Aakash GuptaHow to Build AI Products in FinTech ($100B Robinhood VP Lessons)
Aakash Gupta and Abhishek Fatehpuria on robinhood VP on AI fintech products, experimentation, and product velocity.
In this episode of Aakash Gupta, featuring Aakash Gupta and Abhishek Fatehpuria, How to Build AI Products in FinTech ($100B Robinhood VP Lessons) explores robinhood VP on AI fintech products, experimentation, and product velocity Robinhood Cortex is positioned as an AI investing assistant that fits existing workflows by explaining stock moves using curated, licensed data sources while avoiding direct recommendations to maintain trust and regulatory safety.
At a glance
WHAT IT’S REALLY ABOUT
Robinhood VP on AI fintech products, experimentation, and product velocity
- Robinhood Cortex is positioned as an AI investing assistant that fits existing workflows by explaining stock moves using curated, licensed data sources while avoiding direct recommendations to maintain trust and regulatory safety.
- Abhishek argues successful fintech product building requires deep regulatory fluency, strong cross-functional partnership (especially legal/compliance), and patience to ship incrementally as customers and regulators build confidence.
- Robinhood’s innovation DNA emphasizes delivering both customer value and a delightful, pixel-perfect experience, using “swipeys” (four-screen customer messaging) as a working-backwards artifact before building.
- IPO Access is described as a retail-demand aggregation product inside the IPO selling group, with product iterations focused on clear value messaging (“Get in at the IPO price”) and emotionally resonant UX moments.
- Robinhood’s product velocity is supported by GM-based org structure, keynote-driven planning, extensive experimentation (e.g., referral program iterations), and internal dogfooding with a high bar for polish.
IDEAS WORTH REMEMBERING
7 ideasStart AI by improving an existing user workflow, not by chasing “AI features.”
Cortex began with a universal workflow moment—users see a 5% move alert and ask “why?”—so the product compresses research steps rather than inventing a new behavior.
In fintech AI, upstream data curation is a core product decision.
Robinhood emphasizes licensed, curated inputs (news providers, research reports, exchange market data, SEC filings) to reduce hallucinations and increase trustworthiness.
Sequence riskier AI capabilities only after trust and infrastructure exist.
They explicitly avoid recommendations today because advisory requires portfolio context and a much higher compliance bar, choosing “informational tool first” to build confidence.
Make legal/compliance a first-class product partner, not a blocker.
Abhishek recommends assuming good intent, selling the product vision to legal like you would to engineers/designers, and deeply understanding the rule behind each concern to find workable solutions in gray areas.
Use a crisp customer message artifact (“swipeys”) to force clarity early.
Writing the 3–4 swipe screens before building pressures teams to define value in simple language; if you can’t earn a “Get Started” in one sentence, the product isn’t ready.
Differentiate within regulatory constraints by picking ‘3–4 things we do infinitely better.’
Since many fintech products (e.g., IRAs) are legislated and hard to reinvent, Robinhood focuses on experience, speed, and emotional resonance rather than changing the underlying rules.
High-velocity experimentation requires iterating past the obvious playbook.
The referral program evolved from fixed cash → variable cash → variable stock, then improved activation by making users claim the reward and reinforcing ownership via later notifications.
WORDS WORTH SAVING
5 quotes“We don’t want to build AI products for the sake of building AI products. We want it to fit into problems we know customers already have.”
— Abhishek Fatehpuria
“We curate almost all of the data that’s going in… and coach it to not make mistakes and to not make recommendations.”
— Abhishek Fatehpuria
“If you can’t convince a customer to hit the Get Started button in, like, one sentence, we don’t have a great product.”
— Abhishek Fatehpuria
“You shouldn’t use bad design as a way to keep people out.”
— Abhishek Fatehpuria
“The goal is the goal.”
— Abhishek Fatehpuria
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsFor Cortex, what specific evaluation methods or monitoring do you use to detect factual errors or “unsafe” outputs in production, and how do you decide acceptable thresholds?
Robinhood Cortex is positioned as an AI investing assistant that fits existing workflows by explaining stock moves using curated, licensed data sources while avoiding direct recommendations to maintain trust and regulatory safety.
How did you operationally implement the “no recommendations” constraint—prompting only, system rules, UI disclosures, or hard filters on certain intents (e.g., ‘should I buy’)?
Abhishek argues successful fintech product building requires deep regulatory fluency, strong cross-functional partnership (especially legal/compliance), and patience to ship incrementally as customers and regulators build confidence.
What did the curated-data pipeline look like (sources, refresh cadence, provenance, citation strategy), and which part was hardest to get compliant?
Robinhood’s innovation DNA emphasizes delivering both customer value and a delightful, pixel-perfect experience, using “swipeys” (four-screen customer messaging) as a working-backwards artifact before building.
On IPO Access, what allocation policy do you use when demand exceeds shares—pro-rata, lottery, capped orders—and how do you minimize perceived unfairness?
IPO Access is described as a retail-demand aggregation product inside the IPO selling group, with product iterations focused on clear value messaging (“Get in at the IPO price”) and emotionally resonant UX moments.
You mentioned ‘swipeys’ as a forcing function: can you share an example where the swipeys exposed a weak value prop and caused the team to pivot or kill the idea?
Robinhood’s product velocity is supported by GM-based org structure, keynote-driven planning, extensive experimentation (e.g., referral program iterations), and internal dogfooding with a high bar for polish.
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome