Case studies in research,
strategy, and product direction
- What changed
- The founders added employer and union channels alongside direct-to-consumer, and started re-engagement emails for trial users who never converted.
- My role
- Solo researcher and strategist for the founders, from study design and recruiting through sessions, behavioral analytics, and recommendations
- Timeline
- About 3.5 months: 6 weeks of research through the January readout, then advisory through March
- What changed
- A 6-person usability test, run before anything was built, showed that leading with the score undermined the advice that followed. The team rebuilt the flow around specific recommendations and moved the score to a secondary, explained role.
- My role
- Solo researcher and strategist, from study design through recommendations
- Timeline
- About 2 weeks
- What changed
- A redesigned navigation, content model, and clickable prototype for a 100,000+ user AI and analytics platform. DoD paused development in June 2024 to fix infrastructure, as recommended, and the $15B recompete centered on a modular architecture.
- My role
- Lead researcher and platform strategist on a cross-functional team
- Timeline
- About 14 months
- What changed
- The client relabeled its transaction states and reordered onboarding after an audit found 17 friction points, most of them in how pending transactions read. Compliance approval went from 6–8 weeks to 3–5, and the roadmap projected a 22% lift in 90-day retention and about half as many support tickets.
- My role
- Research lead and strategic advisor
- Timeline
- 6 weeks: 1 of discovery, 2 of research, 1 of synthesis, 2 of readout and implementation support
AI Enablement Measurement Tool
An AI-powered research design assistant that helps practitioners measure what AI enablement actually does to the people using it. You answer 19 questions about your tool, your employees, and your organizational context, and it generates a tailored study design that’s honest about what’s feasible and specific about what the findings can and can’t claim. In testing, it overstated the statistical power of a sample study, so I rewrote the prompt to correct it.
Confidence Calibration in AI Model Outputs
A Python tool that checks whether an AI model’s stated confidence matches how accurate it is. Built as a companion to the AI evaluation framework in Ground Truth series 1: A Framework for Rigorous AI Product Practice.
Still asking the same question
I’ve wanted to know how things work for as long as I can remember. Mostly that meant taking them apart, and my dad still hasn’t fully forgiven me for the pool filter. It also meant sorting the world into categories, from backyard insects to wrapping paper patterns. The question that fascinated me longest, though, was about people. Why did adults say one thing and then do something else? Just resting their eyes while snoring loud enough to rattle the windows, or promising they knew a shortcut right before making three U-turns in a row.
I’ve been chasing some version of that ever since. After starting in marketing intelligence, I moved into product research at Fisher-Price, where the end users were kids. Then I earned a PhD in psychology and quantitative methods and went on to lead research for agency and consulting clients, including pharma and healthtech companies, a bank, and the Department of Defense. Behavioral data shows what people do, and talking with them shows what they say. I still go looking for the places where those don’t match.
These days a lot of my questions are about when people trust what AI tells them, when they check, and what happens to that trust the first time it’s wrong.
Ground Truth on Substack
Where I write about research rigor in AI products, and the gap between AI output that sounds confident and output that’s confident and accurate.