Every fund can build their own software

The venture industry has traditionally been underserved by quality software to assist with fund operations, for a variety of reasons. Too small, too varied, too specific, for years it has generally been outside the scope of the revolution in software to run the operations stack of modern companies. But AI has changed that.

I've spoken to countless people at funds that are working on building apps to better their manual workflows, ranging from inbound email triage, deck analysis, company research and due diligence, financial analysis, portfolio company tracking, LP communication, collaborative CRM and founder communications, and more. So many things to improve with their daily workflows, so much unstructured data, so much opportunity to do something with it with AI.

Why now?

The opportunity to do it now is better than ever:

• LLMs are actually good at this. Modern AI models can read a 40-page due diligence memo and extract a structured summary in seconds. They can compare that memo against your current portfolio to find thematic overlaps. They can expand the range of metrics and KPIs you can track on a regular basis. They can flag when a portfolio company's quarterly update contains language that suggests trouble, even if the numbers look fine.
• The cost of not doing it is rising. Fund II is harder to raise when you can't demonstrate systematic learning from Fund I. LPs are asking sharper questions about process. And as fund lifespans stretch out, institutional knowledge that lives in people's heads becomes a liability.
• It's a real competitive edge, for now. Most funds are not doing this well yet. The ones that build structured data practices now could have a meaningful advantage in sourcing, portfolio monitoring, and LP storytelling within the next few years.
• Pattern recognition across deals. If every investment memo is a structured document with consistent fields — sector, stage, thesis, key risks, lead partner — you can start asking real questions. Which types of companies get to the third round? Where do your thesis bets cluster? What risks appear most often in deals that underperform?
• Faster, better LP reporting. Instead of rebuilding context from scratch every quarter, you pull from a living document of portfolio events, milestones, and narratives. AI can draft the LP update section for each company in minutes, grounded in actual data.
• Institutional memory that survives turnover. When a partner leaves or a new associate joins, they can query the fund's history. "What's our thesis on vertical SaaS?" shouldn't require a two-hour conversation — it should be searchable.
• Better inputs into your models. The real payoff is when unstructured data starts feeding structured models. Graduation rates, holding periods, exit timing: these assumptions should be grounded in what actually happened in your portfolio, not just market benchmarks. That requires structured data.

A practical starting point

You don't need to build a data infrastructure to start. Here's a minimal viable approach:

1. Pick one part of the operations stack to focus on first.
Do you want to start with inbound deal analysis, or pre-deal due diligence, portfolio KPI reporting, or LP letters? Pick one part of the operations stack to start testing with.

2. Pick one document type and standardize it.
Start with investment memos or portfolio updates — whichever you have more of. Define a template with 8–12 structured fields. Use AI to backfill your existing documents against that template.

3. Build the extraction once, run it every time.
Write a prompt that reliably extracts the fields you care about from a raw document. Test it on 10–15 of your documents. Once it's reliable, it takes minutes to run on everything new.

4. Put the structured output somewhere queryable.
Even a well-organized spreadsheet is a significant upgrade. A simple Airtable or Notion database is better. The goal is: when you want to answer a question about your portfolio, you don't have to dig through files.

How I started

I started with portfolio KPI reporting with Portfolio Reporting, a source-available platform for running a major component of fund operations around portfolio tracking and reporting, and built on LP aggregation and LP letters, have continued exploring additional tools and talking to GPs about what they are looking to build.

At the same time, I'm putting together a workshop on How to build fund infrastructure with AI because I am interested in diving into creating processes to identify, design, and deploy solutions to improve fund operations. If you are in fund operations or investing and you are interested in building and using these tools, not just understand it conceptually, I hope to see you there.