A federal program supporting service members returning to civilian life needed to understand where those service members were going by state, and ideally far finer. The source was DMDC data: hundreds of thousands of records, updated annually. The team’s only way to use it was by hand. Analysts compiled the figures manually and assembled a single PDF for each state. One report, one state, built start to finish by a person. The work was heavy enough that it could only be done once a year. By the time a report was finished, the question behind it had often moved on, and anything more specific than a state-level total simply didn’t exist. There was no way to ask “what does this look like for this region” without commissioning another round of manual work.
We built an engine that took the raw DMDC data and did the compilation automatically. Instead of a person assembling one PDF per state, the system generated reports on demand, instantly, and let the user filter to whatever subset they actually cared about. Most importantly, we unlocked a level of detail the old process couldn’t reach: per–zip-code granularity, nationwide. The same dataset that used to surface a single number per state could now answer questions down to an individual zip. Because the data was sensitive, the build respected that throughout and the visible output gave the program what it needed without exposing what it shouldn’t.
A reporting cycle that ran once a year now ran in seconds, as often as anyone needed it. Questions that previously required a fresh round of manual compilation: “show me just this region,” and “filter to these attributes” became self-service. And granularity that had never existed in the old workflow, down to the zip-code level across roughly 34,000 zip codes, was suddenly on the table for every user. The program stopped waiting on its own data.
