Customer service touchpoints you should never automate

A man's grandmother died. He needed to fly interstate for the funeral. Before booking, he asked the airline's chatbot whether he could claim a bereavement discount.

The chatbot said yes. He booked the ticket. The airline said no. When he took them to tribunal, their defence was that the chatbot was essentially a separate entity, not their responsibility. The tribunal called that argument remarkable. They lost. (Moffatt v. Air Canada, 2024 BCCRT 149.)

If a company with an entire legal team didn't think through what their automated system was communicating to customers, what does that mean for the rest of us? Your automated system speaks for you. Whether you know what it's saying or not.

Why are Australian clients specifically afraid of automated customer service?

Not worried. Afraid.

Qualtrics surveyed 1,500 Australians in 2025. Fifty-eight per cent said they were afraid that if something went wrong, they wouldn't be able to get to a real person. That's the specific fear: not that AI is impersonal, not that it's imperfect, but that it will stand between them and a human when something actually matters.

Only 33 per cent of Australians trust organisations to use AI responsibly. The lowest of any country in the study.

Zoom and Morning Consult found that 51 per cent of Australians are more frustrated by AI customer service than any other country surveyed. And 86 per cent said they'd stop buying from a brand entirely if their issue wasn't resolved.

This isn't a vague preference for human connection. It's a specific, documented fear that your automated systems will be the thing standing between your client and help when they actually need it.

"Your automated system speaks for you. Whether you know what it's saying or not."

What makes a customer service touchpoint a "bucket three" task?

The bucket framework gives you one question to sort every task in your business: would the client you'd least want to lose feel less valued knowing a system handled this?

Yes means bucket three. It stays with you. Always.

Bucket three isn't about the complexity of the task. It's about what the task means to the client. And the tasks that belong there share three characteristics.

1. The client is worried or upset

When something has gone wrong, or when a client feels uncertain about money, a deadline, a job outcome, or their own decision-making, they need to know a real person is across it. An automated response in that moment doesn't just fail to help. It actively confirms their worst fear: that they're dealing with a system that doesn't know them and doesn't care.

2. The client is making a significant decision

Hiring you. Agreeing to a scope of work. Signing something. Committing budget. These moments carry weight. When a client is deciding whether to trust you with something that matters, an automated response tells them the decision doesn't matter to you.

3. The client is in a relationship, not a transaction

When something happens, a question, a concern, a check-in, and it routes through automation, you're telling a client who chose you for the human element that the human element is no longer available. That's the moment they start comparing.

The three-bucket framework gives you the full decision filter and how to sort every task in your business.

Which specific customer service touchpoints should never be automated?

This section gives you the specific list, and for each one, what happens when you get it wrong.

1. Client complaints

This is the clearest bucket three task in any service business. When something has gone wrong, your client needs two things: acknowledgement and accountability. Both require a person.

In 2023, Klarna stopped hiring altogether and went all in on AI for customer service. By 2024, their CEO was publicly declaring AI could do every job a human does. About a year later, satisfaction had collapsed and they were hiring human agents again. The CEO told Bloomberg: from a brand perspective, he believes it's critical that customers always know a human is available if they want one. (Bloomberg, May 2025.)

An automated response to a complaint doesn't just fail to resolve the issue. It tells the client their problem isn't worth a real person's attention. Most of them won't tell you that. They'll just leave.

Roy Morgan's Snap SMS Survey (October 2025) found that 65 per cent of Australians say AI creates more problems than it solves, up eight points since 2023. An automated complaint response is one of the fastest ways to become a data point in that statistic.

2. First real conversations with new clients

The initial enquiry can be acknowledged by an automated system, and should be, so the person knows you've received it and when to expect a reply. But the first real conversation, the one where they tell you what they actually need, has to involve a human.

This is where the client is evaluating whether they can trust you. An automated response that attempts to qualify, propose, or move the conversation forward without a human reading their actual situation is making commitments on your behalf that you haven't agreed to. That's the airline problem, scaled down.

When a new client reaches out for the first time, ask yourself: is there a real person on the other side of this first interaction, or is there an automation pretending to be one? The distinction matters, and clients can feel it.

3. Sensitive financial or legal conversations

An ATO letter arrives and your bookkeeping client panics. A tradie's quote comes in higher than expected and the homeowner needs an explanation. A consultant's client discovers the project has gone over scope.

These conversations involve money, trust, and often a degree of client vulnerability. Routing them through automation, even a well-written automated response, signals that you're not available for the hard conversations. That signal lingers long after the conversation is resolved.

4. End-of-relationship conversations

When a client says they're going to pause, move on, or not renew, that conversation needs a human. Not to talk them out of it. To understand what happened.

Thirty-one per cent of Australians say nothing after a bad experience. They don't complain. They don't ask for a refund. They just leave.

The clients who do reach out to tell you they're leaving are giving you a gift: the chance to understand what went wrong, make it right if possible, and leave the relationship in a state where they'd refer you anyway. Routing that conversation through an automated system closes that door permanently.

"The clients who do reach out to tell you they're leaving are giving you a gift."

5. High-stakes onboarding moments

The intake form can be automated. The scheduling can be automated. But the moment when a client has said yes to something significant, a major project, a retainer, a service they've never used before and are nervous about, that moment deserves a real person.

A personal message or call at the start of a high-stakes engagement tells your client: you're not just a transaction in my system. It costs you ten minutes and it's remembered for the life of the relationship.

6. Anything the client is worried about before it becomes a complaint

A client who sends a message asking "just checking in on the progress of X" is not asking for an automated status update. They're signalling mild anxiety. The automated response might answer the literal question. It won't address what's actually going on.

When you train yourself to read client messages for what they're actually saying, not just what they're literally asking, you start to catch these moments before they become complaints. That's a human skill. It doesn't automate.

How do I know if my automation is handling something it shouldn't?

One exercise. Write down every way a new client first interacts with your business: enquiry form, website, DM, email, phone. Then go through each one and ask: is there a human here, or is there an automation pretending to be one?

If the answer is mostly automated, that's where you're losing people. Not in your product. Not in your pricing. In the gap between what you set up to run without you and what your client actually needed from a real person.

The silence test

Thirty-one per cent of Australians say nothing after a bad experience. They don't complain. They don't ask for a refund. They just stop calling.

The next time a long-term client goes quiet, or a promising enquiry drops off without explanation, consider whether an automated touchpoint might have been the last thing they experienced before they decided to look elsewhere.

You won't always know. That's the point. The silent leaver is the hardest problem to measure and the most important one to prevent.

The practical audit

Pick one client who has been with you for two years or more. Trace the last three touchpoints they had with your business. How many were automated? How many involved a real person? If the ratio is mostly automation, you already know what to work on.

So what should you automate in client-facing interactions?

Automation belongs at the front door in a specific, limited role: acknowledging contact, confirming information, and handling the administrative details that surround real conversations.

The common thread: none of these tasks require your judgement, your relationship knowledge, or your accountability. They just need to happen consistently. Automation does that better than you do.

The moment a task requires you to know this specific client's situation: stop. That's a bucket three moment, and it needs you.

The airline didn't brief its chatbot properly. It didn't check what the chatbot was telling people. And when it went wrong, the tribunal held them responsible for every word it said.

Your automated system speaks for you. The question is whether you know what it's saying.

Nardia Barrett

Founder of BossKit. Small business systems consultant with over a decade in mortgage broking and eight years running WEBS, a free networking group for small business owners in Western Sydney. BossKit helps service businesses work out what to automate, what AI can assist with, and what to keep human.