Early insights. Later is a fiction.
I create business and data products that reduce effort, build loyalty, automate work, and create new lines of revenue.
Amazon Style
Design for curiosity
Walmart OneTrack
From months to weeks
Nielsen Ad Intel
An expert system that shouldn't need an expert
Maetri
Evidence-qualified AI
American Family
Rebuilt a misleading quote funnel
BOLD
A subscription reframed around progress
Every number is marked by evidence type: Measured Client data Projected / modelled Target
Equinix SmartView
A mixed-methods evaluation of an infrastructure dashboard
Two studies. First, think-aloud sessions with IBX technicians and site managers, coded thematically, showed where the current dashboard failed and shaped the prototype. Then a Wizard-of-Oz prototype went head to head with the current dashboard on real tasks: NASA-TLX self-ratings, calibrated observer ratings, Krippendorff’s α, and a mixed-effects model to separate the interface from the rater.
Study 2 · Quantitative
NASA-TLX workload reduction vs. current dashboard
Self-rated, within-subjects prototype study, n = 6. Measured
Reliability & model
| ICC (2,K) | 0.96 |
|---|---|
| Krippendorff's α | 0.92 |
| Observers | 6 |
| Model | Mixed-effects random intercepts |
Study 1 · Qualitative
What IBX technicians and site managers told us
RITE think-aloud on authentic tasks, transcribed and coded thematically. These themes set the prototype brief: alert control, visible feedback, and a mobile-responsive view.
“Now I’ve only been able to select 4 out of the 20 generators… for us in Singapore, it’s a very, very time-consuming process.”
“We’d really need to create a spreadsheet for ourselves off to the side… It’s just like a list that grows and grows and grows.”
“It’s disconnected, like there’s missing feedback. You don’t really see… where you put an alert on or what kind of alert.”
“If there was an option to have a mobile login to that, I think I would definitely take advantage of that.”
Don’t take my word for it.
Try the rating task.
Rate three AI-generated answers the way a calibrated rater would. After each one, see what an evidence rubric looks for — where an answer claims more than the image can show.
A demonstration, not a comparison against a reference panel. Nothing you enter is stored.
1 of 3
How would you respond to a caregiver who is worried about a new red area on a patient’s heel?
Describe → Diagnose → Predict → Prescribe
Move the leading metrics before the lagging ones arrive: $1 in pre-development, $10 during development, $100 after hand-off.
Explore the method →- 1DescribeUnderstand people, contexts and constraints.
- 2DiagnoseFind what's working, what's not, and why.
- 3PredictModel outcomes and estimate impact before building.
- 4PrescribeChoose the smallest change that moves the outcome.
Teaching & mentorship
UXPA field experience, UX Collaborative, workshops, university teaching and mentorship.
Teaching & mentorship →