⚡ Building ChatGPT for Finance — Joseph Kim, OpenAI's Head of Financial Services
October 31, 2026
AI Summary
5 min readBuilding ChatGPT for Finance: A Practical Guide from OpenAI's Joseph Kim
In a recent conversation at AIE Finance, Joseph Kim, OpenAI's Head of Financial Services, revealed a striking finding: when OpenAI tested MCP connections from major data providers, they found error rates exceeding 40%. This discovery drove a counterintuitive product decision — bundling proprietary data directly into ChatGPT for Financial Services, even though on first principles it looked like customers would feel they were paying twice for data they already licensed. "On paper it looks bad," Kim said. But customers loved it. "The number one thing they call out is the data."
The 80/20 Rule: Building on Top of ChatGPT
Kim's team built ChatGPT for Financial Services as a first-party OpenAI product, but he emphasizes that 95% of what they built is something any startup could replicate. "We didn't really have to go leverage our ability to go make changes in the model to go about building fantastic product," he said. The core insight: ChatGPT gets you about 80% of the way to a successful financial product, but the remaining 20% is decisive. Without it, "if you just took ChatGPT and launched it as a finance product, it would fail."
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Latent Space: The AI Engineer Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:05) **Opening & Keynote Teaser** - Joseph Kim introduces the new ChatGPT for Financial Services product and previews his keynote’s forward-looking predictions for model capabilities through 2027.
- 2 (01:49) **Research-to-Product Pipeline** - Kim explains how financial training data feeds into the main model run, making intelligence accessible to startups and firms alike.
- 3 (03:38) **The Codex Adoption Curve: Why Product Matters** - A chart shows finance adoption of Codex jumping from near-zero to 90% after the desktop app launch, illustrating the power of co-developing model and product.
- 4 (04:49) **Building for Enterprise Reality vs. First Principles** - Kim shares a key product lesson: ignoring industry-specific forces leads to building the wrong thing, even if it’s logically defensible.
- 5 (07:06) **95% of the Product Is Startup-Buildable** - Kim reveals that the vast majority of the finance product was built on standard ChatGPT plugin primitives, not model-level changes.
- 6 (09:22) **Safety Architecture: Humans in the Loop Today, Co-Worker Analogies Tomorrow** - The current product assumes a human-in-the-loop, but future autonomy will be modeled on how humans work together.
- 7 (12:36) **The "One Mistake" Myth vs. Real Model Gaps** - Kim addresses the fear that a single error could end a career, arguing that models are rapidly improving on factual accuracy but still lack human taste and judgment.
+ Full timestamped outline available in the app
Show Notes
Watch the full show on YouTube: https://www.youtube.com/watch?v=v-d1WTqIULs