How I work
Land the decision with data, then build something leaders will use
Most of what I deliver depends on people I don't manage, so I lead through influence: setting standards, building operating rhythms, and bringing senior leaders decision-ready insights.
Land decisions with data
I run the analysis that turns a debate into a recommendation, and I state the tradeoffs plainly.
See it in the performance rating modelPilot, iterate, scale
I ship a usable first version, learn from the people using it, and release improvements on a steady cadence.
See it in store talent planningBuild for the leader who has to use it
A simple process that leaders adopt beats a perfect one they ignore. I favor designs that scale.
See it in the AI audit agentSeeing more difference between people, without changing how managers rate
How six ratings become one overall rating
Pick an example or set the ratings yourself. Behaviors and tier names are illustrative.
Go deeper: what one rating changes, the three-tier view, the math
Before: three tiers
Proposed: five tiers
Bars count rating combinations (of 729), not people.
| Behavior | Weight | Rating | Points | Weighted |
|---|
What the analysis showed
Weights change little. Cutoffs change a lot.
Scoring 1, 4, 9 and a plain 1, 3, 5 ranked people almost identically (correlation 0.99). Where the cutoffs sit moved the distribution far more.
Fitting tightly to a target creates weight cliffs.
One cutoff set matched the target best on paper but left roughly 1 in 6 evaluations where the same pattern of ratings landed in different tiers, decided only by a 20% versus 15% weight. The proposed cutoffs cut that to under 1 in 50.
The proposed cutoffs land close to the target.
In every tier the result is within about five points of the target, with the middle tier running slightly high. Calibration, not the formula, closes the rest.
The recommendation
Keep the six behaviors, weights, and points as they are. Add two cutoffs, placed in the gaps between populated scores, so no rating tips on a small change in weight. Close the remaining gap at the middle tier through calibration.
For managers, nothing changes. They rate the same behaviors the same way, and the same ratings now show more of the difference between people.
Moving store talent planning from spreadsheets into one Workday Extend experience
Excel, built by hand
A district process I built and iterated in Excel before it had a system home.
Excel and Teams at store scale
Hourly and leadership workflows split across files and conversations.
One Workday Extend experience
Hourly and leadership workflows in one place, improved through pilot, iterate, scale.
Pipeline identification rate
Promoted within two years of pipeline placement
Bars share one scale, 0 to 40%. Figures describe the store talent planning program overall, driven by greater visibility and leader action.
Building an AI agent that finds the riskiest performance evaluations first
Manual comment audit
Surfaced high-risk evaluations for HR business partners to review.
About 1,400 corrections
High-risk evaluations were corrected before compensation planning.
The audit became an agent
The audit logic was built into an agent that ranks evaluations by risk.
Enterprise tool
Used by about 250 HR business partners across populations.
Human in the loop: people first, AI second
The agent builds the list. HR business partners review and make the decision. Strict guardrails prevent drift.
Target. Under 0.1% of evaluations with an issue at year-end close, and fewer associate escalations after release. This is the goal, not a reported result.