Building AI Systems for Capital Markets

August 24, 2026

AI Summary

5 min read

Chris Churchman, head of the Marquee platform at Goldman Sachs, has spent the last four years building AI products for capital markets. His central insight is that the hardest part of the job is not the flashy demo—it is the "boring, hard, actually pretty thankless work" of making a system reliable enough that a client or a trader can trust it on its worst day. In a conversation with Alison Nathan and George Lee, Churchman walks through the practical lessons learned from building in an environment where the foundational technology shifts every few months, and where the stakes for accuracy are extremely high.

The hallucination problem is structural, not fixable

Churchman is blunt about the core difficulty: large language models cannot distinguish between a fact and an interpolation. "Everything goes into the context window, it goes through the same sausage machine, comes out the other end, and it really just doesn't know." The problem is baked into the training objective. As Churchman explains, there is a famous paper on why LLMs hallucinate: "They're doing multiple choice tests and they're not rewarded for abstaining. They might as well guess because that is optimal."

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

Never miss an episode of Exchanges

Get every new episode summarized in your inbox — free, ~5 minutes to read.

No spam. Unsubscribe anytime.

What you'll learn

1 (02:04) **Marquee’s Mission & the AI Concierge** - Chris Churchman introduces Marquee as the firm’s digital platform for institutional clients, explaining that AI is layered in to serve the client’s investment process, not the other way around.
2 (04:10) **Marquee AI in Action: From Intent to Grounded Answer** - A walkthrough of the system’s internal prototype, showing how natural language input triggers a multi-step retrieval and analysis pipeline.
3 (05:28) **The Hallucination Problem: Why Plausible Isn’t Good Enough** - Chris explains the fundamental challenge of factuality in institutional AI, where a confident but wrong answer is worse than no answer.
4 (06:26) **The Frustrating Pace of Change: Why You Can’t Bet Against the Model** - Chris recounts how rapid model improvements (context windows growing from 4,000 to 1 million tokens) rendered months of engineering work obsolete.
5 (09:50) **The Evolution of Engineering: From Prompts to Mandates** - Chris maps the progression of AI development eras, concluding that the next frontier is legal and liability frameworks for autonomous agents.
6 (11:49) **The Demo vs. The Product: Building for the Model at Launch** - Chris’s key lesson on avoiding wasted engineering effort in a fast-moving field.
7 (14:06) **First Principles of UX: Trust, Transparency, and Human Reasoning** - Chris explains how the user experience design must prevent “cognitive atrophy” and ensure human capital stands behind the AI’s output.

+ Full timestamped outline available in the app

Show Notes

Key takeaways:

The "demo versus product" gap:  An AI demo is judged on its best day, but an institutional-grade product is judged on its worst. Building reliable financial tools requires rigorous, auditable grounding frameworks to avoid hallucinations and ensure every output can be traced back to a verified source.
Don't bet against the model:  Developers should avoid wasting resources on complex workarounds for temporary constraints on models that are evolving rapidly. Instead, they should focus on elements that models cannot natively learn: details specific to the firm, such as proprietary data or how data sets are connected to each other.
Tackling legacy constraints: Successful innovation lies in redesigning workflows from first principles rather than simply automating legacy bottlenecks to do the same tasks faster.

In this episode of Goldman Sachs Exchanges, Chris Churchman, head of Marquee, Goldman Sachs’ digital platform for institutional and corporate clients, talks about building effective artificial intelligence (AI) products for institutional investors.

Churchman, who is also co-chair of the Global Banking & Markets AI Working Group, describes an ongoing shift from a world where users must learn software to one where software learns the user. He tells hosts Allison Nathan of Goldman Sachs Research and George Lee, co-head of the Goldman Sachs Global Institute, about the challenges of grounding generative AI in hard facts, and emphasizes the need to design systems that enhance human reasoning rather than outsourcing critical thinking to machines.

The opinions and views expressed herein are as of the date of publication, subject to change without notice, and may not necessarily reflect the institutional views of Goldman Sachs or its affiliates. The material provided is intended for informational purposes only, and does not constitute investment, legal, or tax advice, a recommendation from any Goldman Sachs entity to take any particular action or be used as a basis for any other investment decision, or an offer or solicitation to purchase or sell any securities or financial products. Any forward-looking statements, case studies, computations or examples set forth herein are for illustrative purposes only. Past performance is not indicative of future results. Neither Goldman Sachs nor any of its affiliates make any representations or warranties, express or implied, as to the accuracy or completeness of the statements or information contained herein and disclaim any liability whatsoever for reliance on such information for any purpose. Each name of a third-party organization mentioned is the property of the company to which it relates, is used here strictly for informational and identification purposes only and is not used to imply any sponsorship, affiliation, endorsement, ownership or license rights between any such company and Goldman Sachs. Th