What can go wrong with an AI receptionist for a consulting business

We sell an AI receptionist, and I still wouldn't put one on every consultant's phone. Your callers aren't asking about opening hours. They may be ringing about a sale, a redundancy, a dispute or a board problem, and that changes what can go wrong.

The evidence is thin. No independent test of these tools on UK consulting lines turned up, and the surveys below mostly come from companies with something to sell, asking consumers rather than finance directors. Read what follows as indirect evidence plus my judgement.

Problem 1: Confidential information lands in a transcript

This is the risk I'd put first. An AI receptionist turns speech into text, and that text is stored somewhere. An existing client might ring and say, "It's about the sale to Acme, and the redundancies won't go down well."

That detail is now in a summary, on your phone and on a supplier's system. Your contract or NDA may promise to keep it confidential. If it names people, it's personal data too, and UK data protection applies. Details about someone's health or trade union membership are more sensitive still.

How to reduce it:

• Give existing clients a direct number or your mobile, so they never reach the AI.
• Keep the script short. It should take a name, a number and a one-line reason, and it shouldn't ask probing questions.
• Have it say, "Please don't share confidential details. I'll ask them to call you back."
• Ask the provider where data's stored, how long it's kept, who can see it and if it's used to train models.
• Check your client contracts for any rule on using third-party processors.
• List the provider in your privacy notice.

Problem 2: A caller treats it as advice

An AI can state something false with total confidence, and you can't necessarily blame the machine.

Air Canada found that out in Moffatt v Air Canada. A Canadian tribunal made the airline answer for wrong information from its website chatbot, and wouldn't accept that the bot was a separate entity. No UK court is bound by that, but I still wouldn't want to argue "the AI said it" here.

For a consultant, the risk has an extra edge. Your whole product is judgement. If a receptionist tells a caller, "Yes, you should be able to make those people redundant", that sounds like your advice. And your professional indemnity insurer may have views on it, so ask.

If your work touches regulated areas, it's more serious. Investment business, reserved legal work and immigration advice all have legal limits on who can do them. Don't let a machine sound as if it's doing any of them.

The risky areas are:

• Advice. Any answer to "Should I...?" or "Can we...?"
• Fees. "We charge £500 a day" is a promise you didn't make.
• Promises. "She'll have a report to you by Friday."
• Scope. "Yes, we do employment law."
• Names. Confirming who your clients are, which may breach confidentiality.

How to reduce it:

• Write a no-advice rule into the script: "I can't advise, but I'll take your details so they can help."
• Give it a written line on fees, usually a range or "they'll confirm that".
• Tell it what you don't do.
• Tell it never to confirm or deny who your clients are.
• Read the summaries weekly, and fix whatever's drifted.

Problem 3: Some callers don't want to talk to AI

The evidence is fairly consistent here.

Gartner's survey is nearly three years old, and I can't say which countries it covered. The hang-up figure is what people say they'd do, and it comes from a business that sells human answering. I couldn't find data on what callers really do, or any on senior business buyers.

I'd expect this to matter a lot in consulting. If someone's about to spend £20,000 on your judgement, their first contact sets the tone. On the other hand, the fair comparison is often AI against nobody. And voicemail is a poor fallback: 6% of UK adults told YouGov, in its survey for Answer4u, that they'd leave one.

How to reduce it:

• Open by saying it's an AI. In a Five9 survey, 76% of 1,000 UK respondents rated being told as very or extremely important. It's vendor research, and I'd do it regardless.
• Offer a human route. In the same survey, across three countries, 55% trusted AI to resolve an issue when a human option was available, against 26% when none was.
• Keep the greeting short: "I can take your details and Laura will ring you back by 4."
• Read your first 20 call summaries, and listen if your provider records them. If calls are recorded, tell callers.

Problem 4: A caller in real distress

Most consultants won't meet this. Some will. If you work with business owners in financial trouble, advise on restructuring, handle redundancies, or coach leaders under pressure, you could get a call from someone at the end of their rope. An owner facing insolvency may not be ringing about insolvency.

An AI receptionist isn't equipped for that, and neither is ours by default. If a caller could be at risk, you must set what it says, and test it. The official routes are these, from NHS and Samaritans guidance:

• If someone's life is at risk, call 999 or go to A&E.
• For urgent mental health help, call NHS 111 and select the mental health option.
• Samaritans are free to call on 116 123, 24 hours a day, in the UK and Ireland.

Keep to those lines. It shouldn't counsel, and it shouldn't ask questions. It should try to transfer the call to you, and the call should reach your summary so you can act on it. I'm not a clinician, so have your own wording approved by someone qualified before it goes live. I only checked the NHS page, so confirm the local route if your clients are in Scotland, Wales or Northern Ireland.

Test it by ringing it yourself and saying the words a frightened person would use, not the tidy ones. If it doesn't work every time, don't use the AI on that line.

Other urgent calls need a route too. A client with a data breach, a legal deadline or a press problem can't wait for a summary. Tell clients how to reach you for those, and don't let them depend on the AI number.

Problem 5: It mishears names and companies

Names are where a consultant gets hurt. Getting "Mr Patel of Northgate Holdings" wrong means ringing the wrong person, or not being able to ring anyone.

I found no reliable figure for AI receptionists mishearing UK callers. So test it with people who sound like your clients, on a bad mobile signal. Have it read names and numbers back. Get the caller's number on every summary, and text to confirm.

Problem 6: Recording a conversation you promised to keep private

Many of these tools record or transcribe calls, and a recording of a person is personal data.

• The ICO says callers must be told that you're recording, and why. That line sits in its guidance on monitoring workers, but I'd apply it to clients too.
• Your privacy notice should explain any third party that handles calls.
• Recording a confidential client call may break your own contract, even if the law allows it.

Providers differ. Fasthosts says its AI receptionist doesn't record calls. Ours can record calls, and the greeting says so. For confidential work, I'd think hard before switching recording on. Whichever you pick, find out what's kept and for how long.

I found no UK rule that specifically requires a business to say a caller is talking to an AI on an inbound call, but my search was cut short, so that's unconfirmed. Tell people anyway. Don't use an AI to make outbound marketing calls without taking advice first.

Problem 7: Paying for calls you didn't want

Pricing is usually per call or per unit of usage, so sales pitches and spam eat your allowance. Ask what counts as a call, and what happens when you pass the limit. I'd expect most consultants to get few calls, so this should be small. Our plans include a monthly usage credit, and any extra goes on your bill.

Problem 8: Summaries pile up during a workshop

I'd bet on this one happening before any of the others, and the AI isn't to blame. Two days in front of a client, 10 summaries waiting, and the caller who needed an answer has moved on.

Book a daily slot for call-backs before you go live, even on workshop days. Use the follow-ups in our text templates. The rest of the chain is in why consultants lose clients after the first enquiry.

What ours does, so you can compare

• It says it's an AI at the start of the call.
• It answers questions from what you've told it about your services, areas, prices and hours.
• It takes the caller's details, and you get a summary.
• On its own, it doesn't book. You call back and book. The full AI Business Engine adds booking.
• It can record calls, and the greeting says so.
• You choose how urgent calls reach you. You can have a separate emergency number that comes straight to you, and an enquiry number the AI answers.
• On a number the AI answers, it only knows a call is urgent if the caller says so. Then it tries to transfer the call to you. If you can't pick up, it takes their details and puts them in your summary.

We make the script changes you ask for after set-up. The safety wording is still yours to agree, get approved and test. No provider, us included, can take these risks away. Whether one's worth having at all is a separate question, and this review answers it for consultants.

David Bell, founder of Real Business Systems

David helps small UK businesses get found, answer every enquiry and keep customers coming back. We sell some of the services mentioned here, and we say so where they come up.