Kekona

Statistician, AI product data scientist, college lecturer, youth sports coach, and weekend builder. Usually not all at the same time.

Some of what's below reached millions of people. Some of it reached my wife, my kids, and whichever neighborhood kids were over that day. I'm proud of all of it, and happy to talk about any of it.

Right now I'm taking on consulting and advisory work in measurement, experimentation, and AI evaluation. I'll also help you build something out of wood, or teach a statistics class people actually enjoy. Whichever it is, email me.

Select Professional Data Science Projects

AI model quality for Microsoft Teams
Microsoft
Model evals tell you the answer was good. They don't tell you the person got what they came for. I worked with applied science and product teams to sharpen the evals, connect them to customer outcomes, and find new ways to measure whether the AI actually helped.
SharePoint: measuring new product launches
Microsoft
A brand-new feature launched to an enormous existing customer base looks like noise on every chart that matters. I built the framework that found real growth signal in small early data, and showed what each launch added to the ecosystem around it.
SharePoint: OKRs and metric strategy
Microsoft
Every team had its own number, and the numbers disagreed. I built the OKR program and the shared definitions behind it, so teams argued about what to do next instead of whose dashboard was right.
OneDrive consumer growth and monetization triggers
Microsoft
People don't upgrade because you asked nicely; they upgrade at particular moments. I found those moments in the lifecycle data and tested them, so the product could ask at the right time rather than the loudest one.
R that scales
Fellows Statistics
The statistical methods people trusted were written for data that fits on one machine. I refactored them to run distributed, so the same analyses could be run on data many times larger without giving up the method.
Retail expansion and site selection
Hot Topic · Torrid · BoxLunch
The real estate team knew the market in their bones, and an outside consultant's model hadn't earned their trust. Working alongside them, I rebuilt it with machine learning over geography, demographics, and customer spending data, so their instincts, the models, and the first-year revenue forecasts finally lined up.
Demand forecasting meets allocation
Hot Topic · Torrid · BoxLunch
Buying and allocation ran on spreadsheets and arithmetic that mostly worked, until they didn't, and the misses turned up months later on the markdown rack. I added statistical rigor to what merchandising already did, connecting demand forecasts to what each store actually received.
Statistics as evidence
Welch Consulting
In employment litigation, an analysis is only worth what it survives. I built the economic and statistical models used as evidence in nationwide cases, where every assumption had to hold up under cross-examination.