The Fix

Chapter 13: The Rate Has Changed

The rational response to previous failed modernisations was to leave legacy alone. That calculation is no longer correct: AI has materially reduced the cost of understanding legacy systems, and bounded reconstruction through process-owning units provides a structurally different approach to replacement.

She was thirty-one when the rewrite started and thirty-four when it was quietly abandoned.

The platform was called Horizon. The brief was unambiguous: replace the claims management system, a fifteen-year-old monolith written in a language that three people in the company still understood, with a modern, event-driven architecture that would “future-proof the business.”

Katrin joined as the second senior engineer. She knew the legacy system better than anyone except the contractor who had left the previous autumn. The monolith was genuinely painful. Deployments required a four-hour maintenance window. A schema change needed sign-off from two teams that no longer existed in the organisation chart, routed through a governance function that reviewed changes it could not evaluate. Three incidents in a single quarter had been caused by the same race condition in the payment reconciliation path, each time patched differently because no one could hold the full system in their head long enough to fix it properly.

The first year was productive. The team built the core domain model, the event infrastructure, and two of the seven process flows. Katrin worked with a payments specialist named Willem who understood the regulatory constraints better than anyone in compliance, and between them they built a reconciliation service that was cleaner, faster, and more auditable than anything the monolith had produced.

Then the programme followed the pattern that every practitioner recognises. Engineers rotated off; their replacements had technical context but not institutional context: the reasons why certain boundary decisions had been made, which edge cases were real and which were inherited workarounds. Scope expanded as stakeholders discovered late that the new system did not accommodate their process variants. The sponsor moved to a different division; the replacement inherited the budget but not the conviction. A consulting firm recommended running both systems in parallel indefinitely. In practice: two claims systems, two data models, two on-call rotations, and a coordination layer that required more engineering effort than either system alone.

Eighteen months later, the parallel approach was abandoned; the legacy system remained in production and Horizon's reconciliation service was quietly decommissioned. Katrin stayed another year, long enough to watch the organisation repeat the same modernisation conversations at the next strategy offsite. Willem left for a smaller company, and the institutional knowledge walked out with them.

The programme budget was € 4.8 million. The opportunity cost of four senior engineers who spent three years building something that was discarded, the knowledge that walked out the door, none of it appeared in the post-mortem. The post-mortem attributed the failure to “insufficient stakeholder alignment in the early phases” and recommended a more robust governance framework for future programmes.

Katrin read the post-mortem. She recognised the language. It was the same language that had been used to justify the rewrite in the first place. The structural corrections described in the following chapters require sustained executive mandate. Process ownership, contract enforcement, and budget envelopes cannot be implemented unilaterally by engineering. The transition moves authority from the coordination layer to the units that own processes, and that redistribution must be authorised by the people who currently hold the authority. A direct address to the CEO at the end of this book states this explicitly.

...

See also: Full contents · Preview chapters · Illusions in the Boardroom