In the first article in this series, we described a fundamental shift: how AI in increasingly many organizations is treated as labor, not just as technology. But it is only when this logic encounters volume that the difference between intention and reality becomes clear.
Volume quickly reveals whether AI is an experiment – or an integrated part of daily operations.
Theory meets reality
Klarna is not a company that can experiment quietly. With millions of transactions and thousands of daily inquiries, the company has an extremely low tolerance for friction. Any weakness in the systems becomes immediately visible – to customers, and thus to the bottom line.
In this context, the company's CEO and co-founder, Sebastian Siemiatkowski, has been clear. He does not view AI as support behind the scenes, but as an active and operational part of the front line. AI has practically become the very front door to customer dialogue.
Traditionally, increased volume has meant increased staffing. More customers have demanded more employees, more coordination, and higher complexity. This linear relationship has long been treated as a natural law.
At Klarna, this law is in the process of being repealed.
When AI becomes the primary delivery
At Klarna, AI has transitioned from being a tool to becoming a central part of the delivery itself. That doesn't mean humans have been removed from the equation, but rather that roles are clearer and more consciously distributed.
AI does not just handle simple questions. It manages large parts of the dialogue related to payments, delays, and transactions — in real-time and on a large scale. It is only when the dialogue requires assessment, context, or clear responsibility that a human is involved.
The division of labor is thus clarified:
AI handles the transactions:
- repeated inquiries
- standardized processes
- immediate response at scale
People own the relationships:
- exceptions and escalations
- empathy and context understanding
- ultimate responsibility
The result is not only higher efficiency but an organization with a completely different flow.
From cows to capacity
When AI takes over the volume, customer dialogue ceases to be a staffing problem. Instead, it becomes a matter of architecture and organizational sustainability.
This forces management to address issues that cannot be delegated to IT:
- Who owns the dialogue when the initial contact is not human?
- How is quality defined when response times approach real-time?
This is not just about efficiency, but about building a engine that can withstand load over time – without wearing down the organization.
Economies of scale
Volume has always had a cost. In the traditional model, operating costs increase proportionally with the number of inquiries. More customers have meant larger teams, more training, and increasing complexity.
When AI takes volume, this relationship breaks down. The marginal cost per inquiry drops dramatically. Capacity can be scaled up and down in seconds, without the organization having to go through painful hiring or downsizing processes.
This provides management with a new strategic freedom of action:
- greater predictability in delivery
- less operational stress
- consistent quality, even under high load
This is not efficiency.
It's tempting to call this efficiency, but it's an underrivelse.
Efficiency is about doing the same thing a little faster or cheaper. What we see here is a more fundamental reorganization of work. Transaction is separated from relationship. AI handles the transactions so that humans can own the relationships.
It is in this division of labor that real delivery capacity is created.
A necessary leadership election
Letting AI take the volume is not a technical upgrade. It is a leadership choice.
When Siemiatkowski describes this development, he is clear that this is not just about cost-cutting per se, but about building an organization that can handle scale without losing control. It is a choice that directly affects how work is organized, how quality is measured, and who bears responsibility.
Klarna does not show that this is easy. They show that it is possible.
What needs to be built for this to work?
This type of operation does not happen on its own. It requires the right architecture, clear ownership, and systems built for continuous production – not for demos and pilot projects.
This is where many organizations come to a halt. They have the technology but lack the structure that allows it to work over time.
In the latest article in this series, we take a closer look at what actually needs to be built for AI to function as real labor – with control, responsibility, and full delivery capacity.
