On-prem is a real pain in terms of delivery. For a lot of customers there’s long lead times and change management processes to navigate to upgrade to a new version. There’s a wide distribution of environments and network topologies that introduce often surprising edge cases. There’s a whole lot you don’t control or don’t have visibility into. Which is almost by design and perfectly fine - it’s why the customer chose on-prem!
But… the trade off is you really want high conviction your new feature is perfect before it goes on-prem. You can just flip a feature flag to temporarily roll back some capability if you uncover an issue. It might take your customer months to negotiate with their internal stakeholders a deployment window for an upgrade.
Now obviously that’s not true of all on-prem customers. But it was true for such a large percentage of them that it was more pragmatic to accept it as the general case rather than have a higher velocity flow for the handful that would adopt new capabilities at a pace closer to Cloud customers.