Although venture capital industry relies heavily on human judgment and relationships, there is an emergent trend of starting to leverage generative AI and analytics to support investors across the whole investment journey. Deal sourcing process stands as one of the critical processes of this journey given the low conversion rates of the industry (<1%).
Our client’s investment team had an ambitious plan regarding startups to invest in the next few years which brought the key challenge of improving the current deal sourcing process. Sourcing relied on inbound or outbound lead generation through market research, working the network, and attending events to source potential investment opportunities. Screening covered understanding startups’ business fit and basic criteria of exclusion of moving on to a more detailed analysis.
Both processes required mostly manual effort including traditional daunting and time-consuming tasks such as visiting startups’ websites to better understand their business. The team was still leveraging from a few sources to identify potential startups leaving a universe of companies off the radar. Screening lacked clear criteria and a structured analysis of key metrics and variables to prioritize startup companies (e.g., FTEs growth, last deal size, founders, investors).
In terms of data and as a starting point, startups with fit and contact history were in a centralized platform but the companies previously analyzed and excluded were stored in several local files.
Given these challenges, this VC firm needed to improve the current deal sourcing process and bring an AI and analytical approach to augment and streamline deal sourcing efforts. Would it be possible to explore multiple sources to identify new startups to fuel the funnel? From those, how to gather as much as possible information and key metrics to prioritize the ones that should go through the investment selection stage? Plus, is it possible to perform all these steps with minimum teams’ efforts?
To address our client’s challenge, we designed and implemented a tool to support the investment team to gather startups from several sources, characterize them with qualitative and quantitative data leveraging the license of Pitchbook, score startups and export results to a centralized platform.
The tool was developed using generative AI to enable extraction and characterization of startups and an analytical model to score startups given a set of criteria discussed with the investment team.
The extraction of startups from multiple sources of structured (e.g., Crunchbase) and unstructured data (e.g., newsletters of the industry, events’ websites, news, startups’ websites) was enabled by a web scraper. Once a bulk of text and information was extracted from these sources, generative AI was applied to summarize and structure contents.
At this stage several prompts are performed to obtain structured information about the startups such as: brief description, region, vertical or horizontal industry, type of technology, contribution to Sustainability Development Goals (SDGs), just to name a few. The quality of the results of this stage gradually increased through better prompting and continuous improvement of collected input data (e.g., inclusion of Google meta descriptions).
The second step of the model was to enrich startups’ characterization with key metrics from Pitchbook. Pitchbook enrichment covered key metrics such as: number of FTEs, last deal size, exit predictor, among others – the tool is flexible enough so that the team can carefully choose which metrics to add or remove overtime.
Once startups have a thorough characterization with qualitative and quantitative data, an analytical scoring model is applied to assess startups’ attractiveness. For this purpose, several criteria are used, having different weights according to historical and market data. At this stage investors are given a list of startups ranked by their attractiveness so that they can proceed with the investment selection stage.
Finally, all the startups analyzed are exported to a centralized platform that allows to gradually build a structured database including startups with fit but also the ones excluded and with the reason behind that choice identified.
For our partner’s investment team to use the tool autonomously, a simple interface was built, ensuring not only easiness to use but also flexibility (e.g., the team can choose the sources to search for startups, fields to be characterized, criteria to fuel the scoring of the analytical model).
The tool became part of the investment team’s deal sourcing process that regularly sources new startups through multiple sources and analyses the results of the screening.
The developed tool guide and the performed training sessions with the team were critical to ensure the team’s adoption.
The tool successfully augmented deal sourcing efforts identifying and evaluating ~3x more startups when compared to the baseline.
Also, it has contributed to notable increase in efficiency, streamlining previously time-consuming tasks that became ‘one-click’ steps, simplifying the retrieval of company descriptions and key metrics. This automation not only saves time but also ensures that crucial information is readily available for assessment.
Finally, the tool has played a pivotal role in improving the investment team’s decision-making process. The incorporation of a scoring system facilitates a more objective evaluation of potential investments.
