Key takeaways
- Klarna’s CEO told Bloomberg in 2025 that customers must always be able to reach a human, after its AI-heavy service produced lower quality.
- Gartner expects agentic AI to resolve 80% of common service issues by 2029, which still leaves the hardest 20% for people.
- In HBR research, 56% of customers had to re-explain their issue. A good handover passes the full context so they never repeat themselves.
- Set clear triggers: emergencies, complaints, explicit requests for a person, low AI confidence and high-value decisions.
The most important feature of an AI agent is not what it can do. It is knowing what it should not do, and getting the customer to the right person, fast, with the story already told. Here is how to design AI-to-human handover that customers actually like.
A lesson from Klarna
Klarna, the Swedish payments company, became one of the most visible examples of AI in customer service when it said its AI assistant was doing the work of hundreds of agents. Then, in May 2025, CEO Sebastian Siemiatkowski told Bloomberg that the company was hiring human customer service staff again.[1] He said AI was cheaper but produced lower quality, and that the company needed to be clear that "there will always be a human if you want."
The takeaway is not that AI failed. It is that AI without an easy path to a human fails. Customers are happy to let AI handle the simple things, as long as they are never trapped.
How much can AI handle on its own?
Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention.[2] That is a large share, and it is a forecast, not today’s reality. Even then, the remaining issues will be the hard ones: complex, emotional, unusual or high-stakes. Those are exactly the cases that decide whether a customer stays.
The hidden cost of a bad handover
In Harvard Business Review’s study of more than 75,000 customers, 59% were transferred and 56% had to re-explain their issue.[3] Customers who experienced low effort were far more loyal: 94% intended to buy again.
A transfer that forces the customer to start again turns a good AI experience into a bad overall one. The handover is part of the product.
When should an AI agent hand over?
| Trigger | Example | What should happen |
|---|---|---|
| Emergency or safety | Chest pain, a gas smell, a car accident | Immediate guidance (e.g. call 998/999) and transfer to on-call staff |
| Customer asks for a person | "I want to speak to someone" | Transfer without arguing |
| Strong frustration | Raised voice, repeated complaints | Apologise, transfer to a senior person |
| Low confidence | The answer is not in the knowledge base | Say so honestly, offer transfer, callback or message |
| High value or risk | A large purchase, a legal or medical decision | Human approval or conversation |
| Repeated failure | The AI misunderstands twice | Stop trying and transfer |
What a good handover looks like
1. Tell the customer what is happening
"I’m going to connect you with Sara from our bookings team. I’ve shared what you told me, so you won’t need to repeat it."
2. Pass the full context
The person receiving the call or chat should see, before they say hello: who the customer is, what they asked, what the AI already did, and why it was handed over. In a unified inbox, that is the conversation thread itself.
3. Choose the right route for the hour
- Office hours: warm transfer to the right team or person.
- After hours, urgent: on-call number or a WhatsApp alert to a manager.
- After hours, not urgent: book a callback slot and confirm it by WhatsApp.
4. Never leave the customer in silence
If no one is available, say so and give a specific next step. "Someone will call you between 9 and 10 tomorrow morning" beats a ringing phone.
Measure the handover, not just the automation
Track these monthly:
- Containment rate: conversations fully resolved by AI.
- Handover rate and reasons: which triggers fire most often.
- Time to human: how long a handed-over customer waits.
- Repeat explanation rate: how often customers say "as I already told…".
- CSAT after handover versus after AI-only conversations.
A rising containment rate is good only if satisfaction after handover stays high.
See how Northline AI routes callers to your team instantly, with context.
The principle
Customers do not mind talking to AI. They mind being stuck. Design every AI conversation with an exit, make the exit fast, and make sure the human on the other side already knows the story.
Frequently asked questions
Sources
Every statistic in this article comes from the studies below.
- 1.Klarna is hiring customer service agents after AI couldn’t cut it on calls Entrepreneur (reporting Bloomberg interview with CEO Sebastian Siemiatkowski), 2025
- 2.Gartner predicts agentic AI will autonomously resolve 80% of common customer service issues without human intervention by 2029 Gartner, 2025
- 3.Stop Trying to Delight Your Customers Harvard Business Review (Dixon, Freeman & Toman), 2010
Researched by
Northline AI Research
Editorial team, reviewed by Abdelfattah Alsaeed
The Northline AI research team reads the studies so UAE business owners do not have to. Every statistic links to its original source.