Sarah Guo made a name for herself as Greylock‘s youngest and only female general partner.
But after nine years at the firm, she decided to strike it out on her own last summer and launched a fund called Conviction Partners. It didn’t take long to convince LPs to get behind Guo’s investment strategy. Despite the difficult fundraising environment, it took her less than 10 weeks to raise a $101 million fund.
Guo had a strong record at Greylock, where she backed companies like family-benefits platform Cleo. She also chose to make AI—an area that captured investors’ imaginations with the introduction of DALL-E—the focus of her fund.
Guo announced her new fund’s first investment last month. Together with Battery Ventures, Conviction led a $7.5 million seed deal in Seek AI, a company that uses generative AI to help business users retrieve data from company databases without the help of data scientists.
PitchBook spoke with Guo about what attracted her to AI investing, what kinds of companies are garnering the highest valuations and the defensibility of newly formed generative AI startups. This interview has been shortened and edited for clarity.
PitchBook: It sounds like you planned to start an AI-focused fund just before generative AI blew up in earnest. Did you feel that a big breakthrough in AI was coming?
Guo: When I was at Greylock, the other partners and I had been investing in AI for years. AI was becoming more effective over time. We saw that machine learning was crucial to Google for its ads ranking, and on-demand companies used machine learning for pricing and driver matching. So, we started to invest in companies that leveraged AI as a tool in some way. These companies were in areas like security analytics or contact center automation.
I started writing about machine learning applications in 2016. And then transformers happened, and it was obvious transformers as a new AI architecture were really worth paying attention to.
But most people didn’t really notice the evolution in AI until generative capabilities like Dall-E and ChatGPT came about.
You must be thrilled that you have a whole fund dedicated to an area that is going through such tremendous technological growth.
Yes. But at the same time, it’s not that great for me as a fund. Having such an aggressive hype cycle around AI is not ideal.
Valuations of generative AI companies are generally much higher than for other startups. Do you think the prices are warranted?
I think there’s a lot of willingness from the broader venture community to pay up for companies started by researchers that come from a few labs. That’s probably because a lot of venture investors don’t feel comfortable making a call about what types of companies are going to be durable and defensible in this market. Instead, people are really leaning into the founder pedigree.
I don’t think that is always right, but I have noticed that the most aggressively priced deals are for companies founded by people who worked as research scientists at OpenAI or [Alphabet‘s] DeepMind because it’s where the state of the art is now.
Do you believe top researchers make good founders? They can build an amazing model, but can they turn that model into a great application that people want to use?
It’s true that a model is very different from having something that works in the real world that is valuable.
If I could change one thing about the ecosystem, it would have to be investors’ focus on founders who are thinking about solving a customer problem instead of just the AI model exclusively.
We’re excited to fund core research, but the majority of our portfolio is going to be in applied companies where we understand the market deeply and we can see how the product is going to offer customer value.
I understand that besides high valuations, there may be many similar companies being built with generative AI capabilities. How do you make sure that you’re investing in something that’s not easily copied?
There are many ways to build defensible companies.
I think there’s a whole cohort of companies right now that are doing some lightly verticalized applications on top of Anthropic or Open AI’s APIs [application programming interfaces]. Those companies are probably not super defensible.
But I don’t think investing in generative AI is actually that different from investing in the prior generation of software.
We are investing in workflow companies like Seek. Its models provide users with answers they can trust for making important business decisions. That’s not different from what made certain SaaS companies win.
It’s interesting that you say that users trust Seek’s answers. Obviously, one of the biggest criticisms of ChatGPT is that its responses are not factually correct. Is that a problem for other companies that are using generative AI?
Seek is not going to make up answers because it’s generating code for your company’s data only. There is no outside data that it relies on. Similarly, I think people will end up building products that make sure that the end user gets accurate answers.
Related read: VCs try to parse through the ‘noise’ of generative AI
Featured image by Chloe Ladwig/PitchBook News
Learn about our editorial standards.