Invited Speakers

David Purdy 

CEO and Founder, Palladio AI

David Purdy is Palladio AI's founder and CEO and an AI leader with nearly 25 years of experience. Previously Chief Data Officer at Chime, he founded and architected Michelangelo, Uber's machine-learning platform, and has held data leadership roles at Apple and Goldman Sachs. He is a named inventor on 12 patents and holds a PhD in Statistics from UC Berkeley.

Title: The double life cycle: How Games and Players grow together  (grow vs growth). Shall I plan it?

Abstract: Few teams think about the customer journey as obsessively as game studios, or instrument it as heavily. At Palladio AI, we work with game studios throughout their growth lifecycle to better engage players on their end to end journey.

We've identified key steps in how games and players co-evolve, each reshaping the other, and that loop is the double life cycle. It looks different by genre, by market, and by stage. We've also found the early player life cycle has surprisingly rich contours.

The same journeys run in every mobile app. Games run it on fast-forward, which makes them the clearest place to surface opportunities for the rest of mobile apps.

Sang Su Lee 

Staff Applied Scientist, Thumbtack Inc.

Title:  Certifying What Helps Customer-Return Timing: A Screen-and-Confirm Test for Conditioning Signals, and Why Decay Is Nearly Enough

 
Abstract: Practitioners enrich customer-return models with ever more signals (lifetime value, category, recency/frequency, calendar, geography), and the temporal-point-process (TPP) literature follows suit with covariate- and external-covariate-conditioned intensities. But does any of it improve the timing, and how would you know? A null (“feature 𝑋 doesn’t help”) is only meaningful if the model could have found a signal. We make two contributions—a method and a measurement—to answer this credibly. (i) A screen-and-confirm protocol that certifies whether a candidate signal improves a TPP’s event-timing likelihood: a positive control plants a coupling of known strength and confirms the model recovers it—for categorical and continuous encodings, and on a real clock-driven dataset (NYC taxi hour-of-day, where it recovers)—so a real-data null can be read as “no signal” rather than “weak method.”(ii) A model-free ceiling quantifying how little of customer-return timing is predictable at all (≲ 5% of gap variance from any covariate; returns are nearmemoryless). With these we certify a clean result on three public benchmarks (Amazon, Taobao, RetailRocket) and a real marketplace (Thumbtack): the inter-event clock—continuous-time decay, long known to beat frozen-intensity models—is nearly sufficient, and the conditioning the field keeps adding is redundant or harmful on top of it (≲0.06 NLL; ≈−64% MAE from decay alone on Thumbtack). We do not claim to discover that decay helps; our contribution is the tools that turn “conditioning doesn’t help” into a checkable, certified statement—plus an honest-evaluation account of the readout/leakage pitfalls we hit and retracted.

Iris (Nan) Wang 

Lead Data Scientist, Databricks

Nan Wang is a Lead Data Scientist at Databricks, where she leads high-impact data science initiatives across Growth, GTM, and B2B product strategy. Her work focuses on translating complex and ambiguous business problems into scalable data science solutions. She specializes in causal inference, machine learning, and optimization.

Title: Tradeoff-Aware Territory Design: A Multi-Objective Decision System for Enterprise Customer-to-Seller Allocation

Abstract: This talk presents a decision-system framework for enterprise customer-to-seller allocation, framing territory design as a multi-objective optimization problem rather than a static planning exercise. The system integrates predictive account value, causal estimates of relationship continuity, operational constraints, and human-in-the-loop workflows to generate scalable and adaptable territory recommendations. More broadly, the framework illustrates how data science can translate complex business tradeoffs into repeatable decision infrastructure for GTM planning.