617: Causal Modeling and Sequence Data

Episode Summary

Executive Summary: Sean Taylor discussed launching Motif Analytics, a startup focused on analyzing event sequences to uncover causal patterns rather than just reporting metrics. He explained how causal modeling informs better product decisions, why sequence data demands new query and visualization approaches, how his Lyft and Facebook experience shaped his thinking, and why rigorous evaluation, uncertainty communication, and strong teams matter in applied data science.

Main Topics: Motif Analytics and the startup transition: Taylor revealed his new role as co-founder and chief scientist at Motif Analytics, explaining why he left larger companies for a smaller, high-alignment startup where the product is still being defined. Sequence analytics as a new category: He described Motif’s focus on event sequences generated by user behavior, arguing that these rich logs are usually flattened into metrics and lose important information that could reveal friction and success patterns. Causal modeling as the core lens: Taylor framed analytics as fundamentally causal: companies need to know what happens when they change something, or what caused an observed effect, not just what happened. Lyft’s marketplace, experimentation, and simulation work: He explained how causal modeling, simulation, reinforcement learning, and Bayesian optimization supported Lyft’s pricing and dispatch systems, where small improvements can have large marketplace effects. Tool skepticism, evaluation, and diagnostics: He argued that models and packages are tools, not magic; proper evaluation, diagnostics, benchmarking, and uncertainty communication are essential to avoid overtrusting methods like Prophet or causal frameworks. Hiring, learning, and the information systems background: Taylor emphasized intellectual curiosity as the most important hiring trait and described his Information Systems PhD as a blend of computer science, social science, and business thinking oriented toward real-world system adoption.

Key Arguments: Most analytics questions are implicitly causal, so making them explicit improves rigor and business usefulness. Sequence/event data contains richer information than aggregated metrics, especially for rare or localized user behaviors. Motif is not just a reporting tool; it aims to help users uncover causes, consequences, and root causes in behavior sequences. Causal inference requires more than one method; practitioners need experiments, DAGs, design thinking, and validation. No model or package can be trusted without evaluation against known outcomes and robust diagnostics. Uncertainty should be communicated visually and explicitly, especially when tools generate hypotheses rather than final decisions. Large, long-term platform investments can unlock major gains, but only if leadership supports slow payoffs and cross-functional teams. Strong hires are intellectually curious, adaptable learners who can generalize lessons across problems. Information systems is valuable because it combines technology with human adoption and organizational behavior. At Lyft, even small algorithmic changes in pricing or dispatch can materially affect rider experience, driver supply, and marketplace efficiency.

Data Points: Episode number: 617 - Super Data Science Podcast episode featuring Sean Taylor LinkedIn/Twitter audience size: 50,000 followers on Twitter - Taylor said his audience grew because he has been on Twitter for a long time Interview volume: 400-500 people - Taylor estimated how many people he has interviewed across his career Bayesian optimization platform rollout time: over a year - He described the long lead time before Lyft could run high-bandwidth experiments with continuous parameters Lyft interview reference: three years - He referred to spending three years at Lyft as a student of ride-sharing Web post engagement: 12,000+ impressions - John Crohn noted the LinkedIn post announcing the episode received large engagement Social reactions: a couple hundred reactions - John Crohn referenced audience interest in the episode announcement Price example: $1 - John described Lyft’s priority option as an extra dollar for faster pickup Pickup time example: 4 minutes vs 5 minutes - John used Lyft’s priority feature as an example of pricing and dispatch tradeoffs

Pivotal Quotes: "The two joys in work are like doing the fun thing and then telling people about it." — Sean Taylor: Explaining why it was hard not to reveal his new startup before the interview "I think of it as like: if we succeed at this tool, it will be like as if you had me helping you do your data analysis at scale." — Sean Taylor: Describing Motif Analytics’ role as a chief-customer style product vision "Models and packages are just tools. We attribute magical powers to them as if their creators have somehow anticipated all the idiosyncrasies of your particular problem." — Sean Taylor: Discussing limitations of Prophet and the need for evaluation

Implications: For data teams, the episode argues for treating analytics as causal, sequence-aware, and evaluation-driven. For industry, it signals demand for tools that surface root causes and uncertainty, not just dashboards, and for startups to build around under-served data structures and workflows.