Where LLMs meet VCs and how budding their relationships can be

Key finding: AI set to boost seed funding and diversity in coming years
Barbara Krassner
🇬🇪 Uniborn Team
3 min read

The AI industry is having its "iPhone moment." Particularly popular are LLMs — large language models, a subset of generative AI that works with human language — text data. 

And just the other day the sensational LLM family "ChatGPT, Bing, Bard and co" was joined by a newcomer, Claude-2, by Anthropic. US and UK residents can already load Claude with their tasks — and by the way, this bot can read and comment on a whole book in less than a minute (his context window is about 75,000 words). By comparison, ChatGPT's ceiling is around 3,000 words.

So what does all this have to do with the venture industry? We at Uniborn think our blog subscribers already know where we are going with this. But, we will try to give others some food for thought.

Here are some clingy numbers

We can agree that more than half the cost of VCs is sourcing and screening. However, sourcing is inefficient, and screening is biased and additionally inefficient.

LLM-driven VCs can quickly and accurately analyze huge amounts of data, including unstructured data: not just financial statements but also social media posts, news articles, and customer reviews. As a result, they can spot trends, anticipate risks, and make the best investment decisions.

Specifically, LLMs are already helping investors with various tasks: finding promising teams and conducting due diligence, adjusting portfolios, hiring the best people, and so on.

So what are the figures?

  • As little as 2% of venture funds generate a whole 95% of the industry's total profits, according to Andre Retterath, partner at Earlybird Venture Capital. To become one of these leaders, investors need to reduce the percentage of lost deals. This means reducing reliance on traditional processes and increasing data-driven approaches.
  • Andre's same report shows that ML models outperform human investors in deal screening, achieving similar accuracy and recall rates as the best investors (80%) while surpassing the average/median investor by 40%/33%, respectively.
  • Leveraging LLM reduces time-on-task by 0.8 standard deviations (SDs), improves product quality by 0.4 SDs, and reduces the gap between high and low-ability teams. This was found by MIT analysts who conducted research on the use of ChatGPT.

From Andre's other work, we can highlight one more good point: "VCs are always busy, distracted, and unable to focus on actual value drivers. Leveraging data-driven approaches and AI dramatically increases the input-to-output ratio in the VC investment process and, consequently, releases time for VCs to focus on the right opportunities at the right point in time. Let computers process the data so that humans can spend more time together."

A quick look at today's AI&VC duo

The venture capital industry is one of the least digitized in the world. While 84% of all global VCs want to become more data-driven, only 1% emphasize data and technology as a critical part of their strategy, which includes adequate resource allocation and building dedicated teams. The latest Data-driven VC Landscape 2023 report revealed this gap between intent and reality. And by its count, only 151 VCs worldwide can be considered truly data-driven.

This 1% includes younger and smaller companies. While the old-schoolers rely on good old-fashioned manual labor, the new generation uses more modern and efficient tools across the value chain than their older and larger counterparts.

Which tools which VCs prefer. (Image: DDVC)
Which tools which VCs prefer. (Image: DDVC)

And even among the data-driven 1%, only some are willing to share their secret sauce with the public. The exception to the rule is companies like Moonfire — a British early-stage VC that openly writes about using text models to locate companies and founders that fit its philosophy — or EQT — a Swedish-born global investor that just recently shared in a podcast about how it incorporates data usage and AI in its work.  

The lack of transparency in the use of AI in venture capital is a serious problem because the power of these technologies lies in the data which needs to be shared. Otherwise, the output of the neural network will not be accurate enough. To paraphrase the authors of the Data-driven VC Landscape 2023 report, a largely closed-source industry is starving the open-source approach.

There is another rub: After reading inspiring stories, many foundations rush to digitize and collect data. They quickly burn out and give up after a few months of playing the game. The key advice here is to take it one step at a time. 

Evaluate your daily workflow — what is automated and what is not? As Ties Boukema of Dawn Capital aptly pointed out, "Many firms talking about automation, whilst still ping-ponging to schedule a meeting and doing all reporting in Excel. If you want a flying car, first upgrade your horse." Then you should test LLMs, at least in free interfaces, and then try to connect them to your enterprise services via APIs and customize them.

A must-do list for tomorrow

By 2025, AI and data analytics will take over startup vetting. Gartner predicts that more than 75% of reviews will be influenced by AI, meaning that AI could determine whether a company even reaches the human hands. 

And that's just primary screening. Pitches, rounds, incubators — all elements of the venture industry are rich in tweets, letters, and publications. If we collect all these words (most of them anyway), investors can gain magic powers by going into the LLM app and typing up, "Hey, I want to raise a $200 million fund in a couple of months to back immigrant founders in the Nordics who build deeptech startups. Who better to call?" — and in a split second, it will generate prewritten letters and a list of potential LPs.

This is not to say that everyone can become an investor or that AI networks will completely replace them. But yes, the routine will go away, and the entry threshold will go down significantly — and that's an inspiring thing.

For this to happen, there are two important things needed. The first is to collect and disclose data, including online. So far, the public ChatGPT does not know anything before 2021, and Claude 2 is trained on data not fresher than December 2022. The second is to recruit a team capable of not only developing and teaching LLM but also creating the most convenient and simple user interfaces.

As James Currier, GP at venture capital firm NFX, has rightly pointed out that we are experiencing the last 10 years of venture capital as we know it. So when the above-mentioned factors converge at last, the venture will begin to evolve exponentially.

For founders, the situation will change in such a way that disruptors and unicorns will be made by the hands of several people and a bunch of smart programs (do you remember that pitch decks from GPT-4 have already proven to be twice as convincing as human-made ones?).

"AI will level the playing field and get rid of the only-top-tier approach. The less burdened founders are, the smaller checks they can accept from investors, and the more diverse people will ultimately be able to join the ride and drive value. VC will be more inclusive, balanced, and meritocratic," says Dmitry Samoilovskikh, founder and CEO of Uniborn.

VCs will have to seek alpha in amplifying portfolio companies than in the ability to pick winners
Dmitry Samoylovskikh, founder and CEO of Uniborn

Cover image: Unsplash

Share this article

Uniborn is a community-led platform for sourcing, sharing, syndicating, and amplifying startups.

Uniborn is a trade name of Uniborn OÜ, registry code 16515953, Rotermanni 6, Tallinn, 10111, Estonia.

We are not intended to be a substitute for legal, tax or financial advice.

Terms and Conditions
Privacy Policy
Risk Disclosure