That situation makes sense but then i have to ask why hasn't it been done already? Software developers are not rare and if the use case is so isolated and discreet then surely it would have been tried by now. Even without genAI, CRUD, RDBMS record management, SSO, row level security... none of those things are new or out of reach until now. I think what you'll find is when you sit down with the users and start asking about the parts of the exiting system they actually need you'll never get agreement nor a clear answer. When/if you finally get a set of requirements and after UAT sign-off and then after go-live the users will say "this isn't what i meant" and you're back to square one. Rinse/repeat for years and then one day an exec will say "why are we wasting all this time, let's just subscribe to an OTS saas and make them configure it to meet our needs".

Nobody gets fired for choosing SAP, Salesforce etc.

Spending tons of money to get a janky, unreliable system of record, or finding out too late it is missing crucial auditing capabilities, or that it has Big Money bugs, on the other hand, is far worse, especially if you have investors asking what the hell you were thinking.

Your point about users not knowing what they wanted until after the fact is also painfully true. The hardest part about these systems is the people most likely to buy are the ones who have been doing it with a lot of human processes for years. Buying a SaaS or other third party product means having leverage to force them to change to more standard practices. Building in-house means that everyone will fight to high hell to make sure that their special snowflake way of doing things is accounted for and you end up in a worse spot as a result.

Complacency, bad management, revenue growing faster than costs for a decade hiding the problem, politics.

There's multiple people highly involved into maintaining the status quo which do everything to take any responsibility out of them.