On 27 February 2024, the Swedish fintech Klarna announced a number that startled the customer service world. The AI assistant it had built with OpenAI had, after one month of global operation, handled 2.3 million conversations, two-thirds of the company's customer service chats. Klarna said the volume was equivalent to the work of 700 full-time agents2.
The other numbers looked good too. Resolution time fell from 11 minutes to under 2. Repeat inquiries dropped by 25 percent. Customer satisfaction was on par with human agents. The assistant ran 24/7 in 23 markets, in more than 35 languages, and was estimated to add 40 million dollars in profit in 20242.
For more than a year, Klarna was the most-cited example in presentations about AI replacing labour.
Fourteen months later
On 8 May 2025, chief executive Sebastian Siemiatkowski told Bloomberg that the push to cut customer service costs with AI had gone too far. As cost had become too predominant a factor in how it was organised, he said, what the company ended up with was lower quality3. Klarna began recruiting customer service staff again, promising that customers would always have the option of speaking to a real person3,4.
AI did not disappear from Klarna. It still handles most simple requests. What changed was that the company stopped treating replacing people as the goal.
Deloitte's 1.6 times
Klarna's story lands squarely on something Deloitte's 2026 Global Human Capital Trends report puts in numbers. Surveying more than 9,000 business and HR leaders in 89 countries, Deloitte found 59 percent of organisations taking a technology-centred approach to AI. That group was 1.6 times more likely to miss its expected returns than organisations taking a human-centred approach1.
The report names three tipping points: integrating humans and machines rather than placing them side by side, shifting focus from cutting costs to creating value, and replacing static workforce planning with real-time orchestration of capability. Seven in ten leaders named speed and agility as their main competitive advantage over the next three years1.
Put Klarna in that frame and the story makes sense. In 2024 success was measured in people replaced and minutes saved. In 2025 the company realised some conversations need a human, and its old measures could not see them.
The part that belongs to HR and learning
Most AI budget today goes on tools and licences. What decides the outcome is redesigning the work: which tasks go to the machine, which stay with people, what skills people need to work alongside it, and who is accountable when it gets things wrong. That is precisely HR and learning's territory, and it is usually the part nobody owns in an AI project.
For HR teams in Vietnam, the 1.6 times figure is the argument for a seat in AI projects from day one, rather than being called in once the project has decided how many people to cut.
What the averages hid
Look back at the February 2024 announcement and every figure Klarna published was a total or an average: total conversations, average resolution time, satisfaction on par with human agents2. They were accurate, and they said nothing about the hardest conversations.
In customer service, most requests are simple: checking a balance, moving a payment date, looking up a transaction. A small share involve customers who are angry, in financial difficulty, or facing a problem that fits no script. That small share decides whether customers stay, and it affects reputation more than thousands of smooth exchanges.
A system that handles the majority brilliantly and the minority badly can still produce beautiful averages. To see the problem you have to measure complex cases separately. That is also what Deloitte means by shifting from cost-cutting to value creation: the question stops being how much work the machine can do and becomes how well people and machines together serve the customer.
