How does AI make sourcing deals easier?

Less than a year into the AI craze, it’s clear that many venture investors are not only doggedly scouting for the most promising AI start-ups, but they are also increasing their own productivity by applying these tools to their internal processes.

Sourcing new deals tops the list of tasks VCs consider ripe for using AI. That may seem an obvious choice to help sift through reams of digital data on companies and markets. Perhaps less apparent is how effective these tools can be in ferreting out worthy companies that are building customer bases and generating revenue far removed from the awareness of human networks.

Headline, a San Francisco-based venture firm managing about $4 billion in assets, has been using machine learning, a rudimentary type of artificial intelligence, for about 20 years. The firm has created several proprietary data-driven machine learning platforms supported by a global team of 17 engineers, data scientists and technologists.

One of those platforms, called Searchlight, has become an integral part of Headline’s sourcing regime. To look beyond major tech hubs to find fast-growing companies, Searchlight scans millions of websites around the world. Machine learning enables Headline to cross-reference, rank and filter companies based on proprietary algorithms that also provide “peer-to-peer benchmarks for analyzing growth and performance of early-stage companies across multiple sectors,” two Headline partners wrote in a recent blog post.

Searchlight led Headline to a little-known open-source password manager called Bitwarden in Florida. “The founder, Kyle Spearrin, was a one-person operation and by the time we got to him he had five of the top 50 banks as paying customers,” Thomas Gieselmann, a Headline founding partner, tells Venture Capital Journal. “NASA was a paying customer. Not a single angel investor, no VC in there… No VC could have helped this individual because he wasn’t in any VC’s network. The only way to find him was through data [by looking] at the exhaust of the pawprints his product was leaving all over the internet.”

Bitwarden is now one of the leading passport managers and last September received a $100 million minority growth investment led by PSG with participation from Battery Ventures. Headline invested in the company’s seed round in 2013 and participated in its Series A in November 2019.

Gieselmann views so-called data exhaust as an indication of how much traction a company has. “If something is successful, people start talking about it on Reddit, Twitter, certain blogs. We can measure the frequency of that. The number of people linking back to a particular company is interesting. Who’s linking to them has predictive power.”

He believes that data “truly does remove bias in that it allows us to talk to companies that have traction but otherwise don’t have access to [traditional VC] networks.”

Searchlight helped Headline identify Segment.com, a customer engagement platform that was snubbed by other VCs but turned into a big hit. After completing Y Combinator’s accelerator program and getting exposure to multiple VCs on Demo Day, Segment reworked its product four times before wearing out its welcome with investors.

“There are only so many meetings you’re willing to take with a founder before there’s diminishing marginal utility, even if you’re very patient,” says Gieselmann. Headline, in contrast, had neither attended Y Combinator’s Demo Day nor met with Segment’s founding team multiple times before its product reached its final iteration.

“As the company started to grow, we saw data exhaust that told us there’s something interesting there,” Gieselmann says. “We reached out and put the first $1 million of institutional money into the company” in 2013. Headline also participated in Segment’s $175 million Series D round in 2019. In late 2020, Segment was acquired by Twilio for $3.2 billion. Headline declined to reveal its return on investment.

Searchlight is also able to make fine distinctions between similar business models when scouring publicly available data. It led Headline to Poplin, previously known as SudShare, a door-to-door laundry service. Gieselmann calls it “the Uber of laundromats.” In online searches, it’s hard to differentiate that sort of service from a chain of physical laundromats. Searchlight “allows us to distinguish between this as a start-up [as opposed to] a chain of local laundromats, which is not as interesting to us as an asset-lite, high-growth laundry service,” he says. Headline led Poplin’s $10 million seed round in March 2022.

“Data truly does remove bias in that it allows us to talk to companies that have traction but otherwise don’t have access to [traditional VC] networks”

Thomas Gieselmann,
Headline

For the most part, Headline avoids early-stage rounds and invests only when it spots an inflection point for a company. That could be internet traffic showing growing recognition for a founder, as measured by links from important sources, which Gieselmann says are highly predictive of business growth.

Headline has also long been using AI to help it derive meaning from unstructured data. One example is routing start-ups to the right team member based on sector and geography so the best positioned investment professional can reach out.

In one of its works in progress, Headline is developing technology to look for patterns in unstructured data to identify promising companies anywhere in the world irrespective of human biases. Gieselmann likens the effort to how a chatbot finds patterns in unstructured data to predict the next word.

Narrowing searches

Like Headline, EQT Ventures of Sweden is a believer in data science. Its proprietary Motherbrain platform, launched in 2016, is now being used by investment teams across the larger EQT Group. Partner Ashley Lundström says Motherbrain is particularly good at interpreting alternative datasets that could be leading indicators for where to invest. EQT Group has established an entire team known as Motherbrain Labs to experiment with this.

“Where venture is doing sourcing on a very broad set of companies, this new approach is very thematic in a very narrow but deep category,” Lundström explains. Placing constraints on a category enables more nuanced experimentation with focused datasets because you’re “not just looking for generalist companies to invest in by stage.”

For example, when looking into certain healthcare technologies gaining traction, limiting the search universe yields “recommendations on what datasets to start searching through, or you can [specify a very narrowly focused dataset] like research papers or academic publications in a subset of categories and then you can go very deep,” she says.

Two portfolio companies Lundström says EQT Ventures would probably have overlooked without Motherbrain are AnyDesk, a German developer of remote work desktop software, and Handshake, a US HR tech developer whose career network for college students democratizes access to job opportunities. EQT first invested in AnyDesk in 2018 via an $8 million Series A round, then participated in its Series B round for an undisclosed amount in 2020. For Handshake, EQT participated in its Series D in 2019 and the company’s subsequent rounds in 2021 and 2022.

“Where venture is doing sourcing on a very broad set of companies, this new approach is very thematic in a very narrow but deep category”

Ashley Lundström,
EQT Ventures

Tracking talent

By mapping where talent is moving across the venture industry on an aggregate basis, Motherbrain can provide leading indicators of verticals that may be poised for a breakout. For example, Lundström is seeing a marked migration of founders who built some of the most successful consumer, B2B and fintech companies in the US and Europe toward climate tech. She’s also able to see which operators from their former companies founders are bringing with them. Getting an early read on those patterns can provide a head start for EQT to begin building relationships with those founders and operators, she notes.

While AI is well-suited to distinct tasks such as sourcing and outreach, it isn’t ready to evaluate companies for investment, says Gieselmann of Headline. “We found product-market fit to be incredibly predictive of whether a company [will be] successful,” he says. “I don’t think an AI today would have higher predictive power for us [than that].”

The primary challenge Gieselmann sees in applying AI to investment decisions is the speed at which technology evolves. “You have a handful of companies every year driving the vast majority of returns,” he says. “By the time they’re driving the vast majority of returns, seven to 10 years have passed.”

Anyone trying to find newer companies that exhibit the same signs of growth as a unicorn did 10 years earlier will learn “there are no more companies that look like [the unicorn did] 10 years ago because the technology has moved on,” he adds. “They all look very different now. There’s very different building blocks for these companies. There’s pattern drift over time. So, trying to train a machine on a very small training set of a handful of companies every year that drive the supermajority of returns is very difficult.”