seanjtaylor.com

What is Sean J. Taylor known for?

Sean J. Taylor is known as a data scientist, social scientist, statistician and software developer who specializes in causal inference and business decision-making, with a particular focus on building tools for practitioners solving real-world problems. His background spans academia and industry: a Ph.D. at NYU's Stern School of Business, research and management roles at Facebook's Core Data Science team and Lyft's Rideshare Labs, and now co-founder and chief scientist at his startup, Motif Analytics.

He is also known as a generalist who deliberately works across fields—collaborating with computer scientists, economists, political scientists, statisticians and machine learning researchers—and as a communicator on experimentation, statistics and causal methods. His site collects talks and podcast appearances on topics such as minimum viable experimentation, when not to use SQL, causal inference with sequence data, and matching in two-sided marketplaces.

Where to start depending on your interest

• Practitioners designing experiments: the "Minimum Viable Experimentation" podcast appearance and his talk on the relationship between experimentation and causal inference.
• Analysts weighing causal methods: the talks on when causal inference is actually needed and on causal models in practice.
• People curious about his career path: the Background section tracing his route from the Federal Reserve Board and software engineering through Facebook and Lyft to founding a company.

A useful next step is to pick one talk or podcast episode that matches a decision you currently face—say, whether an experiment is warranted—and use it as a concrete lens rather than browsing the whole list.

What does Motif Analytics do and how does it relate to Sean J. Taylor's work?

Motif Analytics is the startup Sean J. Taylor co-founded and where he serves as chief scientist, according to his site. The page itself does not describe Motif's product or services beyond that role, so the site establishes the connection to his work rather than the company's specific features.

How it connects to his work

Taylor describes himself as a data scientist, social scientist, statistician and software developer focused on causal inference and business decision problems, especially building tools for practitioners. His background—research scientist and manager on Facebook's Core Data Science Team, then data scientist and manager on Lyft's Rideshare Labs team—points to applied experimentation and causal methods in large-scale product settings. Motif is the commercial continuation of that interest: turning methods for causal and business-decision problems into tools for people doing the work.

His listed talks reinforce the same theme: minimum viable experimentation, when not to use SQL, causal inference and sequence data, and the relationship between experimentation and causal inference.

A practical way to use this

If you are evaluating Motif, treat Taylor's public writing and talks as a way to judge the team's approach to experimentation and causal analysis, and check Motif's own site for what the product actually does. For learning the underlying methods, his talk "Minimum Viable Experimentation" and his essay-style keynote "In Defense of Curve Fitting" are concrete starting points. His site is Sean J. Taylor.

When should a practitioner avoid using causal inference versus simpler methods like SQL analysis?

A practitioner should avoid causal inference when the decision does not depend on a cause-and-effect claim: when you need a description, a count, a trend, a ranking or a monitored metric, plain SQL analysis is usually faster, cheaper and easier to explain. Causal methods earn their cost when someone will act on the answer and a wrong "this caused that" conclusion would lead to a costly or harmful decision.

Sean J. Taylor's page frames his work around "methods for solving causal inference and business decision problems" and building tools for practitioners, and his talks include "When do we actually need Causal Inference" and "When not to use SQL" (Sean J. Taylor). Read together, those titles point to the same judgment call from both sides: neither tool is the default.

A practical way to choose

Example: an analyst asked why signups dipped last week can answer most of it in SQL — traffic mix, channel shifts, broken funnel steps. That is description, and causal machinery adds delay without adding truth. The same analyst asked whether a new onboarding flow caused more signups needs a comparison group; if a randomized test is possible, that is the cleanest route, and if not, causal methods with explicit assumptions are the alternative.

Decision criteria

• Ask whether a counterfactual is required. If no one will compare "what we did" with "what would have happened otherwise," skip causal inference.
• Check whether an experiment is feasible. A randomized test is often simpler and more credible than elaborate observational adjustment; causal inference is most valuable when randomization is impossible or unethical.
• Weigh assumption risk. Observational causal estimates rest on assumptions (no unmeasured confounding, correct model form) that are hard to verify; if stakeholders cannot tolerate that uncertainty, a descriptive answer honestly labeled as such is better.
• Match method to audience. A descriptive SQL result is easy to defend in a meeting; a causal estimate needs its assumptions explained, so reserve it for decisions that justify that conversation.

A useful next step is to write the decision and the counterfactual in one sentence before choosing a tool. If the sentence reads "we need to know how many and where," query the warehouse. If it reads "we need to know whether changing this will improve that," design the comparison — ideally an experiment — and treat SQL as the data-prep layer rather than the answer.

How does Sean J. Taylor connect experimentation methods with causal inference for business decisions?

Sean J. Taylor frames experimentation and causal inference as complementary tools for the same job: making defensible business decisions from data. On his site, he describes himself as a data scientist, social scientist, statistician, and software developer who specializes in "methods for solving causal inference and business decision problems" and is "particularly interested in building tools for practitioners working on real-world problems." That practitioner focus is the thread linking the two areas.

How the two areas connect on the site

The site groups his work under headings including "Data Science," "Social Science," and "Experimentation and Statistics," and lists talks that treat the relationship directly:

• "The Relationship between Experimentation and Causal Inference," a talk at the Nubank Data Science meetup.
• "When do we actually need Causal Inference," a talk at the New York Open Statistical Programming Meetup.
• "Minimum Viable Experimentation," on the Analytics Engineering Podcast.
• "Causal Inference and Sequence Data," on the Super Data Science podcast.
• "Causal Inference Approach to Matching in Two-Sided Marketplaces."
• A keynote, "In Defense of Curve Fitting," at the 2020 Causal Data Science Meeting.

Read together, these titles sketch a practical position: randomized experiments are the cleanest source of causal evidence, but they are not always feasible or sufficient, so observational causal methods fill the gaps—and the choice between them should be driven by the decision at hand, not by methodological fashion. The "when do we actually need causal inference" framing suggests he treats causal rigor as something to invoke deliberately rather than by default.

What the background adds

His career path runs from the Federal Reserve Board and software engineering through a Ph.D. in Information Systems at NYU Stern (dissertation on social influence from online social signals) to research science at Facebook's Core Data Science team and Rideshare Labs at Lyft, and now co-founder and chief scientist at Motif Analytics. That mix of academic social science, large-scale product experimentation, and tool-building explains why his treatment of causal inference tends to be operational: the goal is methods practitioners can actually run.

A concrete reader scenario

Suppose you run a subscription business and want to know whether a discount causes customers to stay longer. A randomized test answers that cleanly but may be expensive or ethically awkward at scale. A practitioner following this line of thinking would ask first what decision the estimate must support and how precise it needs to be, then choose between a randomized design and an observational approach with explicit assumptions—rather than defaulting to either.

Next step

For someone weighing this trade-off, the most direct starting points are the Nubank talk on experimentation and causal inference and the "When do we actually need Causal Inference" talk, both listed on Sean J. Taylor. The newsletter linked there is the way to follow new material.

What practical resources does Sean J. Taylor offer for learning causal inference and statistics?

Sean J. Taylor's site is most useful as a curated entry point rather than a structured course: it points to talks, podcasts, and written work where he explains causal inference and experimentation for practitioners. The practical value is in hearing how these methods are applied to real business and product decisions, not in step-by-step tutorials.

What the page offers

• Talks and interviews on causal inference. The page links to a talk on "When do we actually need Causal Inference," a Nubank meetup talk on the relationship between experimentation and causal inference, a TWIML interview on causal models in practice, and a talk on causal inference for matching in two-sided marketplaces. These suit someone who already knows the basics and wants judgment about when causal methods are worth the cost.
• Experimentation content. "Minimum Viable Experimentation" on the Analytics Engineering Podcast and a keynote, "In Defense of Curve Fitting," address how to run and interpret experiments without overbuilding them. Useful for product analysts and data teams setting up A/B testing.
• Statistics and methods discussions. Episodes on the Learning Bayesian Statistics podcast and the Super Data Science podcast cover statistical reasoning and sequence data, aimed at people comfortable with modeling.
• A practitioner angle on tooling. "When not to use SQL" (NormConf) and the "What's New in Data?" podcast speak to working data scientists choosing tools.
• Background and links. His history at Facebook's Core Data Science team, Lyft's Rideshare Labs, and his NYU Stern PhD, plus links to Google Scholar, GitHub, and a newsletter, give you a way to find his papers and code.

How to use it

Pick by your gap. If you can run an A/B test but struggle to decide when a causal claim is justified, start with the "When do we actually need Causal Inference" talk and the experimentation-versus-causal-inference talk. If you want applied modeling examples, the TWIML and Nubank talks are closer to case work. For formal foundations, his Google Scholar page is the better route than the podcast list.

For structured coursework alongside these, university-hosted materials such as MIT OpenCourseWare or Stanford Online cover statistical foundations systematically; the site itself is best treated as practitioner commentary, not a curriculum.

What career path led Sean J. Taylor from the Federal Reserve and Facebook to founding a startup?

Sean J. Taylor's path runs from economics and software engineering through large-scale data science to co-founding Motif Analytics, where he is chief scientist. The through-line is causal inference and experimentation methods applied to business decisions, plus a stated preference for building tools practitioners can use.

The path in order

• 2000–2004: Undergraduate at the University of Pennsylvania in economics, finance and information systems.
• 2004–2006: Research assistant at the Federal Reserve Board — an early research role rather than a data science one.
• 2006–2008: Software engineer at Matrix Group, International, adding engineering experience before graduate study.
• 2008–2013: Ph.D. at NYU's Stern School of Business in information systems; dissertation on social influence from online social signals, advised by Sinan Aral.
• 2012–2019: Research scientist and manager on Facebook's Core Data Science Team.
• 2019–2022: Data scientist and manager on the Rideshare Labs team at Lyft.
• 2022–present: Co-founder and chief scientist at Motif Analytics.

What each stage contributed

The Federal Reserve and Matrix Group years supplied the two halves he later combined: quantitative social-science research and working software. The Stern Ph.D. turned that into a research identity around online social signals, which maps directly onto Facebook's questions about influence and measurement. At Facebook and Lyft he moved from research into managing data scientists, working on causal inference and experimentation at scale in two very different marketplace settings. Motif Analytics is the natural next step: productizing methods for practitioners, consistent with his stated interest in tools for real-world problems.

A decision criterion for readers

If you are mapping a similar route, the pattern to notice is not the job titles but the accumulation of one method area across contexts — academic social science, a social network, a two-sided rideshare market, then a vendor tool. Depth in experimentation and causal inference transferred; the industry did not. For methods background, his site collects talks and podcast appearances under Data Science, Social Science, and Experimentation and Statistics. Related practitioner writing appears at Motif Analytics.