klarna replaced about 700 support staff with a bot. then it started hiring people back.

support volume is mostly routine, the routine part automates cleanly, and on a spreadsheet the headcount math looks finished. what the spreadsheet does not carry is that customers judge you almost entirely on the remainder: the billing error, the thing that broke twice, the refund just outside policy.

klarna's satisfaction fell on exactly those cases. the cost arrives two quarters later as churn that gets blamed on price, followed by a rehire at a new title and a higher rate. the fix being sold for this is an escalation button, which routes the angriest customer into a queue and tells them their call is important to us.

automate the routine and staff the remainder on purpose. route by complexity, not by volume, and keep a named person on anything involving money, an error you caused, or a customer asking for the second time. the org chart will call the rehire a new role. the customer will call it the third time they explained this.

do this: route by complexity. a named person owns money, errors and second asks. not this: cut the team and call the escalation queue a fallback.