Man Group's AlphaGPT

Hey, it's Matt. Here’s what’s up in AI + Wall Street.
Was this email forwarded to you? Sign up below:

IN THIS ISSUE: Man’s AlphaGPT, adoption at Morgan Stanley & Goldman, and is audio the next frontier for quants?

BABY ON BOARD 👶 

The AI Street family grew this week with the arrival of our second child. Unlike AI, us human writers do need sleep. I am going to try to keep roughly the same publishing cadence. Wish me luck!

HEDGE FUNDS

Man Group Builds AlphaGPT to “Disrupt” Quant Process: CTO

"The next level of thinking is how do we systemize how we build systematic strategy in the first place?"

Quantitative hedge funds use algorithms to test and deploy their profitable trading strategies. Now, hedge fund Man Group thinks LLMs can help create the strategies themselves.

The firm, which manages around $150 billion in assets, is developing a system called AlphaGPT that uses AI to automatically generate and test new investment ideas.

"Fundamentally, if we think, what does tech do? How does it add value? Well, we aspire to understand workflows, then automate those workflows," said Gary Collier, Chief Technology Officer at Man Group, on a recent Reuters’ podcast. "The next level of thinking is how do we systemize how we build systematic strategy in the first place?"

_{Collier discusses AlphaGPT around the 15:40-16:30 mark and 32:10.}

The goal of AlphaGPT is to automatically scan sources like research documents and academic papers, extract investment ideas, and quickly test them across various financial instruments. It seeks to automate the trial-and-error stage of quantitative strategy development that humans usually perform.

"We're aiming to fundamentally disrupt the creative quant process," Collier said. "It’s early days, but we're showing some significant promise already."

Man Group has been at the forefront of quantitative investing and AI technology for years, notably pioneering algorithmic trading through its systematic investing arm, AHL.

The hedge fund has also built ManGPT, which uses both internal and external LLMs, for coding, drafting documents, and customizing information flows for portfolio managers. It’s currently used by half the firm on a monthly basis, according to Collier.

One of the most interesting aspects of the current AI boom is that the technology doesn’t rely on anyone writing the rules or what to look for—it instead ‘learns’ directly from data. It’s already leading to medical breakthroughs. Is finance next?

VOICE

Are Speech Patterns the Next Frontier for Quants?

Did you know that most earnings transcriptions scrub filler words like, “umm,” “you know” and “like” from recordings? I didn’t.

I learned this from David Pope, whose startup Speech Craft Analytics mines vocal patterns for trading signals. Pope, who previously built quant products at S&P, says
earnings call audio is a new frontier for quantitative investors.

The model-driven investors have long used algorithms to parse text for trading clues. Pope sees audio analysis as the next logical step in this evolution.

Standard transcripts strip out “ums,” “ahs,” repetitions, and hesitations-- subtle but revealing vocal cues that can indicate nervousness, anxiety, or deception. Audio preserves these signals, along with tone, pacing, and other emotional markers that text doesn't capture. Speech Craft’s technology detects voice tremors at 22 microseconds beyond human perception.

"When you're nervous it's physiological, your body changes... One muscle that tenses is our vocal cords, and that changes the way that sound emanates," Pope tells me.

The startup runs nearly 40 cloud servers simultaneously to transcribe and analyze its own recordings to preserve all those "umms," "ahs," "you knows." Pope predicts analyzing audio will become as ubiquitous as mining text. 

RELATED:

• Analyzing Earnings Call Tone with AI 

• Huh? The valuable role of interjections 

• Facial, Vocal Cues of Central Bankers Can Impact Markets

BANKS

AI Helps Advisors, Doesn't Replace Them: Morgan Stanley

Morgan Stanley sees AI as a growth engine, not a cost-cutting tool, according to Sal Cucchiara, CIO of wealth management technology. "We're not looking at [AI] as a cost-reduction exercise," Cucchiara told ThinkAdvisor. "It's a productivity enhancer."

The bank has created:

• The Assistant: An AI chatbot providing access to 100,000+ internal documents (98% adoption rate).

• Debrief: Tool that records, summarizes client meetings, and drafts follow-up emails.

• Mobile Debrief: Coming this spring for in-person meetings.

Morgan Stanley is also building AI agents and tools using OpenAI's GPT-4 in a secure environment that keeps client information private and separate from OpenAI's training data, according to Cucchiara.

How Goldman Measures Its AI Rollout

In a recent interview with Fortune, Goldman Sachs CIO Marco Argenti said the bank is tracking AI adoption by looking at how frequently tools are used, how often AI-generated code is accepted, and whether employees are integrating AI into their daily work.

• AI tools now cover half of Goldman’s 46,000 employees. The rollout began with engineers about 25% of the firm’s staff, who received tools like GitHub Copilot and Gemini Code Assist. Usage metrics include frequency and code acceptance rates.

• GS AI Assistant, launched in 2024, is used by 10,000+ bankers, traders, and asset managers. It can summarize documents, draft emails, analyze data, and generate client-facing content. Full rollout is expected by the end of 2025.

• Goldman’s `agentic AI’ remains in testing. These tools could automate tasks like compliance checks and transaction processing, but Goldman is holding back until it can ensure high data quality, proper training, and appropriate risk controls. (Fortune)

FURTHER READING

Ex-Point72 Exec Launches Pay-Per-Use Data Platform

Former Point72 executive Kirk McKeown has launched Carbon Arc, an alternative data platform that charges by usage, instead of locking customers into fixed-price contracts.

By eliminating high upfront costs, Carbon Arc is betting on volume, offering cheaper data to more users rather than catering solely to the biggest money managers, a model similar to cloud-computing pricing. (Business Insider)

New AI Agent Automates Financial Risk Monitoring

AI technology Auquan has launched an autonomous AI agent to monitor financial risks for portfolio management. The tool monitors hundreds of companies across millions of global data sources, providing early warning detection of emerging non-financial risks while eliminating up to 80% of manual monitoring work, according to the company. (Press Release)

NatWest Strikes Deal With OpenAI

NatWest and OpenAI have partnered to enhance the lender's digital assistants and customer support processes using AI, in the first collaboration of its kind with a UK-headquartered bank. (Reuters)

ICYMI: MARKETS EDITION

Demoing Primer: AI Analysis in Your Inbox

Dividing companies into peer groups seems like it should be pretty easy. But there are some oddballs → Uber is an “industrial” company per the Global Industry Classification Standard, the industry benchmark, which was created by S&P back in 1999.

(One of the first companies I covered at Bloomberg News in the 2010s was S&P’s predecessor, conglomerate McGraw Hill, which funnily enough was classified as a media company because it owned BusinessWeek and other publications, though most of its profitability came from financial services).

Because of these weird classifications, analysts often create their own peer groups. With large language models, you can now make custom cohorts at scale.

Primer, a U.K.-based AI for equity research platform, recently used this ability to create an AI Disruption Report, which ranks 2,000 companies by their vulnerability to AI-driven market shifts.

While we focused on the AI Disruption report, Primer also analyzes companies' filings and earnings calls in real-time, automatically identifying meaningful changes in guidance, outlook, and competitor landscapes that human analysts might miss.

Rather than treating businesses as monoliths, Primer evaluated each segment individually, determining which revenue streams are at risk and where AI-driven margin pressure is most likely to emerge. The report is based on multiple passes of AI analysis, using models like Claude 3.5 Sonnet and OpenAI o1 to assess how AI could impact different divisions within a company.

Companies in professional services dominate the AI risk rankings—firms like Cognizant and Accenture, which have built their businesses on high-touch knowledge work that AI is now automating. Customer service outsourcing giant Teleperformance also stands out, as AI-powered chatbots and voice assistants take over roles traditionally performed by human agents.

For more on how Primer did the analysis see their white paper here.

In the video below, Liam Rogan, Primer’s Head of Growth, and Ruggero Gargiulo, Head of Research, demo the platform to show earnings analysis in real time, such as flagging material changes.

When JD Sports reported earnings, the company told investors its guidance was “unchanged.” But Primer’s AI spotted a subtle tweak: the guidance now included a £25 million contribution from an acquisition that hadn’t been factored in previously. That meant the company was effectively lowering expectations for its core business.

“Seven out of eight sell-side brokers expected the stock to be flat or up,” Ruggero said. “But shares opened down 3% and closed the day down 10% as the market realized what had actually happened.”

Primer also identified a read-across effect when Interpublic reported weak results, triggering a delayed reaction in WPP’s stock. Interpublic’s earnings signaled broader softness in ad spending, a risk that extended beyond just one company. Primer flagged the update not only to clients following Interpublic but also to WPP investors who weren’t tracking the company directly.

“Some of our clients who weren’t even following Interpublic received the report anyway,” Ruggero said. “Primer recognized that the company’s outlook had worsened, which could have systemic effects on its peers. Sure enough, WPP initially held steady, then dropped 2% later in the day and kept falling in the following weeks.”

Current Primer users are from Allianz, Jefferies, Blackrock, Fidelity, North Rock, PMI, Panmure, among others. The platform covers 2,000 companies across the US, UK and EU.

If you’re interested in Primer, check them out here.