He wasn't the only one. I built a couple of systems there integrated into the accounts system and "no ML" was an explicit upfront design decision. It was never regretted and although I'm sure they put ML in it these days, last I heard as of a few years ago was that at the core were still pages and pages of hand written logic.
I got nothing against ML in principle, but if the model doesn't do the right thing then you can just end up stuck. Also, it often burns a lot of resources to learn something that was obvious to human domain experts anyway. Plus the understandability issues.
He wasn't the only one. I built a couple of systems there integrated into the accounts system and "no ML" was an explicit upfront design decision. It was never regretted and although I'm sure they put ML in it these days, last I heard as of a few years ago was that at the core were still pages and pages of hand written logic.
I got nothing against ML in principle, but if the model doesn't do the right thing then you can just end up stuck. Also, it often burns a lot of resources to learn something that was obvious to human domain experts anyway. Plus the understandability issues.