In 2025, it started hiring agents again.
Before you hand a decision in your supply chain to a machine, ask three questions.
By May 2025, chief executive Sebastian Siemiatkowski was telling Bloomberg that cost had been "a too predominant evaluation factor". The result, he said, was "lower quality".
Klarna began recruiting people so a customer could always reach a human. The savings were real. So was the bill that arrived later.
A June 2026 MIT CISR briefing (Sebastian, Weill, Haskamp and vom Brocke) sorts decisions by ambiguity and risk. Three questions do the sorting:
- Would two competent people, given the same data, reach the same answer? If yes, ambiguity is low.
- What does one wrong answer cost, and who carries it? A wrong refund costs a few dollars. A wrong recall decision costs far more.
- Who learns when it goes wrong? If nobody owns the exceptions, the machine repeats the mistake at scale.
Low ambiguity and low risk: automate and monitor.
High risk: a named human handles the exceptions and owns the learning, however well the pilot performed.
Automate the decisions a stranger could check. Keep a name on the ones a customer would remember.
Which decision in your supply chain would you never hand to a machine?
