Framework for Vertical AI

How I invest in Vertical AI

Over the past few weeks, the job of a private investor has become increasingly hard. Having seen dozens of companies, I toy with various frameworks to help me more easily filter companies that come through the pipeline. Over the past few weeks, I keep thinking about one factor: disintermediation.

My mental model as an investor is quite simple, especially given that growth-stage investing is consensus investing. In publics, on the other hand, investors are forced to be contrarian (otherwise everything is already priced in). My mental model for investing boils down to two traits: unique distribution and product.

However, the competitive risk is becoming more apparent each day. For example, Perplexity released Comet yesterday, and a few hours later, OpenAI said that they’re working on an agentic browser too. Another enterprise browser, Island, which previously raised a $5 billion Series E financing earlier this year, will likely be disintermediated as well.

OpenAI and Anthropic, especially Anthropic given that 70-80% of their revenue come from APIs, feel like the near-term (and maybe long-term) winners in the AI race. As the world becomes increasingly agentic, the largest beneificiaries will be a select few model companies. What is even scarier is that model companies are becoming increasingly application-layer focused (OpenAI with Operator/Deep Research/Codex and now their Browser). In my view, this is where the trillion-dollar outcomes lie, and it appears that they can easily disintermediate many application-layer unicorns extremely quickly.

As I think about where I want to make bets as an investor, it has become clear that horizontal will be won by model companies, while vertical is still an open playing field with white space to be captured. Last fall in 2024, I made a call where I believe certain winners will emerge in the vertical AI space. And it did: Harvey and Abridge. One covers legal, and the other is healthcare. As I think about a framework to evaluate future vertical AI businesses, I created a simple framework to identify winning companies in the vertical AI space.

Here are 5 key frameworks I consider when investing in vertical AI businesses:

1. Underserved Market with Huge TAM: I believe that companies historically built in an underserved market have the potential to succeed. This is akin to blue-ocean strategies, where companies want to compete in a vast, but nascent market. A good way to tell if it is underserved is by 1) asking yourself how sexy this space is and 2) seeing how many billion-dollar vertical SaaS companies exist in this industry. Freight and construction, for example, are great industries where there are only $3 and $6 billion unicorns in the space, respectively. At its core, less competition equates to a higher likelihood of success.

2. System of Intelligence: The next generation of businesses will be systems of intelligence. This can be defined as software that orchestrates actions across a fragmented system of records. Here’s a helpful mind exercise. Companies have a bunch of structured and unstructured data. Think emails, pictures, scans, Excel sheets, OneDrive, Salesforce, Google Drive, etc. A system of intelligence, as I like to call it, is this layer of software you overlay on this bedrock of data for agentic workflows. The company should be able to win one wedge or workflow (such as agentic scheduling) and expand to other agentic offerings. As they win, they also become the de facto system of record, truly becoming a sticky offering that can beat incumbents.

3. Proprietary Data: The company needs to be able to access and use proprietary data (ideally unstructured data where general LLMs can’t easily parse). An example of proprietary data access is Harvey’s alliance with LexisNexis. The company must also be able to develop and fine-tune in-house models, which should improve with usage-driven network effects. This means that as the company incurs more usage, these models become better over time, beating out generic LLMs that will never win in this vertical. It’s even better if there is some form of patented IP (like a patented RAG system) incorporated in the business.

4. Distribution and Mindshare: This one is simple, as I’ve harped on the topic of distribution on this Substack. A company must be able to capture positive customer mindshare and user love before incumbents can innovate. Examples of this strategy are open source companies like n8n, where they also have a robust community and content strategy across YouTube, Reddit and Discord. This can also be seen in an influencer-led founding team and/or where the startup has figured out a form of viral k-factor.

5. Founder Experience: This is perhaps one of the most important traits in investing. Does this founder have the necessary experience to turn this from a Series C business to a company that will go public? And you develop this through pure taste. If you’re not impressed by the first impression, then neither will the market.

Moats, moats, moats and there are so many more traits that companies can incorporate to be more defensible. See here for Elad’s piece that I recently read today.

Again, this is a raw entry with my thoughts on the vertical AI space. My goal is to share more thoughts about investing and startups in the coming weeks.