At the Rise of AI Summit 2020, Dr. Christian Nagel, co-founder and managing partner of Earlybird Venture Capital, explains why he thinks the largely unchanged venture capital value chain, sourcing, screening, due diligence, closing and portfolio support, is about to be reshaped by AI. He traces the shift to an anecdote in which a US venture capital firm knew a promising Berlin entrepreneur that Earlybird had missed, prompting the firm to build its own tech-supported sourcing engine. That engine pulls from more than 100 sources, including GitHub, Product Hunt, public registries, social media, Crunchbase, PitchBook and CB Insights, to identify and enrich data on startups, and ranks those sources for reliability based on a study by the Technical University of Munich. The same study found the engine's picks matched the roughly 80 percent accuracy and recall of Earlybird's best human investors, though Nagel stresses the final investment decision still rests with people. He describes how auto-sourcing widened the firm's visible European deal flow and answers audience questions on gaming the algorithm, market sizing and current selection criteria.
Key takeaways
- Nagel says a US VC's knowledge of a Berlin entrepreneur Earlybird had missed triggered the firm to build its own AI sourcing engine.
- Earlybird's sourcing engine pulls from more than 100 sources, including GitHub, Product Hunt, Crunchbase, PitchBook and CB Insights.
- A Technical University of Munich study found the engine matched the roughly 80 percent accuracy and recall of Earlybird's best human investors.
- Auto-sourcing raised Earlybird's visible investment opportunities from about 8,000 to roughly 30,000 a year, including about 13,000 in Western Europe.
- Nagel says the investment decision still rests with human investors, and expects due diligence, closing and portfolio support to be automated next.
Detailed notes
Why the VC industry is starting to change
Nagel says the trigger was an anecdote: a US venture capital firm called Earlybird about a promising Berlin entrepreneur that Earlybird, despite its network, had never encountered. After this happened more than once, the firm concluded competitors had tech-based support it lacked. He walks through the classic VC value chain, sourcing, screening, due diligence, closing, portfolio support and exit, describing it as mostly manual and inbound, with the investor waiting for ideas to arrive. He expects this to flip toward outbound sourcing, already underway in the US where capital is abundant, and increasingly necessary in Europe, where he estimates roughly 40,000 to 50,000 promising ideas emerge each year against only about 12,000 actual investments and Earlybird's own database of around 12,000 companies.
Building an AI-powered sourcing engine
Earlybird built its own technology to source and enrich startup data, first identifying companies through web crawlers covering more than 100 sources, including GitHub, Product Hunt, public company registries and social media, then enriching that data with news, payment and traffic data, Crunchbase, PitchBook and CB Insights investor information. A study by the Technical University of Munich compared these sources and ranked them by how precise, current and valuable their information is, giving the firm a learning starting point for weighting sources over time.
Can the algorithm be trusted?
Nagel frames the goal as reaching high automation with reliable results, since investors otherwise trust direct, manually verified sources over combined algorithmic ones. The same Technical University of Munich study tested this by giving investors and the engine the same sets of ten companies to rank; the engine matched the roughly 80 percent accuracy and recall hit rate of the firm's best-performing human investors. Nagel is clear the engine does not make the investment decision itself, only supports and increasingly will support it with more transparency and accuracy.
Widening and then filtering deal flow
Auto-sourcing raised the number of investment opportunities Earlybird sees each year, Nagel says, from about 8,000 to roughly 30,000 overall and to about 13,000 in Western Europe alone. He argues the real value is not simply seeing more opportunities but using the engine to filter that larger pool down to the matches most relevant to Earlybird's investment strategy, so the firm ends up with fewer but more relevant deals to review manually. He expects similar support to extend next into screening, due diligence, closing via smart contracts, and competitor analysis for portfolio companies, though Earlybird has not yet built tools for those later stages.
Q&A highlights
Asked whether founders could game the algorithm, for example by inflating GitHub stars, Nagel says engines can always be gamed but will keep improving, and jokes that simply calling yourself a founder already attracts VC interest. Asked about market sizing, he says it remains difficult to automate, especially for ideas that open entirely new, 'blue ocean' markets with no defined size, and still requires manual analysis. Asked what currently drives selection, he says it is the combination of the entrepreneur, including whether they are a repeat founder, and early product traction, which the engine can help detect as usage spikes. Asked which AI techniques Earlybird applies, he says the current focus is widening and training the top-of-funnel sourcing engine on relevance criteria such as sector and geography, rather than natural language processing for due diligence.
Quotes
It's mainly because nothing has changed for a long period of time.
Christian Nagel
There's always a way to cheat engines, then the engines will come better.
Christian Nagel
If you want to get attraction by VC, just change your title to founder and everybody will jump on you.
Christian Nagel
Questions answered in this talk
How is Earlybird Venture Capital using AI to find startups?
According to Christian Nagel at the Rise of AI Summit 2020, Earlybird built an engine that crawls more than 100 sources, including GitHub, Product Hunt, public registries and social media, then enriches that data with news, payment data, Crunchbase, PitchBook and CB Insights information to identify and rank promising startups.
How accurate is Earlybird's AI sourcing engine compared to human investors?
According to Christian Nagel at the Rise of AI Summit 2020, a Technical University of Munich study found the engine matched the roughly 80 percent accuracy and recall hit rate of Earlybird's best-performing human investors when both were asked to rank the same sets of companies.
Does an AI algorithm make the venture capital investment decision?
According to Christian Nagel at the Rise of AI Summit 2020, no; the engine supports sourcing and screening with more transparency and accuracy, but the final investment decision still rests with the human investor.