Data science leader in growth & marketing with 10 years in fintech — 8 at PayPal, plus consulting for American Express — specializing in experimentation, attribution, incrementality, and AI-powered analytics tools you can try live.
| Segment | Net inc. $ | ROI | Decision |
|---|---|---|---|
| Segment 2 | $1,062,000 | 472% | Fund |
| Segment 5 | $624,000 | 650% | Fund |
| Segment 4 | $332,500 | 64% | Hold |
Diagnose where growth is leaking, build the measurement to quantify it, and translate the numbers into decisions.
I lead growth and marketing analytics at PayPal — targeting and audience selection, behavioral segmentation, propensity scoring, attribution modeling, and experimentation standards across A/B tests, holdouts, and incrementality measurement. I've supported campaigns across email, onsite banners and placements, direct mail, digital (paid search and social), and sales channels, along with the data partnerships and vendors behind them, working cross-functionally with Marketing, Product, Commercial, Risk, and Finance.
Before PayPal, I built data engineering and reporting solutions for American Express as a consultant — the start of 10 years working with financial-services data. Increasingly, I build AI-enabled analytics tools with Claude: deterministic math does the math, AI does the synthesis. That principle cut recurring reporting turnaround from about two days to under four hours and standardized how experiment readouts and budget recommendations get made.
From targeting and audience selection through channel launch, performance tracking, and reporting — I've owned the full loop across email, onsite placements, direct mail, paid search and social, and sales channels, including the vendors and data partners that power each one.
I take an ambiguous business question, design the measurement to answer it, and land on a clear call — Fund / Hold / Suppress, Ship / Kill / Iterate — with the revenue case attached, not just the analysis.
Cross-functional partnership with Marketing, Product, Commercial, Risk, and Finance, vendor and partner management, and executive reporting that leadership actually reads — performance framed as decisions, not dashboards.
Self-serve dashboards and reporting infrastructure in BigQuery, Python, and Tableau, plus channel pacing and forecasting across weekly, monthly, and quarterly horizons — built to catch underperformance early and reallocate budget while a campaign is still live, not after the fact.
Working AI tools built with Claude — each one solving a real problem from my day-to-day analytics work. Open them and use them live.
Fund what moved the needle — not what merely followed it. Scores segments on true incremental lift, tags each Fund / Hold / Suppress by ROI, and generates the executive budget recommendation, validated against a randomized holdout.
Open the live tool
Is the lift real, or just noise? A two-proportion significance test for the holdout tests that validate every incrementality call — control vs. treatment, lift, confidence interval, and a clear verdict.
Open the live toolOffline channels can't be A/B tested at the user level — there's no impression to randomize and no holdout to carve. A billboard goes up in Chicago, applications rise, and everyone claims the win. But how much of that rise was the billboard?
Find Chicago a statistical twin — a city matched on income, customer mix, traffic, business density, and seasonality. Detroit qualifies. Chicago gets the billboards; Detroit doesn't. Then compare each city to itself, before and after: a difference-in-differences design.
Chicago rose by 300 applications — but Detroit rose by 70 with no billboard at all. That 70 is the market tide. Only the gap between them is the billboard's real effect.
Illustrative numbers. The method — matched-market difference-in-differences — is the point.
New-application volume (PIF) had been trending down quarter over quarter. Total portfolio volume looked fine on the surface — which made it tempting to file the PIF decline as noise and move on.
An end-to-end funnel review told a different story: portfolio growth wasn't coming from a healthy new-customer pipeline — renewals were doing the work. Meanwhile, federal Do-Not-Contact suppression volume was climbing quarter over quarter, shrinking the population the PIF funnel could even target in the first place.
I brought the split to cross-functional stakeholders and reframed the question — not "why are applications down," but "how do we solve the suppression constraint that's shrinking the funnel." That moved the roadmap from optimizing PIF campaigns to fixing the actual input constraint.
Illustrative numbers. PIF = new-customer application funnel; DNC = Do-Not-Contact suppression list.
Questions about the tools, a measurement problem worth talking through, or just want to connect? My inbox is open.