Situation: Franchises, hotels, and retail chains with 50+ locations sought to improve pricing and operations but relied on informal tests, leaving revenue impact unclear.

Task: Build a B2B platform enabling enterprises to run scalable brick-and-mortar experiments on pricing, promotions, staffing, and operations across stores.

Action: Founded the company; developed patent-pending methods for attrition diagnosis and convenience sample adjustment; built software to operationalize methods; raised $2M and led 11-person team.

Results:

• Secured $2M at ~$10M post-money valuation.
• Learned customers wanted answers, not tools, and fragmented budget ownership made selling a experimentation platform uneconomic.
• Shut down based on evidence; returned remaining capital and placed entire team.

Situation: Windows 11 launched under time pressure, with leadership relying on observational scorecards vulnerable to selection bias.

Task: Identify and estimate the causal impact of Windows 11 on core outcomes, and determine whether OS-level experimentation was feasible under rollout constraints.

Action: Identified upgrade notifications as random variation source; designed encouragement experiment using IV/LATE; led A/A tests, SRM checks, and 38-person team through implementation.

Results:

• Reversed executive conclusions driven by selection bias.
• C-suite extended experiment from six weeks to eleven months.
• Informed product strategy shift toward customer jobs-to-be-done.

Situation: In Mexico, billions in anti-poverty infrastructure funds flowed to municipalities, but misuse risked denying basic services to poor communities.

Task: Design a politically sensitive randomized audit trial the federal audit agency could run in practice and defend to Congress.

Action: Embedded with audit agency to design a 3-arm randomized audit trial (85 municipalities, 17 states) using transparent and defensible public lottery assignment.

Results:

• Produced credible causal evidence on whether audits change compliance.
• Null result redirected attention from audits to enforcement incentives.
• Published in the Journal of Development Economics.

Situation: The Asian financial crisis revealed the World Bank was making credit decisions with 3-12 month-old data.

Task: Extend an advanced-economy forecasting system to deliver monthly global nowcasts under severe data sparsity.

Action: Built 148 country trade matrix to impute missing trade data for poor economies using partner-reported flows; adapted OECD forecasting system using ARIMAX models; automated daily updates with structural-break detection and validation.

Results:

• Reduced macro data lag from 3-12 months to real-time with daily-updated monthly nowcasts.
• Enabled real-time credit decisions with daily automated updates.
• Methods adopted by the Netherlands Bureau for Economic Policy Analysis (CPB World Trade Monitor); cited by The Economist.