FocusAlpha

Inspiration

• As an investor, I saw how powerful Claude, ChatGPT, and Codex could be for research—but also how limited they are without reliable data.
• Everyone is building investment agents, but most still rely on fragmented, unstructured, and noisy data sources.
• We started by building an agent ourselves and realized that with the right data, 80% of the problem was already solved.
• That led us to build FocusAlpha — Bloomberg for investment agents.

What it does

• FocusAlpha provides investment-ready data for AI agents.
• It helps investors monitor portfolios, screen for changes, and research ideas with better accuracy and less token usage.
• Our data covers financials, filings, transcripts, events, ETFs, ownership, and Asia supply chains.

How we built it

• Built proprietary data pipelines to crawl, clean, normalize, and structure financial data at scale.
• Developed entity resolution, relationships, and connections across companies, securities, events, funds, and supply chains.
• Made the data directly accessible to agents through API and MCP.

Challenges we ran into

• Institutional investors are still slow to adopt AI due to concerns around data privacy, client information, regulation, and lengthy compliance processes.
• Instead of asking customers to move sensitive data into our software, we shifted to an MCP and API-based solution that plugs into their existing AI workflows while we focus purely on data.
• We also learned that better models and better agent harnesses cannot solve a data problem. Reliable coverage, entity resolution, freshness, and data quality remain the hard part.

Accomplishments that we're proud of

• Interviewed 100+ investors, wealth managers, and traders to understand how they are actually adopting AI.
• We are now working with five pilot funds, building directly alongside their investment and engineering teams.
• Built an AI-native financial data infrastructure from the ground up, turning hundreds of fragmented sources into structured, agent-ready datasets.

What we learned

• We initially targeted traditional hedge funds, but many face organizational, compliance, and workflow barriers to adopting AI.
• We found much stronger adoption among newer, AI-native funds—especially those with software engineers on the team. They can already build their own agent harnesses and understand that high-quality data, not another AI interface, is the critical missing layer.
• Our other early users include individual investors, global funds researching Asia, and family offices without Bloomberg or FactSet.
• We also learned that financial data is highly competitive. To build a real data moat, we need differentiated information rather than simply broader coverage.
• This led us to focus on Asia supply-chain intelligence, where valuable information remains fragmented across languages, markets, and local sources.

What's next for FocusAlpha

• Talk to customers: Interview at least 100 users in each target segment to determine where we create the most value—particularly wealth managers versus investment funds that need differentiated Asia data.
• Build an open-source portal: Make the software layer free and open source, while charging only for the proprietary data subscription.
• Deepen our Asia coverage: Continue building proprietary datasets around Asian companies, supply chains, capacity, competitors, and market-moving events.

Built With

• python
• supabase