A ministry-funded system that broke a schema incompatibility standing for ten years. The reconciliation an expert used to perform by hand, over a working week, is now a rule applied once and an analysis anyone can run before lunch.
Role Engineer, Ministry Partner in Innovative Technology Transformation. Period June 2025 to March 2026. Funding USD 3,200.
An analysis that used to take five expert-days now takes twenty minutes. The thing standing in the way had been standing in the way for ten years, and it was not a hard technical problem so much as a problem nobody had been paid to sit with.
Turnaround falling from a working week to the length of a coffee break is the change that matters, because it decides who gets to ask. An analysis you can run before lunch is one people will actually run. One that costs a week of a specialist’s calendar only gets requested when somebody is already fairly sure what the answer will be.
The whole project was funded at 3,200 US dollars, which is the number I would put next to the other two if I were assessing this from the outside.
Two systems held data that needed to be read together, and their schemas could not be reconciled. That incompatibility had stood for a decade. In the meantime the work still had to be done, so it was done by hand: an expert pulled from both sources, reconciled them by judgment, and produced the analysis.
This is the shape of a very common category of problem, and it is worth naming precisely because it does not look like an engineering problem from the inside. Nobody was blocked. Work was shipping. There was a functioning process with a known duration, and that duration was absorbed into somebody’s job description rather than showing up anywhere it would be questioned, which is exactly what lets it survive for ten years.
Five days of a domain expert is not five days of typing. It is the time needed to hold two incompatible representations in one head simultaneously and decide, case by case, which records correspond to which.
That is genuinely skilled work and the expert was doing it correctly. The problem is that the skill was being spent on reconciliation rather than on analysis, and reconciliation is the part a machine can do once somebody works out the mapping properly.
The mapping was worked out and made executable, which turned a judgment call repeated thousands of times into a rule applied once. After that the analysis itself is quick, because the analysis was never the slow part.
I want to be careful about how much credit that deserves. This was not a novel technique. It was a piece of work sitting in the gap between two organisations, where the burden of not doing it was invisible to both, and the entire contribution is that somebody was funded to sit with the schemas for long enough to reconcile them.
The engineering lesson I took from it is that the highest-leverage problems are frequently not the hardest ones. They are the ones whose weight is hidden inside somebody’s job description.
The interesting effect is not that the same work now happens faster. It is that a different set of questions becomes askable.
At five expert-days, an analysis is a decision in itself. Somebody has to justify requesting one, schedule the specialist, and wait. It happens when the answer already matters enough to defend the effort, which means it rarely happens speculatively, and speculative analysis is where a lot of the useful findings live.
At twenty minutes, it stops being a decision and becomes a question you can just ask. You can run it on a hunch, run it again with the parameters changed, and throw away the ones that come back uninteresting. That is a different institution, and it is why the turnaround number is the one I lead with rather than any measure of effort saved.
Delivered within the ministry partnership programme, June 2025 to March 2026.
I have not published the underlying data, the sector, the counterpart systems, or the cost figures here. The work was done inside a government partnership and that detail is not mine to release unilaterally.