Pinned post
Paul Klay
Verified account Begin Capital@PaulKlayVC
2.7K Following 4.5K Followers 1.2K Posts
Latest posts 5
using AI to score pitch decks is making VCs miss their best investments.
a 2026 study tested AI models on 171 real startups and found that when the models had no prior knowledge of a company, their predictions of whether it would survive had no statistically significant relationship with the actual outcomes (p = 0.31).
meanwhile, in an October 2026 LinkedIn poll, 72% of VCs said they were already using AI to review pitch decks. 39% used it before human review, while 29% used it afterward.
it's a self-selected poll, so we can't treat it as representative of the entire industry. still, it gives us an idea of how quickly these tools are becoming part of the investment process.
a broader survey by Affinity of 275 private capital professionals found that 28% were using AI for investment decisions, up from 13% in the previous survey.
1. why AI scoring is particularly difficult at the earliest stages
at pre-seed and seed, there's often very little reliable data to work with. the product might not exist yet, the market might still be emerging, and the business model could change completely over the next few years.
investing at this stage often means betting on the founders' ability to execute, adapt, recruit exceptional people, and build something that doesn't exist today. and most of that is extremely difficult to judge from a pitch deck. AI is good at finding patterns in existing information. but the startups capable of generating exceptional returns don't necessarily look like companies that have succeeded in the past.
there's also the question of how we define success. model might be good at predicting which startups will raise another round or survive for five years. neither tells us much about which one could eventually deliver a 100x return. venture returns are heavily concentrated in a small number of winners. missing one exceptional investment can be far more costly than correctly rejecting hundreds of average companies.
another 2026 study ran 26,000 AI evaluations of startup pitches, changing founder names while keeping the underlying presentations identical. the researchers found that identity cues influenced the results, and that models appearing to avoid discrimination didn't necessarily produce fairer selection outcomes. in other words, a scoring system can look objective while still making biased decisions.
2. where AI actually seems to work
i think the strongest use cases are those with enough structured data and clear evaluation criteria.
private equity and M&A are good examples.
in January 2026, McKinsey described a company that used AI to identify and score more than 500 potential acquisition targets in less than a day. the team narrowed the list to 15 leads and completed three acquisitions within a few months.
that's a very different challenge from picking pre-seed founders. in this case, the goal is to evaluate existing businesses based on financial performance, customer segments, market size, growth, and strategic fit. with clear requirements, AI can compare hundreds of companies much faster than a human team.
3. when i'd actually use AI scoring in venture capital
for accelerators receiving thousands of applications, the situation is different. it's simply not realistic to give every founder the same amount of attention.
for example, EHE Venture Studio reported receiving around 400 pitch decks for just eight available places. it uses AI to provide feedback at a scale that would be difficult for a small team to handle manually. in cases like this, AI can help organize applications, spot missing information, compare startups against specific criteria, and highlight those that deserve a closer look.
i see much more value in using AI to evaluate more opportunities than in letting it decide which founders deserve funding. the more structured the information and the clearer the criteria, the more useful these tools become. but at the earliest stages, when success depends heavily on exceptional founders building something unconventional, i'd be very careful about using AI to filter out potential investments.
a 2026 study tested AI models on 171 real startups and found that when the models had no prior knowledge of a company, their predictions of whether it would survive had no statistically significant relationship with the actual outcomes (p = 0.31).
meanwhile, in an October 2026 LinkedIn poll, 72% of VCs said they were already using AI to review pitch decks. 39% used it before human review, while 29% used it afterward.
it's a self-selected poll, so we can't treat it as representative of the entire industry. still, it gives us an idea of how quickly these tools are becoming part of the investment process.
a broader survey by Affinity of 275 private capital professionals found that 28% were using AI for investment decisions, up from 13% in the previous survey.
1. why AI scoring is particularly difficult at the earliest stages
at pre-seed and seed, there's often very little reliable data to work with. the product might not exist yet, the market might still be emerging, and the business model could change completely over the next few years.
investing at this stage often means betting on the founders' ability to execute, adapt, recruit exceptional people, and build something that doesn't exist today. and most of that is extremely difficult to judge from a pitch deck. AI is good at finding patterns in existing information. but the startups capable of generating exceptional returns don't necessarily look like companies that have succeeded in the past.
there's also the question of how we define success. model might be good at predicting which startups will raise another round or survive for five years. neither tells us much about which one could eventually deliver a 100x return. venture returns are heavily concentrated in a small number of winners. missing one exceptional investment can be far more costly than correctly rejecting hundreds of average companies.
another 2026 study ran 26,000 AI evaluations of startup pitches, changing founder names while keeping the underlying presentations identical. the researchers found that identity cues influenced the results, and that models appearing to avoid discrimination didn't necessarily produce fairer selection outcomes. in other words, a scoring system can look objective while still making biased decisions.
2. where AI actually seems to work
i think the strongest use cases are those with enough structured data and clear evaluation criteria.
private equity and M&A are good examples.
in January 2026, McKinsey described a company that used AI to identify and score more than 500 potential acquisition targets in less than a day. the team narrowed the list to 15 leads and completed three acquisitions within a few months.
that's a very different challenge from picking pre-seed founders. in this case, the goal is to evaluate existing businesses based on financial performance, customer segments, market size, growth, and strategic fit. with clear requirements, AI can compare hundreds of companies much faster than a human team.
3. when i'd actually use AI scoring in venture capital
for accelerators receiving thousands of applications, the situation is different. it's simply not realistic to give every founder the same amount of attention.
for example, EHE Venture Studio reported receiving around 400 pitch decks for just eight available places. it uses AI to provide feedback at a scale that would be difficult for a small team to handle manually. in cases like this, AI can help organize applications, spot missing information, compare startups against specific criteria, and highlight those that deserve a closer look.
i see much more value in using AI to evaluate more opportunities than in letting it decide which founders deserve funding. the more structured the information and the clearer the criteria, the more useful these tools become. but at the earliest stages, when success depends heavily on exceptional founders building something unconventional, i'd be very careful about using AI to filter out potential investments.
3 key questions for every consumer AI product:View quoted post on X
1. Does your product give people new superpowers?
2. Does your product get better as more people use it?
3. Does your product get better when models get better?
recently, i introduced someone i know to a team for a potential partnership. within the first minute of their call, he texted me: "they're Indian, they have no money. waste of time."
another person wrote in a founders' group chat: "i can only work with young people as an advisor. everyone should have their age visible."
so many people miss out on huge opportunities simply because they believe things that aren't even true.
i get that these opinions might come from past experiences. but being open-minded is a much better approach, and when it comes to business, it's actually more profitable.
of course, that doesn't mean making the same mistakes over and over. the problem is taking one characteristic and using it to judge an entire group of people.
a couple of bad experiences don't define everyone who shares a certain background, age, or nationality.
there are much better ways to evaluate people and decide who to work with.
another person wrote in a founders' group chat: "i can only work with young people as an advisor. everyone should have their age visible."
so many people miss out on huge opportunities simply because they believe things that aren't even true.
i get that these opinions might come from past experiences. but being open-minded is a much better approach, and when it comes to business, it's actually more profitable.
of course, that doesn't mean making the same mistakes over and over. the problem is taking one characteristic and using it to judge an entire group of people.
a couple of bad experiences don't define everyone who shares a certain background, age, or nationality.
there are much better ways to evaluate people and decide who to work with.
Our education system rewards CPU like minds. Sequential thinking, one problem at a time.View quoted post on X
The future might reward GPU like minds.
ADHD might finally have found its product-market fit.
the best way to become great at sales and build a network of ideal customers is to become one of them yourself.
i have a framework for doing this without spending years on it, while still building your product.
the idea is simple: if you're building a product for influencers, become an influencer yourself. this helps you understand their problems firsthand and build relationships with exactly the people you want to sell to. most importantly, they'll see you as one of them rather than someone trying to sell them something. that puts you in a much stronger position when it comes to sales.
here's how to do it faster:
1. find someone who's already where you want to be.
starting from scratch takes time, and there's no guarantee of success. instead, find someone who's already part of your target audience, has relevant experience, and knows the right people. the goal is to build a close working relationship with them.
2. become their assistant or bring them on as an advisor.
the right approach depends on your ideal customer.
let's say you're building a product for CMOs at companies doing $1B+ in revenue. building a billion-dollar company just to understand those customers obviously isn't realistic. neither is landing a CMO role at one overnight. but becoming an assistant to one of those CMOs is much more achievable.
if the target audience is more accessible, like influencers, there's another approach. start building an account and bring a successful influencer on as an advisor to help it grow.
3. give them a reason to work with you.
the simplest answer is money.
if you're becoming someone's assistant: don't ask for a salary. you could even offer to pay them for the opportunity. this removes the risk of hiring someone who might waste their time. once you're in, focus on being genuinely useful. gradually take on more responsibility, learn how they work, and build relationships with people in their network.
if you're working with an advisor: pay them, but also offer a share of the results they help achieve. for example, let's say an influencer is helping grow an account. the goal is to understand the market and build connections with other influencers, so advertising revenue isn't really a priority. offering the advisor almost all the advertising revenue the account generates could be a good deal. this gives them a strong incentive to help the account grow while providing valuable experience, an audience, and the right connections.
that's the whole idea. becoming part of a market doesn't have to take years.
relatively small investment can provide access to experience and connections that took someone else decades to build.
all while continuing to build a startup.
i have a framework for doing this without spending years on it, while still building your product.
the idea is simple: if you're building a product for influencers, become an influencer yourself. this helps you understand their problems firsthand and build relationships with exactly the people you want to sell to. most importantly, they'll see you as one of them rather than someone trying to sell them something. that puts you in a much stronger position when it comes to sales.
here's how to do it faster:
1. find someone who's already where you want to be.
starting from scratch takes time, and there's no guarantee of success. instead, find someone who's already part of your target audience, has relevant experience, and knows the right people. the goal is to build a close working relationship with them.
2. become their assistant or bring them on as an advisor.
the right approach depends on your ideal customer.
let's say you're building a product for CMOs at companies doing $1B+ in revenue. building a billion-dollar company just to understand those customers obviously isn't realistic. neither is landing a CMO role at one overnight. but becoming an assistant to one of those CMOs is much more achievable.
if the target audience is more accessible, like influencers, there's another approach. start building an account and bring a successful influencer on as an advisor to help it grow.
3. give them a reason to work with you.
the simplest answer is money.
if you're becoming someone's assistant: don't ask for a salary. you could even offer to pay them for the opportunity. this removes the risk of hiring someone who might waste their time. once you're in, focus on being genuinely useful. gradually take on more responsibility, learn how they work, and build relationships with people in their network.
if you're working with an advisor: pay them, but also offer a share of the results they help achieve. for example, let's say an influencer is helping grow an account. the goal is to understand the market and build connections with other influencers, so advertising revenue isn't really a priority. offering the advisor almost all the advertising revenue the account generates could be a good deal. this gives them a strong incentive to help the account grow while providing valuable experience, an audience, and the right connections.
that's the whole idea. becoming part of a market doesn't have to take years.
relatively small investment can provide access to experience and connections that took someone else decades to build.
all while continuing to build a startup.
"We all must have experienced how lazily, and, consequently, how ineffectually, instruction is received when forced upon the mind by others. Few have been taught to any purpose who have not been their own teachers."View quoted post on X
— Joshua Reynolds