How Semler IT's CIO got control of token spending: his method

Semler Gruppen, the Danish group behind Semler IT, controls its AI token spending with usage limits and alarms. The method is described by the company's CIO, Morten Rye Christensen.

His reasoning is that AI has changed how fast costs can get out of hand. In his words, the scale is "markedly more violent" than before: when something goes wrong, it goes badly wrong. Before AI, a problem of this kind might build up over a weekend. Now, he says, something can accelerate within half an hour.

That speed is why limits and alarms are the core of the approach. The limits cap how much can be consumed, and the alarms flag unusual consumption early enough for someone to react before the bill escalates.

The material available here is only a short excerpt, so it gives no budget figures, no thresholds, and no detail on how the limits and alarms are configured or which AI tools or teams they cover. What it does establish is the principle: with token-based AI services, cost monitoring has to work in minutes rather than days, and Semler IT's CIO treats automated guardrails as the way to achieve that.