Primer on the use of AI in Early-Stage Venture Capital

September 12, 2024

This article explores how AI is revolutionizing venture capital by enhancing deal sourcing, reducing biases, and improving decision-making, offering VCs a competitive edge in discovering high-potential startups earlier.

The landscape of venture capital is evolving at a rapid pace. What was once an industry built on intuition, personal networks, and gut feelings is now being reshaped by data and automation. As competition intensifies, firms become indistinguishable and access to information is democratised, the VCs who will succeed are those who can move faster, spot opportunities earlier, and eliminate the inherent biases in human decision-making.

This primer is the result of my vicarious learning through interactions with multiple VCs about their deal lifecycles. There are a number of points in this primer that are highly debatable and can spark a lot of interesting debates which I am always open to. The aim of this primer is to provide a broad and opinionated starting point for VCs to begin interactions around the use of AI and data in their operations for greater successes. Success in venture capital hinges on two key metrics: the ability to attract high-quality inbound investment opportunities and delivering superior returns to limited partners (LPs). Both are intricately linked—being the first choice for top founders leads to outlier deals, which in turn generates extraordinary returns. Thus, it is essential to do the following in the deal sourcing and analysis stage:

  1. Uncover invest-worthy start-ups before other firms and potentially before the companies themselves start their fundraising process
  2. Rely on a data-driven framework to analyse companies for potential of outsized returns by eliminating biases and betting on return generating qualities
  3. Maintain high NPS scores with founders in terms of both acceptance and rejections
  4. Spend less time sifting through junk pitches and more time on helping portfolio companies reach outsized returns

To that effect, AI can help a Venture Firm with the following:

  1. Sourcing deals by uncovering/ scraping founder and start-up data daily from Tracxn, Crunchbase, LinkedIn to reach new and to be founders first
  2. Rank pitches received through the website in order of potential investability for faster sorting and responses
  3. Do preliminary quantitative and qualitative company analysis and score the companies on a proprietary scale
  4. Be updated with industry trends and news through daily/ weekly AI curated updates from YourStory, LinkedIn and various other internet sources

While all of the following can initially be individual offerings to be used as and when required to save hundreds of man-hours every month, it’s true power can be harnessed when an entire workflow will be developed out of the individual stages of the deal pipeline. This system can be developed into a tech-first, AI-integrated CRM system for the VC Fund.

This system will help do the following:

  1. Save 100s of hours of man-hours spent on manual tasks within a deal’s analysis such as competitor research, problem intensity check, buyer persona creation etc.
  2. Help the firm reach founders first through a daily updated list of new companies
  3. Rank the both inbound and outbound requests in decreasing order of investability for faster deal closing/ rejection times
  4. Maintain a one-stop repository of all investments with their theses to eventually train not just future employees on the characteristics of an outlier but also train the model to help rank companies based on their investability better
  5. Ensure that the firm remains the first choice for founders within India through the adoption of technology and highly efficient operations
High-level structure of AI-Powered CRM for VCs

AI’s use in Venture Capital around the world:

The use of AI in venture capital is not novel. VCs have been trying to incorporate data, automation and AI into their workflows for a while with varied amounts of success. Here are some of the most successful examples of how the following VC are using AI in their workflows:

  1. Tribe Capital: Tribe has grown to a 1.6B AUM fund in just 6-years using a highly-data driven approach to venture. They rely on data to the extent of quantifying metrics for Product Market Fit. The firm follows an extremely data-driven approach across the whole value chain, from automated sourcing and prescreening, through custom syndication tools to real-time portfolio monitoring.
  2. EQT Ventures: Their AI, referred to as Motherbrain, enables deal sourcing and helps make investment decisions. It allows the firm to reach out to promising start-ups well before they are inundated with offers from competing firms and with a high degree of confidence. Nine investments with outsized returns were fully sourced through Motherbrain - investments that would otherwise not have been identified. Some of these were AnyDesk, PeakOn, CodeSandbox, Handshake, Griffin, and WarDucks. Motherbrain was developed inhouse, Pedram (Head of Data Management @ EQT) discusses the intricacies here.
  3. Signalfire: Signalfire is a 2.1B AUM fund with a 85 NPS score, self-described as “most quantitative fund in the world” and “the only VC that brings a data platform to its portfolio companies.”  This is powered by an end-to-end real-time data platform called Beacon. It powers the entire value chain of a venture — from deal origination to picking the right investments, and deal syndication to portfolio support. Beacon tracks the performance of more than 6 million companies in real-time and flags companies that are outperforming or doing something notable on a dashboard, allowing Signalfire to see deals earlier than traditional venture firms.
  4. Hone Capital: Hone Capital, the VC arm of Chinese PE firm CSC Group, became Silicon Valley’s most active seed investor of 2017 despite being founded just two years earlier. Hone Capital "hacked" its way into the somewhat exclusive SV venture world by establishing a strategic partnership with Angelist. Through the use of ML models based on a dataset of 30,000 deals over the last decade, 50% of Hone’s seed stage deals have led to follow-on investments (far above the industry average). Hone estimates its unrealized returns put it among the top 20% of investors

AI-CRM for VC: Concept Abstract

This concept note discusses the possibility of an AI-powered co-pilot system to assist investors in sourcing and evaluating early-stage startup investments. The goal is to discover investable deals before the market, minimise decision-making biases while maximising financial returns. Common biases that can adversely impact investment decisions include selection bias, historical bias, representativeness heuristic, herd mentality, and anchoring bias. To maximise returns, the system would analyse factors such as founder investability, industry valuation multiples, revenue growth potential, and expected timelines to liquidity events.

The proposed system would have capabilities to create buyer personas, analyse market size, assess problem intensity, evaluate founder-market fit, benchmark key performance indicators, predict founder fundability, model exit opportunities, and present easy-to-understand business models. This would be achieved through an ensemble of technologies including fine-tuned language models, search engine APIs, web scraping, and databases.

After several rounds of iterative training on startup investment data, the system could potentially take on an AI venture capitalist role of not just assisting but actually leading investment decisions. While the human element can never be and should not be fully eliminated, this co-pilot could significantly augment human investors via rigorous quantitative and qualitative deal analysis.

Problem Statement

Develop a co-pilot for early-stage VC Investors that discover companies before the market does and evaluates investments to minimise bias and maximise returns. We won’t discuss the obvious CRM features such as transcript uploads, meeting scheduling etc and limit our discussion to things that make the product a VC-centric AI CRM.

The metrics to optimise for are:

  1. Start-up Discovery
  2. Analysis Biases Minimisation
  3. Firm Returns Maximisation

Typical Investment Discovery Process:

Sajith Pai of Blume Ventures once compared a VC to enterprise and like any sales jobs, VCs also have to jump through many hoops to develop a healthy, return-generating deal flow. The crude definition of deal flow is where the firm has start-ups to analyse for investments. There are broadly 2 form of deal flow:

  1. Inbound opportunities - Founders reaching out on LinkedIn, Pitch Deck emails and warm intros etc
  2. Outbound opportunities - Primarily being associates/ analysts reaching out to companies to apply for certain programs or hop on to call for discovery

While a VC often has a high volume of inbound opportunities, this doesn’t always translate to high quality opportunities. The opportunities with the possibility of outsized returns are either in very high demand, making the entire process extremely difficult or are completely under radar where most firms miss out on the opportunity.

Very evidently, this entire process is incredibly manual in the era of automated enterprise sales processes. Leads (aka opportunities) are not qualified objectively, too many people run behind too few deals leading to unjustifiable valuations and high quality leads are discovered incredibly late in their lifecycle. This entire process needs structure, efficiency and objectivity.

Biases that creep into an investment analysis:

There are some common biases which creep into an investor’s analysis. Here’s a list of biases segregated according to the section of the investment memo it affects:

  1. Selection Bias - Founders and investors often mis-represent the competition that exists in the market. This is very well explained through an example used by Peter Thiel - the only two Mexican restaurants in Kolkata are not competing just with each other but also with all the restaurants of other cuisines.
  2. Historical Bias - Investors can skip over a business with a truly unique insight because in the past some other company failed to deliver in that segment. For example, Amber which helps students get PG in countries outside India. Multiple companies tried to start building for India first. The problem in India while real was not enough to drive customers. Thus, Amber focused on solving this problem where the situation for customers is a lot more dire. It’s essentially the same business model with a different target market - an unique insight.
  3. Representativeness Heuristic - This is commonly seen when judging founders. Big brand names on their resume, either educational or work related, can lead to mis-judgement of the founder’s quality. Implementing this in practice is relatively more complicated. Unfortunately these are metrics that are necessary to judge a founder, but we can minimise the effects of biases here by quantifying this.
  4. Herd Mentality - Seeing big names as advisors or known investors on the cap table of the company often triggers a herd mentality in firms curbing their individual thoughts and analysis.
  5. Anchoring Bias - This can be seen when a founder inflates the size of their target market by placing themselves in a superset of their true market. This creates a larger headroom for growth and the investor is made to believe that a higher valuation is justified given the greater scope for growth. An example of this is when Uber defines itself as a logistics company instead of a transportation or a mobility company.

There are other biases which may be at play too, but for the sake of this discussion we’ll focus primarily on the one’s mentioned above.

How is an early-stage (read pre-Series A) investment return maximised?

Typically, there are 3 exit options for an pre-Series A investor:

  1. Their shares are bought out by a later stage investor
  2. The company is acquired
  3. The company IPOs

An early investor doesn’t typically stay on the cap table till IPO, so the dynamics of public market valuations do not come into play. Even if they do, we can subsume the IPO case in the later stage investor case with the caveat that these later stage investors are in fact the retail investors.

To maximise returns, we need to maximise exit multiples while minimising time to exit. The metrics to analyse for this are the following:

  1. Founder investability - How proficient are the founders in raising capital?
  2. Valuation multiples of industry - Typically what are the revenue multiples in the industry (For eg 35x in AI v/s 12x in marketplace business - numbers are just representative)?
  3. Revenue growth in industry - At what rate can a company in a particular industry scale?
  4. Time between rounds - Based on capital requirements and industry standards what is the average time between liquidity events?
  5. Comps - While at an early stage, financial comparable analysis is difficult and oftentimes futile, but certain KPIs like DAU, MAU, Churn Rate etc can be compared within the company’s industry.

Pitfalls with attempting to maximise returns with AI:

The downside with this analysis is also where the upside lies. Most venture backed companies fail and very often the ones that succeed, break all traditional norms. An AI cannot model industry breaking successes - by very definition they are a black swan event. What this analysis will do is highlight all the red flags in a potential deal. All investment opportunities have inherent risks and it’s up to the investor to define their risk appetite.

One approach to solve this problem is to introduce a metric akin to beta or a risk factor in a later stage company. This can help quantify the risk of the investment even though we cannot quantify the upside potential.

To help investors maximise returns, the other approach to the problem would be to help inform the sell decision. Relatively lower number of companies falter massively post Series A which is a sweet spot to start doing quantitative and comp analysis. Since this is outside the scope of the problem statement, I’ll refrain from delving deeper into this.

In conclusion, this co-pilot can potentially help investors make fewer mistakes and gain average returns. The model will break down for industry-altering opportunities. Thus, it’s important to treat any opinions offered by the analysis with a pinch of salt if you’re looking to make outsized returns.

Early-Stage Investment CRM

Sr

Feature

Feature Description

Benefit

 

1


Pitch Deck Emails Centralisation

Every inbound email with a pitch deck is immediately added to the CRM

This can help systemise the pitch deck review process with the help of the evaluator in next step


2


LinkedIn Inbound Centralisation

Every member in the team will have access to a UI to add inbound requests on LinkedIn

Helps include visibility into inbound requests and reduces risk of overlooking a good opportunity

 

 

3

 

 

Tracxn, Crunchbase, Other Database trackers

Companies often voluntarily list themselves on a number of databases for multiple reasons. This is a gold mine for pre-seed and seed funds. Every new company added to these databases will be added to CRM

 

Helps automate discovery without having to go through the same steps in multiple databases daily


4


LinkedIn Scraper

We scrape LinkedIn for an unique queries to try and discover founders before they blow-up. This is typically useful only for pre-Seed and Seed funds.

Develop relationships with potential category leading founders before they start their journey 


5


Twitter Scraper

While the start-up ecosystem is quite active on Twitter, its developer ecosystem is stronger. Very often these are people who are prone to starting up themselves and can be an incredible resource to find potential founders and even possibly nudge them into founding.

Develop relationships with potential category leading founders before they start their journey

This is a primarily automated system that runs automatically and enriches the CRM system with data. For instances where the user needs to input data, there will be an easy-to-use form to add it.

Early-Stage Investment Evaluator

Sr

Feature

Feature Description

Benefit

 

1

Buyer Persona Creation

A descriptive buyer persona based on the business idea presented

Helps confirm the target market

 

2

 

Competition Search

List of competition in the niche, their monthly traffic, funding information

Helps judge the competitive landscape

 

 

3

 

 

Market Size Analysis

Broader target market analysis. For a company like Urban Clap, may not be able to narrow down to Tier-1/2 and instead calculate the market size of entire home service industry

 

Provides a ballpark figure to base calculations or judge mentioned market size

 

4

 

Problem Intensity Search

Use metrics like search volume, reddit threads and other proxy metrics to create a problem intensity score

 

Estimates the need of the solution in the market

 

5

 

Founder-Market Fit Check

Gains information on the founder from their work experience, LinkedIn, Twitter etc to assign a founder market fit score

 

Judges the founder's know-how in the market

 

6

 

Comp KPI Analysis

Comparative analysis of industry/ segment specific metrics for a company such as DAU, MAU, Churn etc

Provides a comparative positioning of a company w.r.t other successful players in similar industries

 

7

 

Founder Investability

A score to judge the investability of the founder based on metrics such as work ex, university, cap table

 

Helps judge potential exit opportunities

 

8

 

Exit Analysis

Analyses exits in similar spaces, venture capital funds that invest in the niche

Helps gauge exit opportunities and feasibility

 

9

 

Valuation Multiples

Typical valuation multiples that a company in the particular space gets

Back-of-the hand estimates for potential upsides

 

10

 

Business Model Presentation

A simple and clear presentation of the business model, clearly mentioning the cost drivers and revenue sources

Easy understanding of the business model and potential pitfalls

Potential UI/UX Ideas:

Since ChatGPT, chat interfaces have become the norm for any type of co-pilot. This puts the onus of asking the right questions on the user, thus I do not prefer a chat interface for most use cases. In my conception of this co-pilot, we’ll generate an analysis for each business and provide the above data in the form of either an investment memo or a standard analysis to the user.

While I believe it’s better for the user to go through this in the form of an investment memo or analysis created by the AI, users will demand a chat interface with which they can interact. This can be achieved using a simple chatbot with a basic information retrieval system.

Training and Implementation of the features:

Algorithm for Competition Search
Algorithm for Problem Intensity Score
Algorithm for Founder-Market Fit Check
Algorithm for Founder Investability

Primary Technologies Required:

  1. Python - For ML models & LLMs orchestrations
  2. LLM - OpenAI or open-source LLM
  3. Crunchbase, Google Search, Ahref, LinkedIn, Reddit, Twitter APIs
  4. LangChain - For interacting with LLMs
  5. PostgreSQL/ MongoDB - For Data storage
  6. AWS EC2 - Hosting Python instances
  7. Docker & Kubernetes - To eventually scale the system for higher throughput
  8. Selenium Webdriver - For web scraping
  9. Etc…

Conclusions

The integration of AI into a VC firm's operations would represent a paradigm shift in how the firm sources, evaluates, and manages investments. By automating and enhancing various stages of the investment process, AI tools can significantly boost the efficiency and effectiveness of the firm. To reach its full potential, we need to make the product described above much more robust but this is more than sufficient for a v1.

Firstly, AI aids in deal sourcing by continually collecting data from platforms like Tracxn, Crunchbase, and LinkedIn, ensuring that the firm is among the first to identify promising startups. This proactive approach enables the firm to discover and engage with high-potential founders before their competitors, thus securing better investment opportunities.

Secondly, AI-powered ranking systems streamline the evaluation of pitches, enabling faster sorting and response times. By scoring companies based on proprietary metrics, AI helps prioritise the most investable startups, ensuring that valuable time is spent on the most promising opportunities.

Furthermore, AI contributes to detailed preliminary analysis, including competitor research and buyer persona creation, saving hundreds of man-hours. This allows the firm to focus more on supporting their portfolio companies, ultimately driving outsized returns.

Globally, leading firms like EQT Ventures, Signalfire, and Hone Capital demonstrate the tangible benefits of AI in VC. EQT's Motherbrain, Signalfire's Beacon, and Hone Capital's machine-learning model showcase how AI can enhance deal origination, investment decision-making, and portfolio support. These firms leverage AI to maintain a competitive edge by identifying trends, benchmarking performance, and providing real-time insights, which are crucial for informed investment decisions.

After 3-5 years of iterative training, I believe this model has the ability to eventually move towards an AI-Venture Capitalist model. Inventus Capital has been envisioning something of this form but hasn’t shown any concrete results yet. Instead of Inventus Capital’s goal of replacing human VCs, I believe this tool can reduce the number of analysts required by a firm for memo writing etc. All of the grunt work can be outsourced to the AI while the investment managers work on the softer aspects of understanding and predicting human behaviour.

If this intrigues you and you want to deep dive into the various aspects of this CRM, I’d love to talk. You can reach out to me here or hit me up at prannaykedia1@gmail.com.

Sign-up to my newsletter

I talk about start-up, investing, tech and everything around it. Sign-Up for my weekly newsletter to never miss an update