4 investments in apps enabled by LLMs

I focus on investing in products using natural language processing and LLMs at the app layer. I focus on business applications today, and my founder and investor experience spans B2B, marketplaces, “prosumer,” online learning, developer tooling, and infrastructure. I started investing in companies leveraging NLP in 2017 and have invested in six such startups so far.

I invest pre-seed through series A, always leading.

How I got into it

One of the reasons I got so excited about this space was my 2019 investment in a company called Keeper. They automatically find tax refunds for self-employed workers, using NLP to decode bank statements. When we made the investment, I wrote about why.

Only I had under-estimated the power of the technology. Within the first year this company had automated 90% of work previously done by humans, and that number has only gone up since. I dug deeper into how this was possible. Having that front row seat led me to the world of LLMs…

Current hypecycle

Regarding the current LLM craze, I believe there are parallels to the ‘90s Internet era (and the ensuing crash), but that doesn’t deter me. People who in the ‘90s claimed the internet was going to be huge could not possibly have been more correct. Yet, during the dot-com crash, they were laughed at. We’re going to go through a similar thing with generative AI over the coming decade. Many AI companies are raising over-priced rounds on hype over substance, and they will produce serious losses for many investors. Yet a small number of founders will create transformational businesses by focusing on fundamentals that matter in every market: making a product people love to use. This dynamic will require patience, and, of course, backing persistent founders with resilient businesses.

Why the app-layer

  1. It’s what I have an intuition for and am most qualified to evaluate.

  2. The most exciting (to me) startup AI products are where the product / UI layer are the innovation. You don’t necessarily need to be doing something super clever with the model. A user can't use a model. A user uses an interface. But you have to nail that part.

4 examples

The majority of startups I see are sprinkling in AI features, but here are 4 examples of apps not really possible prior to NLP of today’s caliber:

Keeper. Uses NLP including LLMs to automatically read bank and credit card statements, match transaction categories to job types and tax law, and ultimately detect write-offs for freelancers. This 2019 seed bet really opened my eyes to the power of the technology and contributed to my decision to focus on it. (Was also one of the earliest OpenAI customers to fine-tune GPT-3.)

Basedash. Uses LLMs to understand databases, specifically to infer the meaning behind column names. With this understanding they are able to deliver a “self-configuring” admin panel for startups. A developer needs only to connect a data source, freeing up their time from internal tooling so they can re-allocate it toward user-facing features.

Fini. Uses LLMs to turn any FAQ or knowledge base into a customized AI customer service chatbot in 2 minutes. Many are working to automate support, but Fini makes it extremely easy for any company to actually see the results instantly without spending weeks training a model or giving away sensitive data. If your FAQ is on a public URL, you can see Fini’s results and plug it in right away.

Outset. Uses LLMs to automate market research interviews and their subsequent synthesis. Today researchers interview subjects and dissect transcripts manually, so the size of their studies is always limited by labor hours and budget. Outset is an example of vertical SaaS for a specific industry and job type that was clearly not possible or particularly useful to build prior to LLMs.

My favorite metric

Retention > growth. Most founders over-invest in new user growth before they have cracked retention.

How I help

  • Recruiting. I source and interview candidates. More often though, after the sourcing and interviewing is done, founders ask me to step in to help sell candidates on joining their particular startup over other attractive options they may have.

  • Retention. I help you workshop ICP, value prop, churn reasons, low-hanging fruit, user segmentation, to help you flatten out those cohort retention curves. Once your users are sticky, I can certainly help you think through growth frameworks and tactics.

  • Raising. Whether it’s your current round or your next, I’ll help you get funded. All of my portfolio companies that have wanted to raise more have gone onto do so at significantly higher valuations following my investment. None have failed.

[1] My bet is on Humanloop which I also invested in.