Enterprise service bus, when point-to-point has stopped scaling

Below about five or six integrated systems, almost certainly not — a bus adds infrastructure, a skill requirement and a new single point of failure. It starts to pay for itself when the number of point-to-point connections has become the maintenance problem in its own right.

The label has fallen out of fashion; the problem has not. Modern equivalents are event brokers and integration platforms, and the underlying question is the same: do you want a shared, monitored layer or a growing set of direct connections?

Longer if attempted as one programme, which is why we would rather migrate a slice first. Moving remaining integrations as they need changing anyway avoids rewriting connections that work perfectly well.

An agreed definition of what a customer or an order means across the estate, so each system maps to one shared shape rather than to every other system. It is the hardest part because it is a business agreement rather than a technical one, and it is where most of the long-term benefit sits.

It becomes a single point of failure, which is the honest cost of centralising. That is why resilience, queuing and failover matter more here than in a point-to-point arrangement, and why a bus is not free once you account for running it.

Usually. Adapters can front systems with no modern interface, using file drops, database views or vendor connectors, which also means the old system can be replaced later without touching everything that talks to it.