Retiring eighty-plus systems from an estate nobody could afford to stop using

Context

Every application in the estate was load-bearing for somebody. The systems were old, the dependencies were tangled, the regulators were real, and replacing the platform was not on the table. That is the ordinary condition of enterprise technology and it is the opposite of a greenfield: you cannot switch the thing off to work on it, and the cost of being wrong is not a rollback, it is an operational outage in a business that has no manual fallback left.

Decision

Treat modernization as a portfolio decision rather than a queue of individual technical retirements. What gets switched off, and in what order, is a question about business impact, dependency risk and ownership — not a question about which system is easiest to decommission.

Approach

• Put ownership, dependency, risk and operating cost into a single decision frame, because the argument was being conducted in four different currencies at once — finance counting spend, technology counting incidents, the business counting disruption, risk counting exposure — and no two of them could compare answers.
• Established a named owner for every application before sequencing anything. An application nobody will claim is not a cheap retirement; it is an unknown risk wearing a low-cost label, and it is the one most likely to take something else down with it.
• Sequenced by dependency and business impact rather than by ease. The cheapest system to switch off is rarely the one whose removal reduces the most risk, and a programme that harvests the easy wins first runs out of goodwill exactly when it reaches the hard ones.
• Ran delivery across analysts in the United States, India and the Netherlands, which forced the decision record to carry its own reasoning. A handoff that crosses that many time zones cannot rely on the conversation that produced the decision, because that conversation is asleep.
• Wrote explicit handoff, validation and decommission criteria per retirement, so a system stayed recoverable until the final gate rather than becoming irreversible the moment work started on it.
• Treated decommission as an evidenced step rather than the absence of further activity. A server nobody has touched in six months is not retired, it is unmonitored — and the difference only becomes visible during an audit or an incident.
• Carried the same frame into two transformation programmes: a $10M manufacturing consolidation moving 1,500 finished goods across six production lines, and a $20M international agricultural build requiring restricted-data handling and synchronization across multiple ERP systems.

Outcomes

• Owned a 30-application portfolio carrying more than $4M in annual operating expense.
• Supported the decommissioning of eighty-plus servers, virtual machines and databases, each with a documented transition path rather than a closed change ticket.
• Ownership made explicit across the estate — the precondition for every later decision, and the thing most often missing when one of these programmes stalls.
• A single decision record shared by product, finance, risk and operations: one artifact all four could argue against, rather than four incompatible views of the same estate.
• The same discipline carried into roughly $30M of transformation programmes across manufacturing and agriculture, on three continents.

Control boundary

All figures are rounded, public-safe summaries. This case intentionally omits employer names, system names, timelines, and proprietary architecture. It describes decisions and method rather than systems, which is the only form in which this work can honestly be published.

What this taught me

None of this was AI work — it predates the current wave entirely, and it is where the discipline behind everything else on this site came from. It belongs beside the agent and governance cases because AI adoption fails on exactly the same question, one earlier than anybody wants to start. Not which model, but: who owns this, what depends on it, and how would we undo it if we were wrong. An organization that cannot answer those about the estate it already runs will not answer them about a model it just bought. Modernization gains momentum when product, finance, risk and operations argue against one shared decision record — not when technology work is measured by tickets closed.