Federal fee-assistance modernization
The situation
A national program that helps military families pay for child care — the benefit that decides whether a single parent keeps their job, whether a family survives a move across the country, whether a deployment is manageable at all. The system behind it was decades old, and the application process ran on paper logic: families guessed at what to submit, coordinators sorted it out afterward, and a complete first-attempt submission was rare enough that operations staff had a nickname for one.
The organization committed to modernizing it. I came in through an early proving-ground engagement, the work expanded from documentation into design, research, and product strategy, and I moved to the organization directly as the embedded design, research, and QA practice.
What I found
Everyone assumed families were careless or confused. The three-layer diagnostic said otherwise.
The governing requirements, operational practice, and system behavior had each drifted from the other two. A document had a validity window measured in weeks — but families apply months before care starts, so by processing time their perfectly valid paperwork had expired. That's not user error; that's a timing mismatch built into the process. Operations assumed physical signatures were required — the governing rules allowed electronic. The system offered generic upload buckets — so families uploaded whatever they guessed, and coordinators inherited the sorting.
Twenty years of workarounds that nobody designed and everyone lived with. The failure rate wasn't a user problem. It was a gap between three layers, each with a different owner who assumed the other two were aligned.
How I worked
Sequential 1:1 sessions with operations — quality control, program directors, provider services, each session refining what the previous one validated. Working prototypes, not presentations.
The key moment: operations saw their own internal checklist — the tracking system they'd always used — surfaced directly to families as a clear, personalized document list. They stopped being skeptical. They weren't reviewing my designs; they were seeing their expertise made visible.
The research practice worked with military families directly. One spouse described submitting documents into "the abyss" — no confirmation, no status, no idea if anyone was reading. Each family taught us something the system couldn't.
As the work matured, the practice grew a data layer: direct measurement against the live production databases — cohort analyses, funnel measurement, evidence chains from the raw record to the decision it informed. When leadership asked whether the new system was actually faster, whether families abandon applications, what happens after submission — the answers came from production data the same day, not from anecdote. Design decisions and leadership readouts stopped running on assumption.
What it produced
First-attempt completion went from rare to the norm in pilot — measured in production, not claimed
Complete applications turned around in a business day instead of weeks
The root cause of resubmission reframed: document timing and hidden requirements, not user error — which changed what got built
A design rationale registry: every decision traces to a person, a quote, and a measured result
A dozen concurrent workstreams run by one person with AI infrastructure holding the context
A research practice trusted at the program's most senior levels — because every finding survived being checked
Leadership questions answered same-day from production data — the evidence base ran on live measurement, not quarterly reports
What I learned
This is where the three-layer diagnostic became a formal methodology. The gap between requirements, operations, and system behavior is where every problem lived — invisible because each layer had a different owner.
And the research compounds. The person doing family outreach is the same person who found the resubmission root cause, designed the fix with operations, and caught the next problem expressing itself differently. That synthesis — holding all the threads at once — is what one person with the right tools can do that a fragmented team cannot.
