains something you didn’t give it, find out where it came from.
• Read the promise. Would you stand behind what this commits you to?
• Read it as the customer would. Especially the first and last lines, which set the tone.

Thirty seconds on a reply is usually enough. What you want is a habit your team will still keep on a busy Friday, and a long checklist won’t survive that.

Setup matters as much as checking. A tool that answers only from a short document you wrote (your prices, hours and terms) makes far fewer mistakes than one left to fill gaps from general knowledge. Tell it plainly what to do when it doesn’t know: “If the answer is not in this document, say a member of the team will reply, and stop there.” The post on using AI to answer enquiries and take bookings shows what that looks like on a website.

Is it still worth using AI with customers?

For most small businesses, yes, provided the checks sit in the right places. People make mistakes with customers too. The difference is that a person usually knows when they’re unsure and the AI doesn’t, so give it a narrow brief and keep a person on the answers that matter.

If you’d like help working out which of your customer messages are safe to hand to AI and where the checks should sit, the AI Assessment is where I’d start: a fixed-fee look at how your week actually runs, with a plain plan at the end.

To make AI writing sound human, give the tool a few samples of your own writing to copy, then delete the stock phrases and tidy patterns that give it away.

Who owns AI-generated content? For an Irish business, you can usually use it as your own, but the legal protection around it is thinner than for work a person wrote.

AI for architects and engineering practices in Ireland earns its keep on the writing around the design: specifications, planning correspondence and fee proposals.

AI for insurance brokers in Ireland earns its keep on the admin wrapped around the advice: chasing renewals before they lapse, writing up why one quote was recommended over another, and keeping a file that stands up to a compliance review.

AI for charities in Ireland earns its place on the admin: funding applications drafted, donor thank-yous written properly, and the reporting pack assembled without losing a weekend to it.

AI for childcare providers belongs on the paperwork, not the children: parent messages drafted, enrolment enquiries answered, and the policy folder made findable again.

AI for farming in Ireland earns its keep in the office, not the yard: scheme paperwork drafted, farm records made findable, and correspondence answered before ten at night.

AI for plumbers and electricians earns its keep on the paperwork, not the pipework: quotes drafted from a voice note, follow-ups chased, cert records kept findable.

AI features in existing software are the cheapest AI you will ever use, and your Microsoft or Google plan, CRM and accounts package probably include some already.

An AI policy for a small business fits on one page: which tools staff may use, what never gets pasted in, who checks the output, and who to ask.

AI for driving instructors is mostly a diary problem: cancellations, rescheduling, reminders and the messages you end up answering at the kerb between lessons.

AI for salon booking earns its keep on three things: no-shows, rebooking, and answering the phone while both your hands are in someone’s hair.

AI for a veterinary practice is best pointed at the admin, not the medicine: vaccination recalls, appointment reminders, and the phone that rings mid-consult.

An AI agent is a tool that carries out a job on its own rather than waiting for you to ask, deciding the steps itself and using your systems to get it done. Most small businesses do not need one yet, and the ones that do usually need exactly one.

AI pays for itself fastest in a recruitment agency on the writing and the follow-up, not on deciding who gets shortlisted. Job specs, candidate summaries, interview notes and the chase that keeps a placement alive are all repetitive, high-volume, and safe to hand over. Screening decisions are not.

AI earns its place in an Irish pharmacy on the counter admin and the phone, not anywhere near clinical judgement or the dispensing process. The wins are the repeat-prescription chase, the same twenty questions asked every day, and the paperwork that piles up behind the dispensary.

Most useful AI automation for a small business is working within days, not months. A drafting habit takes an afternoon, a single automation that runs by itself usually takes one to three weeks, and anything quoted in quarters is either much bigger than you need or badly scoped.

AI for a gym or fitness studio is most useful on the admin around the training, not the coaching: bookings, no-shows, lapsed members and the weekly content.

No business is too small for AI, but plenty are too small for the version of it being sold to them. If you are one person with a phone and a spreadsheet, the useful question is not whether to adopt AI, it is which single job in your week you would hand over first.

An AI course teaches you to use the tools; an AI consultant works out which jobs in your business are worth handing to them. Most owners need the second thing first, then the first thing to stick.

Real AI automation examples for a small business are duller and more specific than the hype suggests: a quote drafted from site notes, an inbox sorted before you open it, an invoice chased on day seven, a Monday report that writes itself.

AI for a construction small business earns its keep on the paperwork, not the building: drafting quotes, writing up site records, and chasing the documents and payments that quietly stall a job.

AI for estate agents in Ireland earns its place on the admin that eats a viewing day: drafting listing copy, chasing enquiries before they go cold, and turning scribbled viewing notes into follow-ups you would otherwise write from memory.

Writing AI prompts that actually work means giving the tool three things every time: the real context, the exact output you want, and an example of good, so it stops guessing at what "good" means.

AI vs hiring in a small business comes down to one test: automate the work that repeats to a pattern, and hire when the job needs judgement or presence.

AI email management works best for a small business when it sorts and drafts while you decide what actually gets sent, which keeps replies personal.

AI helps a Cork clinic write appointment reminders, draft patient information sheets and answer common questions faster, but anything touching patient records needs a properly configured, GDPR-compliant tool, not a free consumer chatbot.

AI helps a Cork shop write product descriptions, answer supplier and customer emails faster, and turn reviews and stock updates into social posts, without needing any technical setup.

AI is useful in a small Irish law firm for drafting, summarising and admin, as long as a solicitor checks every output and client confidentiality is settled first.

The EU AI Act applies lightly to most Irish small businesses: if you use ordinary AI tools, your duties are transparency and staff training, not paperwork.

An AI receptionist for a small business is software that answers calls, texts, or web chats when nobody is free to pick up, and either resolves the enquiry or books a slot straight into your calendar.

The best AI tools for small businesses in 2026 aren’t one big platform, but a short list matched to the job: a chatbot for drafting, a notetaker for meetings, and one scheduling tool that actually talks to your calendar.

To choose an AI consultant in Ireland, look for fixed pricing, a guarantee tied to a real outcome, and someone who names the tools before you pay anything, not after. Beyond that, most of the job is spotting who is selling hype and who is going to sit with your actual week.

Is AI actually worth it for a small business? Usually yes, but only for the dull, repeatable parts of the week, not as a blanket upgrade to everything you do.

How much does AI automation cost for a small business in Ireland? Honestly, less than most owners fear, once you split it into three plain tiers. The trouble is that the true answer starts with “it depends”, and nobody enjoys hearing that. So I’ll break the range down properly, with real figures where they exist and clear bands where they don’t.

The AI Assessment is a 60-minute look at how your week actually runs, ending with a plain plan and a four-day quick-start you can follow.

Is it safe to put client data into AI tools? Usually yes, if you pick the right tier and follow a few plain house rules that keep GDPR onside. The honest answer, for most businesses I meet around Cork, is that the tool itself is rarely the problem. The habits around it are. A lot of the worry I hear from owners is vague. “I don’t want to get in bother, so we just told the team not to touch it.” That is understandable. It usually means the team is using AI anyway, quietly, on their own phones, with no rules at all. Which is the worst of both worlds.

Seven low-risk tasks a small business can automate this month, each quick to set up and chosen to give a Cork owner back real time each week. None of them need a big project, a new platform, or a member of staff who understands code. Most are jobs you already do by hand every week: the small repetitive stuff that takes an hour here and forty minutes there until half your Friday is gone. So here are seven, with a plain reason each one is worth doing.

AI for hospitality doesn’t have to mean losing the personal touch. Here’s how Cork cafés, restaurants and hotels can use it to cut admin. Picture a Saturday in a busy spot off Oliver Plunkett Street. The floor is full, the kitchen is flat out, and somewhere between the lunch rush and the evening covers a pile of small jobs is stacking up. Three reviews to reply to. A supplier waiting on an order confirmation. Two enquiries in the inbox asking the same thing about the set menu. None of it is hard. All of it eats the hour you should be spending on the floor, or at home.

AI for accountants and bookkeepers in Ireland saves real time on drafting, summarising and chasing, but the figures still need a human eye. If you run a small practice in Cork or anywhere across Munster, you know the shape of the week already. Client work in the day, then the admin tail after: the email you owe three people, the meeting notes you meant to write up, the reminder about a VAT deadline you keep forgetting to send. That tail is where AI earns its keep. Not on the numbers themselves. On everything wrapped around them.

AI for tradespeople in Cork can take quotes, invoice chasing and job notes off your evenings, so the paperwork stops eating your nights. Picture the usual. It’s half nine at night. The van’s back, the kids are down, and you’re at the kitchen table with a cold cup of tea, working out a quote you should have sent on Tuesday. You’re grand on the tools. Nobody ever told you the job came with two hours of admin a night.

Do you need an AI consultant, or can you sort it yourself? Here’s an honest guide for Cork and Munster owners on when DIY works and when help pays. A lot of the time you can do it yourself, and I’ll tell you when. I’d rather you spend nothing and get a small win than hand me money for something you could have set up over a wet Sunday afternoon.

To know if your business is actually making money, you need one view that refreshes itself and gives a straight answer on a Monday. A lot of the owners I talk to here in Cork are flat out. The vans are moving, the diary is full, the team is stretched. Then I ask one simple question, how much did you actually make last month, and the honest answer is a shrug. That shrug is not a sign anyone is doing a bad job. It is a sign the numbers live in too many places to see at once.

ChatGPT vs Claude for business? For most Irish SMEs both are excellent, so the real win is picking the one your team will actually use. I get asked this a lot by owners around Cork, usually after someone in the family or the office swears blind by one of them. It’s a fair question. Here’s how I’d think about it, without the hype.

One team I worked with started every week the same way. Someone exported four files, cleaned them by hand, pasted them into a template, and sent it round by ten. Two hours gone before the real work even started.

There is a lot of noise about AI taking jobs. On the ground, in ordinary businesses, that is not what I see happening.