Terry Zeng

I'm a data scientist at Rokt with a product remit alongside it: I own the roadmap and execution

Most of the work is experimentation and causal measurement: incrementality tests that separate our lift from what would have happened anyway, difference-in-difference on new bidding products, churn and budget models in production, auction and recommendation models, and an AI agent that debugs campaign delivery.

Before Rokt I was the founding data hire at Humankind, an embedded shopping service later acquired by Tulip, where I built the analytics function and their first experimentation program.

Senior Data Scientist, Product at Rokt (2024-07 – Present)

Marketplace for e-commerce confirmation page and in-transaction ads

• Lifted revenue per transaction +20% above baseline for Rokt Upsells by designing the A/B experimentation program that accelerated hypothesis testing for formats and offers for ~2 million transactions per day
• Tripled revenue to $75M for Rokt payment page ads by productizing do-no-harm holdout experiments that drove partner confidence by communicating statistical significance, cutting onboarding time by two-thirds
• Led the 0 to 1 launch of conversion elastic bidding, increasing spend +15% on campaigns by dynamically adjusting bids to maximize cost-per-conversion/CPA efficiency, measuring impact using difference-in-differences causal inference
• Won $10M in net-new advertiser spend by designing the incrementality experiments that isolated product-driven conversions from baseline, then equipping GTM and cross-functional stakeholders to sell the measured lift
• Prevented $2.5bn in partner revenue loss by productionalizing a logistic regression model that predicts shopper churn and reorders the auction to prioritize advertisers that minimize churn risk
• Drove a 7 of 10 one-year promotion among the new-grad data scientists and analysts I mentored, by building the mentorship program that paired technical coaching with exposure to senior stakeholders
• Unblocked ~$100M in advertiser spend by shipping an AI agent that diagnosed infrastructure and operational failures in campaign delivery, cutting resolution from 1.5 weeks to 24 hours

Senior Data Scientist at Humankind Technologies (2022-08 – 2024-02)

Embedded chat connecting e-commerce shoppers with product experts (acquired)

• 3x'd shopper onboarding and grew usage-fee revenue 230% by owning the funnel and building the company's A/B experimentation program to decide which formats, copy, and placements shipped
• Lifted purchase conversion 26% by developing the logistic regression model that identified personalized recommendations, expert response time, and retargeting as drivers, then reprioritizing the roadmap
• Retained 10% of business's total ARR by building a client health scoring model that identified accounts at risk of churn and directed product and CS prioritization

Analyst, Venture Equity at CircleUp (2021-11 – 2022-08)

Consumer-brand VC fund built on an in-house data platform (acquired)

• Deployed $17M at roughly 3x return by building a PySpark NLP pipeline that TF-IDF vectorized purchase-criteria language from reviews, narrowing 30 candidates to 6 investments
• Opened up a $500M in investable universe by building the Python/SQL/BigQuery models that identified and validated new value indicators for startups

Strategy Consultant at Partners in Performance, Part of Accenture (2020-10 – 2021-11)

• Eliminated roughly $33M in annual production loss at a $350M/year gold mine by building the Python/SQL pipeline that quantified the unmeasured operation and prioritized its loss drivers
• Added $25M in annualized revenue across 10 declining sales teams by defining their north-star and KPI framework in a SQL/Tableau dashboard, then driving upsell, cross-sell, and churn plays to adoption

McGill University, BA in Economics, Computer Science, and Finance