A senior data architect we placed earlier this year told us something that stuck: the technical interviews she sat through barely touched on the insurance domain she had spent a decade learning. Every question centred on model architecture, Python frameworks, MLOps pipelines. Nobody asked how she would handle the actuarial edge cases that make insurance data so difficult to work with.
New research highlighted by Diginomica puts a name to this pattern. The paper argues that hiring teams are increasingly filtering for AI proficiency while treating domain expertise as a nice-to-have. The result is a generation of technically capable hires who lack the contextual knowledge to apply those capabilities where they matter most.
In our experience, this shows up most clearly in regulated industries across the DACH region. A pharmaceutical company hiring a machine learning engineer needs someone who understands GxP compliance and TensorFlow. A bank building fraud detection models needs people who know how payment flows actually work alongside the technical stack. Strip out that knowledge and the AI becomes a sophisticated system that misses the obvious.
The uncomfortable truth is that domain expertise takes years to acquire. AI skills can be taught in months. When hiring managers optimise for what can be tested in a two-hour interview, they end up with teams that can build technically impressive systems with no grounding in the problems those systems are supposed to solve.
We have started asking clients a direct question: who on your data team has been in this industry for more than five years? The hesitation that follows often tells us more than the answer.
Prompted by reporting from Diginomica.