A payments company announced that its AI assistant was doing the work of seven hundred people. A year later its chief executive was hiring humans again.
Ten items. Say your answer out loud before you reveal it.
Read the sentence, guess the meaning of the word in bold, then reveal.
Three terms: backtrack · hand over · go all in. Say the right one, then reveal.
One missing word fits every phrase in the group. Say it, then reveal.
Read each part. After it, answer the quick check from memory, then discuss one question.
Klarna is a Swedish company that lets customers buy now and pay later. It has millions of customers, and they ask a great many questions: where is my order, why was I charged, how do I change a date. In February 2024 the company published a figure that travelled around the business world. Its new AI assistant, it said, had handled 2.3 million conversations in its first month. That was about three out of every four customer chats, and Klarna said the work was equal to that of seven hundred full-time agents.
The assistant worked in more than thirty languages, around the clock. Klarna said customer satisfaction was similar to what people had given human agents and that the time to solve a problem had fallen sharply. The chief executive, Sebastian Siemiatkowski, spoke about the result often and with obvious pride.
It is worth noticing one word in the announcement: "equal to". The seven hundred was a company's own estimate of workload, not a list of people who had lost their jobs. Klarna's numbers of staff had been falling anyway, by about a fifth, because the company had stopped hiring and people had left.
1. What did Klarna say about its AI assistant in February 2024?
2. What does "equal to seven hundred agents" actually mean?
3. Why had the number of Klarna employees fallen?
1. That in its first month it handled about 2.3 million chats, around three-quarters of the total, with the work of seven hundred full-time agents.
2. It was the company's own estimate of the amount of work, not a count of people dismissed.
3. The company had stopped hiring and people had left without being replaced, which is called attrition.
In 2025 the story changed. In an interview with Bloomberg, Siemiatkowski said that Klarna was hiring human customer-service workers again. He admitted that the company had put too much weight on cost, and that this had produced work of lower quality. Cheaper, he said, was not the same as good enough.
The new plan was not a return to the old system. Klarna intended to build a pool of remote workers, a little like the model used by Uber, that customers could reach when they wanted a person. The company recruited students and people who already loved the product, and it said it wanted to be sure that a human was always available on request.
Klarna itself rejected the idea of a retreat. It described the change as two tracks running together: automation for scale, people for quality.
1. What did the chief executive admit in 2025?
2. What kind of workforce did Klarna plan to build?
3. How did Klarna describe the change?
1. That the company had focused too much on cost, which had made the quality lower.
2. A pool of remote workers, a bit like Uber's model, who customers could reach on request.
3. As two tracks together: AI for scale and humans for quality, not a retreat from automation.
Reports of the reversal vary. Some outlets wrote that the company had laid off seven hundred people and then regretted it, which Klarna's own account does not support, and some described customer satisfaction as having collapsed, for which the evidence is thin. The safest summary is the plainest one: a company that moved very fast to automate found that some conversations needed a person, and it adjusted.
What should be taken from that? Nobody would argue that a chatbot should not tell a customer where a parcel is. Almost everybody would want a human when the problem is unusual or the stakes are high, such as a fraud case or a dispute about a large bill. The skill lies in deciding which is which before the customer has to complain.
That is a decision a small consultancy can make for its clients in an afternoon, with a pen and one simple grid, and it is much cheaper than finding out the hard way.
1. Why should you be careful with some reports about this story?
2. When do customers usually want a person?
3. How can a small consultancy help a client with this decision?
1. Some outlets say 700 people were laid off or that satisfaction collapsed, claims the company's own account and the evidence do not clearly support.
2. When the problem is unusual, or when a lot is at stake, such as fraud or a dispute about a large bill.
3. By using a simple grid to decide which tasks to automate and which to keep with people, before problems appear.
Before you automate a task, put it on a grid with two questions. How routine is it, meaning does it come up all the time in the same shape? And what is the cost if it goes wrong? The answers tell you who should do it, a machine, a person, or the two of them together. The Handover Rule is simple: automate the routine and cheap, keep a person for the unusual and the costly, and make the human door easy to find.
Each task could be handled by a customer-service tool. Where does it belong on the grid? Choose, then check.
1. Automate — routine, and nothing much is lost if it is slightly wrong.
2. Person only — unusual and costly. Trust is at stake and the answer cannot be generic.
3. Automate — same shape every time, low risk.
4. Automate and check — routine in form, but the stakes are higher, so a person reviews it.
5. AI helps, person decides — the tool can summarise and sort the issues; a person chooses the response.
6. Person only — unusual and expensive if it is wrong.
7. Automate — the classic routine question.
8. AI helps, person decides — unusual but low stakes, so a tool can suggest while a person chooses.
The grid is a starting point. Your own data will move some tasks. Check what customers actually ask before you decide which are routine.
The handover is part of the product. A bot that makes the human hard to find saves money for a month and costs trust for years.
It works for internal tasks too. The same grid helps with first drafts, scheduling and research: routine and low stakes go to the tool; unusual and costly stay with you.
Stories about companies are mostly stories about change. A small set of structures lets you say what was true then, what is true now, and what stayed the same.
Complete each sentence with the word or phrase from the box. Use each once: used to · no longer · still · since.
1. Klarna handle every customer chat with human agents.
2. It depends on the bot for every conversation.
3. The bot answers the routine questions every day.
4. 2025 the company has described its approach as two tracks.
1. used to — a past habit that has ended
2. no longer — it was true before, it is not true now
3. still — it was true before and continues to be true
4. Since — from a point in the past until now, with the present perfect
Write one sentence for each pair, using the structure in brackets.
1. 2023: hired agents. 2025: hires agents again. (no longer / again)
2. Before: cost mattered most. Now: quality matters most. (used to)
3. February 2024: 75% of chats by AI. 2025: the figure is lower. (fell)
4. 2024 and now: AI answers parcel questions. (still)
1. The company was not hiring agents in 2023, but it is hiring them again now. (other answers are possible)
2. Cost used to matter most; now quality does.
3. The share of chats handled by AI fell after the first announcement. (or: fell from about 75%)
4. AI still answers parcel questions.
Now describe a change in your own work or business over the last two years, in four sentences, using used to, no longer, still and since.
Three tasks about deciding what to automate. Read the situation, take a moment, and speak. Aim for about a minute each.
Tell the Klarna story in about a minute, from the first announcement to the change of direction. Use one sentence with used to, one with no longer and one with since.
A client with a small online business wants to hand all customer enquiries to an AI assistant to cut costs. Take them through the Handover Rule with three examples of questions their customers might ask, and say where you would keep a person.
You recommended an automatic tool to a client three months ago and it has not worked as well as you hoped. Phone them. Say what you recommended, what has happened, what you now think, and what you propose. The person you are speaking with plays the client, and may be annoyed.