Start with what cannot stop
Before any target architecture is drawn, the useful question is which services, lines, or facilities cannot be interrupted, and for how long. That answer determines the change windows available, and change windows determine the shape of the programme far more than platform preference does.
In practice this means mapping dependencies at a level of detail most inventories do not hold: which application depends on which storage path, which control network depends on which power feed, which maintenance activity assumes a particular system is available.
Modernization is a sequencing problem
Once constraints are understood, modernization becomes a question of order. Which changes reduce risk immediately, which unlock later work, and which should wait until a dependency has been retired.
Sequencing badly is what produces the familiar pattern of a stalled programme: a platform delivered but not adopted, because the systems that were meant to move onto it could not be interrupted at the time the platform became available.
- Do first: changes that remove single points of failure without requiring application change.
- Do next: platform work that can be run in parallel with the existing estate.
- Do carefully: migrations that require an operational outage, batched into as few windows as possible.
- Do last: decommissioning, once evidence shows nothing still depends on the old path.
Ownership decides whether it survives
A modernized platform that no team feels accountable for degrades quickly. Every element introduced should have a named owner, a monitoring path, and a defined maintenance obligation before it carries production.
This is why we treat handover as part of delivery rather than an event at the end. Documentation, runbooks, and access arrangements are produced while the work happens, when the detail is still fresh.
Measure improvement in operational terms
The credible measures of modernization are operational: fewer unplanned interruptions, shorter maintenance windows, faster recovery, less manual intervention. These are specific to an estate and should be baselined before work begins, not asserted afterwards.
Where a business case needs numbers, take them from your own environment. Generic industry percentages do not survive contact with a specific plant or facility.
The estate you have is the constraint and the starting point. Any plan that ignores it is a plan for a different company.