Published on
The Bottleneck Is Not Modelling. It Is Validation
My last three customer conversations point the same way: the bottleneck is not modelling. What costs the most time is validation.
That did not surprise me. Our experience across almost every project of recent years shows that building a driver model is manageable. It gets genuinely laborious where assumptions have to be checked and scenarios made plausible in aggregate.
That is also where the AI question gets interesting. Many people want support. But that final checking step is the one they are reluctant to hand over.
The reason, as I see it, is simple: accountability hangs on validation. Anyone defending a simulation result in front of the management team needs the story behind the number. Traceable assumptions and explainable logic.
This also fits what you see in the market. AI in finance today often helps with knowledge work, automation and anomaly detection. Judgment is a different level.
Calculating can be delegated. Defending the result cannot.