Three teams have independently chosen three different message brokers. Design the selection and governance process going forward.

Understand why it happened before correcting it

Three independent choices usually mean there was no easy default, no forum for the decision, or a central process slow enough that teams routed around it. Treating this as non-compliance produces resentment and another workaround.

Ask each team what it needed and why the existing options did not fit. Frequently one or two of the three choices were reasonable and one was arbitrary.

Count the real cost of the divergence

Three brokers means three sets of operational expertise, three on-call runbooks, three upgrade paths, three security reviews, three monitoring integrations, and engineers who cannot move between teams without relearning. That is the argument for convergence, and it should be quantified rather than asserted.

Weigh it against the cost of migration and against the genuine differences between the tools — some workloads really do need a log with replay, and others really do need a broker with per-message acknowledgement and delayed delivery.

Evaluate on the dimensions that matter over a decade

Not features and benchmarks — public benchmarks are constructed to favour their publisher.

Fit for the actual workloads, tested with a spike against your own traffic. Operational burden in engineer-days per year. Team capability now and the cost of acquiring it — a technically superior tool the team cannot operate is the worse choice. Ecosystem and longevity, including the hiring market. Exit cost. Total cost of ownership. Security posture and vulnerability response.

Score them openly in a matrix — not to produce an answer, but to surface disagreement about the weights, which is where the real dispute usually is.

Choose a paved road, not a mandate

Centralise the default, not the decision. Pick one broker as the supported path — managed, with a platform-owned client library, standard observability, documented patterns, and genuinely easier to adopt than anything else.

Then an exception process: a team with a requirement the paved road cannot meet documents it, and accepts the operational ownership that comes with leaving the road. Exceptions are approved, recorded and reviewed.

A rising exception rate means the road is wrong, not that teams are non-compliant. That is the signal the process exists to produce.

Migrate pragmatically

Do not mandate a big-bang convergence. New workloads use the paved road; existing ones migrate when there is another reason to touch them, or when the operational cost is demonstrably higher than the migration.

One of the three may reasonably remain as a documented exception forever.

Record it and keep it current

An ADR with the options, the criteria, the weights and the rejected alternatives — which is what stops this recurring in eighteen months.

And a technology radar with adopt / trial / assess / hold, reviewed on a cadence, so the standard is a living thing rather than a decision nobody may revisit.