Feb 8, 2024 · 39m · no-priors
No Priors Ep. 50 | With Stripe Head of Information Emily Glassberg Sands
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of No Priors, Stripe Head of Information Emily Glassberg Sands discusses how Stripe leverages generative AI internally and across its financial product suite, while sharing insights into data-driven decision science, macroeconomic trends, and the unique economics of AI startups.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 20% of the talking time here. How this is scored →
The hosts as informed peer 5.0 Guest teaching 4.3 Guest disagreement 0.2 The hosts pushing back 0.1
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 31:45 Rejecting Macro-Driven Micromanagement
Emily playfully rejects Elad's suggestion that macroeconomic swings dictate internal team headcount, explaining that shifting allocations based on short-term macro trends would create whiplash.
Hardest push from the hosts ▶ 30:08 Elad Pressing on Macro Signal UtilizationElad draws a comparison with Google AdWords to probe whether Stripe reacts to macro recessionary indicators by curtailing internal team investments.
Biggest teaching moment ▶ 33:45 Labor Economics Lens on Education AIEmily shifts Elad's framing away from purely elementary and college tutoring tools to the fundamental labor economics reality that education's primary economic value lies in credentialing and skill signaling.
The host holds their own ▶ 37:38 Elad's 1970s Four-Year Vesting Historical ParallelElad connects modern generative AI startup monetization dynamics to the historical origin of four-year stock vesting schedules in the 1970s and early profitable internet IPO waves.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Emily's Dual Role and the Information Org at Stripe | 4 | 3 | 0 | 0 | Elad welcomes Emily and sets up a broad opening inquiry into Stripe's information org and initial LLM exploration. Emily explains her dual role overseeing data foundations and self-serve business, as well as the initial bottoms-up LLM Explorer experiment. | |
| Scaling Internal AI Usage and Prompt Presets | 6 | 3 | 0 | 0 | Elad frames a three-part model for enterprise AI adoption (external, internal, vendor) and asks how Stripe encouraged internal adoption. Emily validates his framework and shares how internal prompt presets and style guides spread to thousands of employees. | |
| Seeding Bets with Applied ML Accelerator Teams | 4 | 4 | 0 | 0 | Sarah inquires about the transition from exploration to exploitation via Applied ML Accelerator teams. Emily describes how ring-fenced one-to-two pizza teams are funded out of the CTO's office as rotation opportunities for internal talent. | |
| Radar Assistant and Natural Language Rule Generation | 6 | 3 | 0 | 0 | Sarah asks about user-facing assistant capabilities, prompting Emily to describe Radar Assistant for natural language fraud rules. Sarah enriches the point by generalizing natural language policy description to broader decision engines across underwriting and risk. | |
| Sigma Assistant for Natural Language Business Insights | 4 | 5 | 0 | 0 | Emily explains Sigma Assistant for natural language SQL queries on revenue data and outlines Stripe's multi-year vision. She educates the hosts on potential 100-200 basis point uplifts from specialized financial foundation models and building an economic operating system. | |
| Organizational Scaling of AI Investments | 6 | 4 | 0 | 0 | Elad demonstrates technical grasp by asking detailed questions on model orchestration criteria including RAG, fine-tuning, open vs. closed models, latency, and cost. Emily explains Stripe's decentralized team selection model backed by centralized infrastructure and internal billing. | |
| Buy vs. Build and Custom Experimentation Infrastructure | 5 | 4 | 0 | 0 | Sarah and Elad explore custom experimentation infra and broader fintech AI white spaces. Emily explains why latency and reliability necessitate building in-house charge-level experimentation, and details merchant identity and compliance opportunities. | |
| Leveraging Payments Data for Real-Time Optimization | 4 | 6 | 0 | 0 | Sarah asks how Stripe leverages payments data back to merchants. Emily schools the hosts with concrete metrics, detailing backend ML retry routing recovering 10% of false declines and Smart Dunning cutting declines by 30%. | |
| Emily's Labor Economics Background and Decision Science | 4 | 5 | 0 | 0 | Sarah asks about Emily's labor economics background. Emily details her college audit study on female playwrights and explains how rigorous causal inference and econometrics form the foundation of her approach to data science at Coursera and Stripe. | |
| Macroeconomic Signals and Long-Term Strategic Planning | 5 | 4 | 1 | 1 | Elad asks if macroeconomic data dictates Stripe's hiring and team allocations like Google AdWords. Emily gently pushes back against short-term micromanagement, clarifying that Stripe takes a long-sighted view guided by user demand rather than macro fluctuations. | |
| AI's Role in Education and Labor Market Signaling | 5 | 6 | 1 | 0 | Elad asks about AI's impact across education levels. Emily reframes the prompt from classroom personalization to labor economics, educating on why labor market signaling, credentialing, and skill measurement matter just as much as learning acquisition. | |
| Unique Growth Dynamics of Generative AI Startups | 7 | 4 | 0 | 0 | Emily outlines four traits of generative AI startups on Stripe (upfront compute costs, instant global demand, subscription models, fast monetization). Elad demonstrates historical expertise by connecting rapid AI monetization to 1970s four-year vesting origins and early internet IPO velocity. |
Statements from this episode (18)
Sands: Stripe Can Learn and Action Interventions Improving Business Success
“Stripe's clearly helping companies run more effectively and also in a position to learn from its data what kind of interventions significantly improve companies' long-term success. And in some cases to actually action those.”
Sands: Three Stripe Engineers Built Internal LLM Explorer Beta in Three Weeks
“So it starts with a story of three engineers who hacked together in three weeks an internal beta for an LLM Explorer.”
Gil: Enterprises Adopt LLMs Internally for Efficiency Before Launching Externally
“I found in, in general, people have tended to follow the pattern that you mentioned, which is they start off kind of thinking, hey, what should we do externally? And then they immediately collapse into doing something internally just so that people get their h…”
Sands: Nearly 3,000 Stripe Employees Use Internal LLM Explorer Weekly
“The weekly active user count of this LLM Explorer is still at almost 3000, which is just shy of half the company using it every single week.”
Sands: Stripe Funds Small Accelerators for Six-Month AI Bets
“So the idea of accelerators is basically ring fencing one to two pizza teams and multiple of them to get new AI bets seated. And one of the accelerators was actually what produced this LLM Explorer. So it's very hard to just pull three engineers off of You kno…”
Sands: Stripe Radar and Sigma Assistants Are Rolling Out Soon
“So on automating code Radar Assistant and Sigma Assistant are two new products that are in beta and rolling out to all users soon.”
Sands: Financial Foundation Models Could Beat Traditional Optimization by 200 Bps
“It doesn't feel crazy to think that a good foundation model could outperform more traditional approaches by, I don't know, a hundred bips, 200 bips.”
Sands: Stripe Operates Four Dedicated AI Incubator Teams
“It's four of them today. And should it be six or should it be eight or should it be 10? And then in parallel, where can we really support the vertical teams or the core product organization in adopting LLMs or generative AI more broadly directly?”
Sands: Stripe has 60 applications built on its internal LLM API
“There are 60 applications built on that now, a bunch internal, but also several external, and I'm happy to talk about a couple of them.”
Sands: Stripe Built Internal Experimentation Platform Due to Latency Demands
“Our experimentation platform is one that we've built internally. We run a lot of charge level experiments and Latency and reliability requirements for charge level experiments are very, very high, and so building and running that internally has been worthwhile”
Sands: Merchant Identity and Regulatory Mapping Are Major Fintech Opportunities
“I think that, that identity piece, like who is this merchant? Are they who they say they are? But also what are they, what's their business? What are they selling? And how does that map to This pretty complicated regulatory environment is a really interesting …”
Sands: LLMs Will Automate Bespoke Financial Integrations Without Payments Engineers
“I think there's almost certainly an opportunity to, you know, whether Stripe does it or somebody else does it, to make sort of financial integrations way more seamless. Stripe has a whole suite of no code products, so you can use, ah, you know, payment links o…”
Sands: Stripe ML Recovers Billions Globally in False Declines
“We use ML to optimize authorization requests for issuers, basically Identifying the optimized retry messaging and routing combinations to recover a big chunk of false declines, about 10%, so billions of dollars globally.”
Sands: Stripe Smart Dunning Reduces Recurring Charge Declines by 30%
“For recurring charges in our billing product, we use Smart Dunning to reduce declines, and actually reduces declines by about 30%.”
Sands: Stripe Radar Evaluates 1,000 Characteristics in Under 100 Milliseconds
“Stripe Radar, which you mentioned, you know, considers a thousand characteristics of a transaction and figures out in less than a hundred milliseconds if each of the, you know, billions of legitimate payments made on Stripe can go through.”
Sands: High Upfront Compute Costs Force AI Startups to Monetize Faster
“Unlike a bunch of the past generations of software startups, we're seeing AI startups have substantial compute costs right out of the gate, and that that's putting a bunch of pressure to build monetization engines faster.”