Workload Fit Assessment
Choosing between warehouse, lake and lakehouse by the workloads that must run, rather than by which one is currently fashionable.
The three options are not points on a maturity curve; they are different optimisations, and the choice follows from what has to run.
A warehouse optimises for SQL over structured data with strong governance, mature access control and predictable performance. If the workload is business intelligence and finance reporting, it is usually the right answer and the argument for anything else is thin.
A lake optimises for cheap retention of anything, including formats nobody has modelled yet, and for engines beyond SQL. Machine learning training sets and semi-structured event data belong here.
A lakehouse attempts both over one copy, and its case is strongest when the alternative is maintaining two copies with a synchronisation job between them — which is where correctness and cost problems come from.
The assessment worth writing down lists the actual workloads, their latency and concurrency requirements, who runs them and what governance applies. The common failure is choosing the architecture first, then discovering the finance team's concurrency requirement cannot be met and standing up a warehouse alongside — at which point the organisation has all three.