Welcome! Let's start with a story
In November 2022, a man called Jake Moffatt lost his grandmother. He went to Air Canada's website to book a flight to the funeral and asked the chatbot about bereavement fares.
The chatbot told him he could book now and claim the discount back within 90 days. That sounded helpful. But it wasn't true. The airline's real policy didn't allow refunds after booking.
Air Canada argued that it couldn't be held responsible for what its chatbot said. In February 2024, a Canadian tribunal disagreed. The tribunal said the chatbot was just part of the airline's website, and the company was responsible for all of it. Air Canada had to pay him back.
Why start here? Because this one story covers almost everything in today's level. AI can be fast and friendly. It can also be confidently wrong. And when it's wrong, the business still owns the answer, and so do you.
So, how are teams actually using AI right now?
You've probably noticed AI turning up everywhere in customer service. Let's look at some dated evidence so we're talking facts, not hype.
In February 2026, Gartner reported that 91% of the customer service leaders it surveyed felt pressure from their bosses to bring in AI. In August 2026, it reported that AI spending by service leaders had risen 38%, even though their overall budgets had grown only 2%. So the push is real, and the money is moving.
But here's the other side. Gartner also surveyed 3,566 customers in February and March 2026. 87% of them said that when a company uses AI for customer service, being able to reach a human agent is essential.
UK customers feel the same. In October 2025, a survey published by 8x8 found that 83% of UK respondents preferred speaking to a real person. Only 4% preferred a chatbot.
And look at what happened at Klarna. In 2024, it said its AI assistant was doing the work of 700 customer service agents. By May 2025, its chief executive was telling Bloomberg that investing in the quality of human support was the way forward, and the company was hiring people again.
What's the pattern? Businesses are investing heavily in AI, and customers still want a person when it matters. That gap is where you come in.
Three boxes for your everyday tasks
Here's a simple way to think about it. Picture three boxes on your desk.
Box one: AI can help. These are tasks where AI saves you time and a mistake is easy to spot and cheap to fix. Think tidying up the grammar in a reply you've already written, suggesting a friendlier opening line, or turning your rough notes into a neat bullet list.
Box two: AI with care. Here AI can do a useful first draft, but you need to check it closely before anyone sees it. Summarising a long chat. Drafting a reply about a delivery delay. Finding the right knowledge base article. The output might be 90% right, and the other 10% could be the bit that matters.
Box three: must stay human. These are decisions and moments that need your judgement, your empathy and your accountability. Deciding whether to make an exception to policy. Responding to a customer who's grieving, frightened or in financial trouble. Making the final call on a complaint. Choosing when to escalate.
So here's a question for you. Where would you put "reply to a customer whose parent has just died and who is asking to close the account"? Box three, every time. AI might help you check your spelling afterwards. The care and the decisions are yours.
Which box a task goes in can depend on your workplace too. We'll look at your organisation's rules in the next level.
How AI gets things wrong
Let's talk about the failures, in plain language.
Made-up facts. AI tools write by predicting likely words. They don't look things up the way you do. Sometimes they produce something that sounds exactly right but isn't: a policy that doesn't exist, a refund window someone made up, a phone number that goes nowhere. People call this a "hallucination". The Air Canada chatbot is a textbook case.
Overconfidence. Here's the tricky part. AI rarely says "I'm not sure." A wrong answer comes out in the same calm, polished tone as a right one. Think of a new colleague who never admits they don't know something. Charming, but you'd double-check their work, wouldn't you?
Missing context. AI doesn't know what you know. It didn't hear the tremble in the customer's voice. It doesn't know this is their fourth contact this week. It can't see the account notes unless someone gives them to it.
Wrong tone. AI can sound cheerful when the moment calls for something gentle, or stiff when a customer needs warmth.
A simple habit to start with
Before any AI output reaches a customer, run three quick checks, in this order.
First, check the facts against your own policy or knowledge base. Is every promise, date and figure real?
Second, check the fit. Does it actually answer this customer's question, with their details?
Third, check the feeling. Does the tone suit the person and the moment?
Only then does it go out. And if you can't confirm a fact, it doesn't go out at all.
Try it on your own work
This week, jot down ten tasks you do regularly. Sort each one into a box: AI can help, AI with care, or must stay human. Don't use any AI tool yet; just think it through. You'll be surprised how much clearer your job looks.
Key takeaways
- Businesses are investing heavily in AI, but customers still want a human option: 87% of customers in Gartner's survey (February–March 2026) said it's essential.
- Sort tasks into three boxes: AI can help, AI with care, and must stay human.
- Judgement, empathy, exceptions and vulnerable customers stay with you.
- AI can make up facts and sound completely sure of itself while doing it.
- The business is responsible for what its AI tells customers, as the Air Canada ruling showed.
- Check the facts, then the fit, then the feeling, before anything reaches a customer.