The right AI pilot in insurance delivers its first measurable results within 2 months.
No buzzwords. No illusions. And not any sooner.
Just hands-on experience from running an insurtech startup through which CZK 1.5 billion worth of insurance was arranged last year.
Pick the wrong one, on the other hand, and you get a presentation costing a few hundred thousand crowns that survives two quarters in a drawer, and then it’s goodbye.
The difference isn’t in the model or the budget. It’s in which process you choose.
I have three criteria for implementing AI at Verzuz.
The process has to be repetitive. Not an exception that happens once a year. Checking documents for completeness, first triage of claim reports, validating input data when a policy is taken out. These are the right candidates.
There has to be a clear metric. How many minutes it takes today, how many after deployment. How many cases go through without manual intervention. The number has to exist before the start, not after it.
It has to be clear where the human stays in the loop. In a regulated environment, you can’t let AI loose on a case without a conscious decision about where the limits of its autonomy lie.
It’s not rocket science. It’s a discipline like any other, one that calls for strategic thinking. Especially now, when the field is swarming with self-proclaimed AI saviors.
Every AI decision has to be traceable after the fact. Because ownership still rests with the specialist. For some, that’s an obstacle; for me, it’s a safety guardrail. And in the world of insurance, that is still the most valuable commodity there is.
