Subscribe
Share
Share
Embed
The "manual era" of capital allocation is in its final chapter. The firms still relying on manual data extraction and analysis aren’t failing overnight, but they are falling behind one week at a time. While most of the industry continues to "white-knuckle" through 200-page documents and legacy databases, and manual Excel extraction, a new breed of Agentic Allocators is quietly rewriting the rules. They aren’t just using AI to summarize emails; they are leveraging AI-augmented workflows that intelligently automate parts of their investment and operational processes that were previously impossible to automate.
Hosted by Victoria Sienczewski, CEO and Founder of AuumAI, The Agentic Allocator is the "behind-closed-doors" look at how the world's most sophisticated Limited Partners (LPs), allocators and General Partners (GPs) are actually deploying AI, and the hard-won lessons from those building the systems.
This isn't a series about high-level theory or technical gibberish. Each conversation features industry leaders, forward-thinking LPs, GPs and experts who are rewriting the rules of capital allocation through agentic AI. Expect real-world case studies, tactical frameworks you can actually use, and moments that challenge outdated norms. You'll come away with a clearer understanding of the critical questions every allocator must ask - about data privacy, team adoption, integration, and governance - before investing in any AI solution. If you're tired of the "black box" and ready to evolve your investment office for what comes next, you're in the right place.
Welcome to The Agentic Allocator. Today, we're joined by Professor Emmanuel Yimfor, Professor of Finance at Columbia Business School, whose research examines the frictions and biases shaping how capital flows through private markets with a focus on racial and gender gaps in VC funding, board representation, and private capital fundraising.
Emmanuel Yimfor:The key idea is to be able to expand the search space. Given that this tool helps me reduce the cost of due diligence, can I expand the set of potential people I meet to those that are outside my network?
Victoria Sienczewski:His work reveals how alumni networks, homophily, and hidden signals determine who gets funded and who does not.
Emmanuel Yimfor:The danger can be a little bit subtle. So picture an LP who feeds a new manager's material, so the private placement memorandum or even the one pager, and asks the AI how similar is this person to the managers who were back before. Now that sounds objective, but if the past portfolio was built through the same kind of narrow networks, then the model is just automating the same types of exclusions.
Victoria Sienczewski:What makes this conversation so timely is the question of what happens when these dynamics get encoded into AI. As the industry moves towards agentic capital matching and digital representations of LPs and GPs, the biases professor Yimfor has documented risk becoming embedded in the technology itself.
Emmanuel Yimfor:These are really exciting times. And so to the extent that the industry adopts some of this technology in the right way, not in ways that are related to how past decisions have been made, I could really see a potential where it not only expands the pool of who is receiving capital, but it also allows this capital to be allocated towards its most productive uses.
Victoria Sienczewski:I hope you enjoy the conversation. Professor Yimfor, thank you so much for taking the time to join me for the conversation.
Emmanuel Yimfor:Thank you for having me.
Victoria Sienczewski:Before we dive into AI, would love to focus on your research. So your research shows that there and documents that there is a long term racial and gender disparity in terms of private capital funding. Would love for you to dive into that for our listeners today.
Emmanuel Yimfor:Oh, absolutely. My research in general is about the core frictions driving disparities in access to capital in private markets. And so you talk about race and gender. Let me take each in turn because, you know, the mechanism is the same, but, you know, the details diff so for race, so far what my work has shown is that, you know, the core friction is networks. Right?
Emmanuel Yimfor:Like, who you can reach as opposed to necessarily how how good you are. And so if you look at more, you know, structured ways of, you know, reaching funders, I'm talking about things like, you know, accelerators or or grant funding. Essentially, the racial gap in access to funding disappears. It mostly shows up when things are more relationship driven. Like if you think about founders, for example, trying to raise black or Hispanic founders trying to raise venture capital funding from GPs, you know, they raise about 40% less even given, you know, identical patent holdings or going to the same school or even having the same track record.
Emmanuel Yimfor:Record. We see, you know, very similar patterns for gender, which are, you know, very striking. In founder space, one thing that we did is, you know, good start ups that were formed by men and women, and then we just track what happened when Anne went on to start a new company relative to the woman. And so what we found is that, you know, the woman raises less funding relative to her male co founder even though they share very much the same track record. And so, that's kind of like what I've seen on the founder space with respect to race and gender, raising capital from GPs.
Emmanuel Yimfor:This core friction of networks, it shows up again when we look at GPs raising capital from limited partners. Here what we find is one of the key barriers is the ability to raise that first time fund, Right? Even if you look at even the allocators that prefer to back, you know, female or minority mentees, a lot of those, you know, allocators need a track record to be able to back them. So I'm thinking about, like, pension funds or even endowments. But how do you get a track record in the first place?
Emmanuel Yimfor:You need to be able to get access to either the type of GPs, family offices, high net worth individuals, now more active in forming funding first time fund. And so given this gap, it's no surprise that when we look overall, there is a gap in funding and how much capital GPs can raise, especially when they are racial minorities or women. So, you know, at the risk of going on forever, I've been thinking about this for a while, that's what I would summarize as like, one of the main takeaway frictions are networks. So like who you can reach as opposed to how good you are.
Victoria Sienczewski:Thank you for that great summary Professor Yimfor. I have a vision of the future in which LPs and GPs will be AI enabled. They'll be using Agentic AI systems and maybe even communicating to each other through those agents. You mentioned your research and the biases you've uncovered in terms of race, gender, and alumni networks. How should we think about those biases and how they could be potentially perpetuated within the context?
Emmanuel Yimfor:If I could start with a caveat, you know, I think whether, you know, AI systems, you know, amplify, you know, existing disparities in capital allocation would depend on, you know, how the AI system is trained. Right? And, you know, what kind of input data the AI system uses. So the the danger can be a little bit subtle. So if you if you picture an LP who, you know, feeds a new manager, you know, a new manager's material, so the private placement memorandum or even the one pager, and ask the AI, you know, how similar is this person to the managers, you know, were back before.
Emmanuel Yimfor:Now, that sounds objective, you know, but if the past portfolio was built through the same kind of narrow networks, then the model is just automating, you know, the same, you know, types of exclusions. But then if you think about, you know, being able to use the time savings that are generated from being able to process a lot more private placement memorandums or process a lot more, you know, one pagers and you leverage that to say, let's expand the set of one pagers that we're looking for. Again, going back to my earlier work, imagine an LP that takes seriously the apply here button. And then based on the apply here button, you know, filters the set of potential, you know, pitch decks to or private placement memorandum to, you know, which ones fit within my fund mandate. And then when they filter that to which one fits within my fund mandate, they're going directly, you know, to your question about agentic AI.
Emmanuel Yimfor:You know, imagine a situation where instead of, like, in person meetings, we're meeting using avatars in the metaverse. Right? You can imagine a structured type approach where, you know, we have the same standardized questions that we ask, you know, candidates that are similar ish. So the AI would have gone through, looked at the materials, and they would have said, you know, Victoria is very similar to Emmanuel in that, like, they have very much the same strategy, you know, and they fit my funds mandate in the exact same way, then there's no reason why, like, ex ante, you can't think harder about the types of questions that you would ask because there have been research that have been shown that, you know, racial minorities or even women tend to be asked different questions. And so in that sense, you know, I'm not worried about potentially showing up to a pitch meeting and reading off your body language because I can't see your body language.
Emmanuel Yimfor:All I can see is an avatar, you know, asking me, you know, objective pre pre planned pre planned questions and, you know, I can focus more on instead of reading and reacting to what I think is your body language just because if we're from different cultures, I might be reading your body language wrong. It's happened to me where people meet me a few times and I'm thinking about what they are saying and what they see is a skeptical look and then they re explain themselves. I have learned that, you know, when they're trying to re explain themselves to tell them, hey, please don't pay attention to my face. I have facial spasms. And so, you know, my facial expressions don't mean like I'm being skeptical.
Emmanuel Yimfor:But imagine a situation where you didn't see my facial reaction. You could just focus on the explanation that you're giving. Just so just to be clear, I want to summarize so, you know, I'm not I'm not rambling and fitting the, you know, assistant professor stereotype. The the key idea is the key idea is to be able to expand the search space. Right?
Emmanuel Yimfor:Who are the given that this tool helps me reduce the cost of due diligence, can I expand the set of potential people I meet to those that are outside my network? Like, they didn't go to the same school as I did. You know, they didn't work at the same places that I did because there are good ideas out there, good ideas for, you know, how to allocate capital efficiently. And can I reduce frictions in the meeting process and the interaction process where you're just trying to learn about what the GP strategy is? Of course, you can't substitute for the in person meeting because at the end of the day, you still need to make sure that this person has ideas and then they're just not parroting something that they're learning from some AI algorithm.
Emmanuel Yimfor:And so you can never replace the in person get to know the person behind the strategy. But that could be deferred until much later while we get rid of like this, you know, disparities in in capital allocation.
Victoria Sienczewski:It's a fabulous point, and I fully agree. If we can get to the point where LPs have the widest top of funnel so today, as you know, there are 56,000 alternative investment managers. So I'm not only including private markets, but also hedge funds in that number. And LP is the most sophisticated. We'll look at $400 or $500 a year in today's world.
Victoria Sienczewski:And so if we're able to widen that funnel, I fully agree with you that we'll get to better investment outcomes over time and also better matches for LPs and GPs. While I see that world in terms of the wider top of funnel being extremely valuable, I want to also touch on some of the subtleties. So you mentioned the risk of embedding based matching in a conversation we had earlier. And would love to understand what are the risks of that? So you have, let's say, top level hard filters that LPs have, which will be maybe less biased.
Victoria Sienczewski:But when we get to the subtler elements around the qualitative, how I feel about a person, what are the risks around that, professor?
Emmanuel Yimfor:Oh, yeah. So just to take a step back, what is when I hear embeddings based matching, like, what does it mean to me in just plain English? So, the idea is that, you know, you can imagine taking a GP's, you know, one pager and their PPM and then, you know, feeding it into a function, you know, a function that is gonna take all that text and then return some numbers. Now with those numbers, you can do a lot of things. Like, for the with those numbers, you could compare how those numbers are similar to say other GPs that you have backed in the past.
Emmanuel Yimfor:And so once you do that, you know, behind the hood, even if you have never, you know, made a decision about, you know, allocating to a GP based on their race, based on their gender, you know, or based on where they went to school. Behind the hood, how the model is going to, like, you know, compute what is similar to what you've done in the past is kind of a black box. You might not be sure exactly what the model is doing, but what it might end up doing is it might end up, like, picking up on those very same signals that even you don't agree you should be using, you know, to allocate capital. And so from that benchmark of, like, you know, let's take a pitch deck, convert it into numbers, and see how it's similar, you know, to what we have done in the past. Those past decisions might pollute, you know, your future decisions in ways that even you might not be able to pick up on.
Emmanuel Yimfor:And so, you know, you know, making sure that even if we're using this embedding space matching, we're opening the black box. Right? Like, you know, making sure that, you know, the specific features that are being fed in, you know, for any algorithm to learn are features that we think can help us improve the allocation of capital. Like, I don't know, off the top of my head, how much competition is there? How many other managers are running a similar strategy?
Emmanuel Yimfor:How long have those managers been running it? What are the key risks? To what extent has this manager mitigated some of the key risks that might arise within their strategy, foreign exchange risk, political risk? So to the extent that we're using some of these embeddings to predict some of those key risks and maybe rank managers or even generate fodder for, you know, the in person meeting and the kind of questions that we ask that we really have the opportunity to be able to, like, you know, expand. I'm gonna use your words.
Emmanuel Yimfor:Instead of saying search space, I think top of the funnel is a lot more intuitive. Expand the top of the funnel and be able to, you know, like, you know, you know, take some of that time savings and transform those time savings into having more, you know, in in person meetings and just making sure not to accept the output of the AI and being more critical about what exactly it's doing in the background or what features it's learning to predict who you get to match with.
Victoria Sienczewski:And professor, you have done a lot of research in this space and also experimented with agentic models and how to reduce some of this bias. What have you found?
Emmanuel Yimfor:Yeah. So like, you know, what I found is that, you know, the most important thing, again, at the risk of, like, going back to some of the themes that we've been looking at, is thinking hard about how, you know, this technology has the potential to expand the top of the funnel. And so what I have found from my work is that if you were to zoom in at places where the top of the funnel is as wide as possible, so let's think about accelerators, you know, where they have an apply here button and everybody, like, you know, gets to enter the top of the funnel in a structured way, you don't find any, you know, gaps in in access to funding for founders that, you know, come in using using using this channel. And so this is where I think, like, you know, this technology holds the greatest promise. I'm imagining almost a one to one time transfer.
Emmanuel Yimfor:A deal team, like you said, can only process so many incoming pitch decks in a year, and so you have to triage, you know, using some strategies like, you know, is this person from a network? How do we get the recommendation? Who else is backing this GP? How does this GP signals tell me something about whether they're going to be successful, you know, in the future? Now we can actually use these tools to expand the top of the funnel.
Emmanuel Yimfor:And I think with that, an added benefit of that is going to be able to include more minorities and female GPs.
Victoria Sienczewski:Phenomenal. And would love to dive into your advice for the industry, LPs and GPs, as they're experimenting, not necessarily building these solutions in house, but they're investing in technology and implementing it internally. What is your advice to them? Where should they start?
Emmanuel Yimfor:If we start with GPs, right, this is an opportunity for GPs to use this technology to do better search. So a GP previously might think about which kind of LP should I send my pitch materials to based on talking to, you know, other GPs in their network or just casting as wide of a net as possible. Here is an opportunity where GPs can actually, you know, gather data on LP preferences. They're not very subtle about this. They have it on their websites, you know.
Emmanuel Yimfor:And so to the extent that you could leverage the technology to curate the list of potential LPs that you're going to reach out to, then you can reduce time in terms of meeting LPs that, you know, are just not close enough to your strategy and wasting time that way. When GPs are preparing their pitch materials, I'll also urge them to, you know, you know, resist the urge to just use automated materials generated by this system, there's still a need, for example, to be clear about your strategy. There's something that has to come across as being very unique. We definitely don't want go to a system where the GPs are using this technology to generate their pitch decks and the LPs are using the same technology to analyze. The pitch decks as were generated by the GPs using artificial intelligence.
Emmanuel Yimfor:And so allocating capital to its most productive uses means that, you know, you have a strategy that you clearly believe in. Just make sure that you police that whatever materials you're submitting actually reflects, you know, the strategy that you're going to As it may be those GPs, the high grade ideas, but were not very good communicators, so didn't have the resources to hire some law firm or some adviser to prepare their pitch decks for them. Maybe this is an opportunity for them to raise raise their game and clearly communicate some of those ideas. On the LP space, you know, as an LP, if I hire a consultant to help me think about, you know, how to implement, you know, AI and how it fits into my decision making, one of the things I would think hard about is how to expand the top of the funnel, right? The consultant help you think about how have you been getting the existing one pagers and pitch decks that you have been evaluating?
Emmanuel Yimfor:Are there opportunities where you could use this technology to expand the top of the funnel? Are there opportunities for you to be able to, like, you know, make sure that, you know, managers with a similar running a similar strategy are being asked the same question irrespective of their race, irrespective of their gender. Is there a way for you to go directly from GP strategy to, you know, thinking about the potential risks that, you know, are are inherent to the strategy that they're going to execute while minimizing the potential distortions that may occur in the middle in terms of like, you know, where the GP is from or asking them personal information about their network or asking them questions, especially minority GPs or women, that you would not ask other GPs. So this is a chance to really have more of a structured process, expand the top of the funnel to essentially make sure that, you know, we're serving the goal of finance, which is again is allocating capital to its most productive uses. These are really exciting times and so to the extent that the industry, you know, adopts some of this technology in in the right way, not in ways that are related to how past decisions have been made, I could really see a potential where, you know, it not only expands the pool of who is receiving capital, but it also allows this capital to be allocated towards its most productive uses.
Victoria Sienczewski:It is. And I fully agree with you. A really exciting time for the industry. But there are risks. And so it's great to speak with you to uncover some of the ways that we can mitigate them for the industry.
Victoria Sienczewski:And would love to just close with one or two more questions. The first one would again be, maybe not for LPs and GPs, as they are on the receiving end of the technology. But for folks that are working on developing this technology, is there any advice you would give them to try to, again, reduce the risk of bias from past decisions and how we carry that through?
Emmanuel Yimfor:Right. So there are two points about reducing the risk of how we have been allocating capital in the past. Any biases that arise from how we have been allocating capital in the past to be projected into the future. And so the first one is, you know, the type of training data that we are using, right, to train the systems as opposed to, you know, just training data that is based on how we have allocated capital in the past, I will try to expand that pool. So instead of saying, among the managers that I have backed in the past, which managers did better or which managers did worse in economics, that's like the intensive margin.
Emmanuel Yimfor:I would try to expand it to the extensive margin. The manager that you chose not to back, you did not follow-up to know exactly what their performance is. And so I'll try to use this technology to expand the training data that comes into the process in the first place. And then how to use the training data, as in making sure that the output is not just a reflection of what you have done in the past, but is actually more aligned, you know, to your objectives of, you know, what types of GPs you are looking for and, you know, what kind of hurdle rate or minimum level of return you are, you know, you are trying to achieve. And so I would again, like, in in the sense of, like, from an LP's perspective, they could use this technology to meet people that they have never met before.
Emmanuel Yimfor:For the people designing the technology, you know, they they could think about how GPs are going to try to use the technology and how LPs are gonna try to, you know, use the technology ex ante. So GPs are gonna wanna very clearly communicate their strategy and only reach out to the set of LPs that they think have a high likelihood of backing them, LPs are gonna wanna achieve, you know, the highest level of return per, you know, per unit of risk. And so from that perspective, making sure that the systems are designed not just to appeal to what the LP has done in the past, but to appeal to their broader objectives is going to be something that's going to serve both sides.
Victoria Sienczewski:And we're at an incredibly exciting transformation point for the industry from an AI perspective, and your research is at the center of it. What are you researching right now, and what is on your roadmap?
Emmanuel Yimfor:Oh, great question. So, right now I'm really obsessed with understanding how artificial intelligence is going to, you know, change the equilibrium dynamics in the industry. So what do I mean exactly by that? On the one hand, you can imagine that, you know, the industry could go one way, which is expanding, you know, the top of the funnel like we discussed, which would mean that you would see more traditionally underrepresented groups raising capital in this space. And then it will be interesting to find out if, like, you know, how these new GPs that are able to enter the market by virtue of LPs expanding their search space, how they perform.
Emmanuel Yimfor:Did they actually deliver on the promises, know, did they actually, you know, deliver on the confidence that was placed in them by these new new LPs. On the other hand is if the industry goes the other way, which is, you know, using, you know, past data or past embeddings of other GPs they have backed in the past and networks become even more narrow, then, you know, we should see even widening disparities in in the allocation of capital in this space and so I'm very closely tracking where we're going to fall. I'm going to be using big data to understand, you know, where we are going to fall on this spectrum moving forward in the future. On I'm gonna be doing this on the GP space, and so GP is raising capital from LPs. I'm also gonna be thinking about this when we look at, you know, founders raising capital directly from GP, so venture capital by founders.
Emmanuel Yimfor:And so I'm pretty excited to see how, you know, the industry has changed with the entry of this new technology.
Victoria Sienczewski:Incredible. Very excited to keep tracking and reading your research. And I'm really hoping we'll end up in the first scenario where LPs widen their top of funnel and we reduce bias across the LP and GP capital raising and matching. Professor Yimfor, thank you so much for taking the time to join me today. It was an absolute pleasure.
Emmanuel Yimfor:Thank you so much. Thank you so much for having me.
Victoria Sienczewski:That's a wrap for this episode of The Agentic Allocator. If today's conversation gave you a clearer vision of where the industry is headed or helped you pinpoint exactly where your own process is stuck, go ahead and follow or subscribe wherever you get your podcasts. Your firm's judgment took years to build. Stop letting it reset every time a team member walks out the door. Visit auumai.com to discover the structured digital intelligence foundation your team needs to leverage Agentic AI to its fullest potential.
Victoria Sienczewski:Until next time.