Manages a team of eight data scientists responsible for recommendation engines, product insights, and measurement tools to enhance marketing effectiveness and customer experience. Regularly partners with PM's, Engineering, UX, and Design to create strategic priorities, integrate ML into product, and deploy end to end data systems to productionize AI/ML models.Deployed Capital One’s newsfeed style recommendation engine using deep learning in pytorch. ML Lead for the Transformer-based recommendation model on the mobile homepage. Used to optimize conversion rate, estimated $30M in revenue lift.Deployed GBM marketing email arbitration algorithm which reduced opt-out rate by 20%.DS lead for Capital One’s Experimentation Platform. Advises stakeholders, builds statistical tools including deployment of sequential A/B testing algorithm on which is estimated to reduce length of digital experiments by 40%.Developed ML ecosystem from ground up for Capital One’s “Business Offers” product. Deployed Recommender model and a series of marketing (propensity) models which increased conversions by 30%.

Managed a team of four data scientists responsible for running experiments and developing customer behavior models to underwrite Capital One’s core credit products.Deployed updates to Capital One’s customer lifetime value models using bayesian hierarchical regression and causal inference, resulted in an annual lift of $80 million in revenue.Led Capital One’s Consumer Card consumer card experimentation program. Advised product leaders and designed analysis of experiments tools which reduced the cost of the credit experiment program by $50 million a year.Built causal inference models (GLM, causal trees) to predict customer spend/engagement, attrition, default risk models used by executives to make product strategy decisions for multi-billion dollar Card programs (e.g. Walmart, expansion market’s, COVID19 response)

Technical lead on Bayesian anomaly detection model used to identify customer impacting application failure – and isolate root cause resulted in 90% reduction of false negatives over vendor product. Deployed Bayesian state space forecasting models used by engineering teams across the Capital One to optimize resource allocationDeveloped gaussian process regression adopted by tech leadership to estimate customer impact of application failure.

Developed survival regression models used to predict risk of critical systems failure in FA-18/F-35 inventory Built forecasting (ARIMA, state space) and optimization tools s used by DoD leadership to inform aircraft acquisitions policy

Built fraud detection models supporting the Small Business Administration, Homeland Security, and Student Exchange Program

As an acting lead auditor on the Recovery Act Improper Payments team, Tom personally identified, research, and reported on over $4.6 million dollars of recoverable funds. In addition, Tom identified and referred $500,000 in fraud to investigators and $2.5 million in recoverable payments to other SBA OIG employees. The report can be found here http://www.sba.gov/office-of-inspector-general/7504/677751