Automating Monthly Portfolio Reports for a VC Fund

The Problem

A venture capital fund based in Taiwan manages a diverse portfolio of companies across multiple sectors and stages. Every month, the finance team is responsible for producing detailed reports for each portfolio company. These reports cover financial performance, key metrics, market developments, and strategic commentary. They need to match a very specific format and tone that the fund has established over years of investor communications.

The manual process was brutal. Finance team members would gather data from spreadsheets, CRM entries, portfolio company updates, and market research. Then they would write each report by hand, carefully matching the fund's established style. With a growing portfolio, the team was spending hundreds of man-hours per month on report writing alone. That left precious little time for the actual financial analysis and strategic thinking that should be the core of their work.

The fund tried templating and partial automation before, but the reports required too much contextual understanding and stylistic nuance for simple mail-merge approaches. Each company's narrative was different, the tone needed to be consistent but not robotic, and the format had to be precise enough that investors would not notice a change in process.

The Solution

BetterBrain built a report generation system that learns the fund's format and tone from its archive of previous reports. The system ingests fresh data about each portfolio company, including financials, milestones, and market context, then generates complete monthly reports that match the fund's established style.

The technical architecture uses RAG to pull relevant historical reports as style and format references. The system combines multiple language models for different parts of the task: OpenAI's gpt-4o and o1 for nuanced writing and reasoning, and Groq running llama-3.1-70B for faster processing of structured data sections. Agentic loops with self-learning allow the system to iterate on its own output, comparing drafts against previous reports to ensure consistency.

The finance team reviews and approves each report before it goes out, but the heavy lifting of data gathering, narrative construction, and formatting is handled automatically. The system also learns from edits the team makes, continuously refining its understanding of what a good report looks like for this specific fund.

The Number

• Hundreds of hours saved per month on report writing
• Generated reports match the fund's established format and tone
• Finance team freed for analysis and strategic work
• System continuously improves from editorial feedback
• Already transforming the monthly workflow with further improvements in progress