- Key insight: Citi Ventures is investing in AI companies that automate lending, coding and manual work of all kinds.
- Expert quote: "In the old world, lending and underwriting used to be driven by rules. You would set and follow rules, and then you say, does the client have a certain cash-flow profile or certain size, or does the customer have a certain FICO score, and then you ended up providing credit based on that. But in the new world, I think it's going to be a combination of some rules being applied, but then some artificial intelligence being brought in." —Arvind Purushotham, head of Citi Ventures
Talk of an AI bubble doesn't disturb Arvind Purushotham, head of Citi's venture capital arm Citi Ventures. That's because he takes a long-term view, focuses on fundamentals and invests strategically in technologies Citi itself is interested in using, he said in a recent interview.
"What does the company do? Is it adding real value? Is there a business model that's sustainable?" Purushotham told American Banker. "It sounds a little bit like motherhood and apple pie, but it's actually very hard to do because when you're in the middle of such a massive trend and shift, and there's so much happening in that world, sometimes those fundamentals can get lost."
Citi Ventures has made more than 200 investments since its founding in 2010, and it currently has 125 active portfolio companies, including non-AI companies. In the interview, which has been edited for length and clarity, Purushotham explained how he assesses companies, the AI use cases he thinks have the greatest potential and why robotics process automation never took off.
You recently recommended on LinkedIn an article on computer use agents, which some people view as the next generation of robotic process automation, or RPA — software that uses simple, non-AI "bots" to mimic human actions and automate repetitive, rule-based tasks. I remember writing about RPA when it was considered the next big thing, and now I hardly ever even see that term used. Do you think computer use agents are the next big thing?
ARVIND PURUSHOTHAM: This is what a lot of venture investors and enterprise CIOs and architects are thinking about. If you step back and think about why RPA was interesting to enterprises in general, and then financial services institutions in particular, it's the amount of knowledge work that happens where there are people sitting at desktops, doing maker-checking and all kinds of manual knowledge work. This technology becomes very interesting when there's a lot of people doing manual knowledge work, and that's true in financial services, insurance and other industries where RPA could bring productivity gains to the enterprise.
Why do you think RPA never took off?
The feedback that we have gotten is that it was hard to make it work sustainably and consistently inside the enterprise. It turned out to be brittle because enterprise environments are very dynamic; they keep changing, so you need to have systems that can adapt to a changing environment. The software ought to be able to handle those changes automatically. Otherwise, you have these agents going down, and then they have to be maintained, and it creates additional work for the company. From an industry standpoint, RPA 1.0 had some challenges.
But now when you come to 2026 and you think about these reasoning agents, where you can give them a task, and then the agent can plan it, and it can actually show you the planning, and then it can go and execute on the individual steps that it has identified, agents are coming to a point where they can handle some very sophisticated tasks. We can see it in our personal work. We can see it in our enterprise as well.
The question is, can these computer use agents bring substantial productivity gains and work reliably within the enterprise? Citi has been working on agents. And with the advent of the newest models, there is an opportunity to bring those kinds of productivity gains to the enterprise for broad knowledge work.
A lot of the initial productivity gains you're seeing are from coding.
Citi Ventures has invested in AI-based lending software providers. What attracts you to that category? Are you seeing an increase in demand among lenders for more AI underwriting platforms?
Yes. Large banks have consumer lending, where we might be doing credit cards or personal loans. We do secured lending of various kinds like mortgages, lending secured by securities, and then we have all kinds of lending that we do institutionally with investor and corporate clients. When people say AI-driven lending or AI in lending, they immediately think about the underwriting piece. Now, underwriting is only one piece of it. There's a huge amount of process involved from the soup to nuts of lending. If it's a new client, you have to actually start to onboard the client. You have to collect the information. You have to make sure the applications are complete and compliant with whatever your internal policies are and regulations. Then it goes through the underwriting process, and then beyond the underwriting process, the loan has to be serviced, and there has to be tax reporting around it, and then eventually the loan comes to an end because it's paid off. So there's a huge amount of work that happens around the core of the underwriting. And this includes, for example, detecting application fraud. This involves doing KYC due diligence on a potential client. This involves the servicing piece of it, which sometimes can be quite operationally intense. For a lot of these institutional loans, it takes a lot of manual work to service those loans and provide a great customer experience for clients.
AI can be used for every single one of those steps. AI can be used for fraud detection. AI can be used for KYC and customer due diligence. AI can be used for servicing and for customer support. Citi Ventures has invested across the board in many of those steps, and it's an important area just because of how big lending is as a part of our business, as a part of any bank's business.
Now, specifically when it comes to underwriting, here's the crux of how I think about it. In the old world, lending and underwriting used to be driven by rules. You would set and follow rules, and then you say, does the client have a certain cash-flow profile or certain size, or does the customer have a certain FICO score, and then you ended up providing credit based on that. But in the new world, I think it's going to be a combination of some rules being applied, but then some additional intelligence being brought in through either machine learning models, which is the old AI, and some of the newer LLMs. You can bring those in to do underwriting itself, based on more real-time information. It could make some subjective decisions based on the text that you have in a loan application. It can potentially provide credit access to more customers than you previously did in a way that is completely compliant from a regulatory standpoint, but also adhering to your own standards of underwriting.
In consumer lending, you can feed a person's utility payment data, rent payment data and other signals of creditworthiness into an AI-based underwriting model, even if they don't have a strong credit score or credit report. What's the corollary with corporate lending? Is it cash flow data or background information about the company?
It's being able to analyze more real-time information about the company. AI is able to detect patterns that sometimes humans can't. But you have to do it in a responsible way too. As a bank and as a lender, you can never go away from putting responsible lending and responsible AI right up front. If you think about the adoption of AI by financial services, it takes a little bit longer than let's say a startup or a tech company because we're regulated, and we just need to be responsible about how we deploy it, both internally or even externally, client-facing.
You talked about AI-based lending and coding as strong use cases. Are there any other use cases for AI in finance that you're really bullish on right now?
I talked about coding, and I talked about general knowledge work. Coding is an important piece of it, given how much of the work being done in any industry, not just in financial services, is becoming more driven by software. The more something is driven by software, the more a coding agent brings you that productivity gain. So it's hard to overstate the impact of what those coding capabilities mean, and it's not just about productivity gain. It also accelerates innovation. It also means that you can introduce new products faster. You can introduce new experiences faster. You can drive up customer net promoter scores because you're able to provide a better experience. So it's not just about productivity gain because you're eliminating some manual work. It's also about being able to innovate faster.
But the aspect of general knowledge work is where we're yet to see those great examples in financial services. That's where Citi is working hard on adopting AI technology, and the same is true for other industries.
Some people argue that there's a bubble in AI investing that could burst, causing various market shocks. What is your feeling about that?
People compare it to — pick your favorite cycle that we have all gone through. Is it like the internet of the late 1990s? When we're thinking about how we invest in these companies at what valuations and what kind of returns are we looking for, number one, we try to take a long-term perspective because how people are feeling about valuations literally changes from week to week. That means that we're looking at the fundamentals. What does the company do? Is it adding real value? Is there a business model that's sustainable? It sounds a little bit like motherhood and apple pie, but it's actually very hard to do because when you're in the middle of such a massive trend and shift, and there's so much happening in that world, sometimes those fundamentals can get lost.
Read more:
As fintechs nab bank charters, what's happening to BaaS? Overheated data center knocked 23 credit unions offline Chris Britt still says banking is broken, but Chime can fix it Credit unions bulk up a business that banks long dominated
Number two, we're always looking at how it's relevant to Citi. We're a strategic investing group, and that aspect of being beneficial to our business, being beneficial to our tech or to our functional groups is an important piece of how we evaluate that, and that's an additional area of diligence for us. That gives us the sense of whether something is sustainable or not. If a company like Citi finds something strategic, that means it's a good sign that it's a sustainable capability, and there's a business model.
The third piece of it is, what is the valuation piece? Is there a potential for return? Here, it's more art than science in venture in general, especially in the earlier stages of venture. The path we try to take there is to make sure that we're investing in the best teams, the teams that can over time build a great company. What is different about this period that we're seeing with all these AI companies is the revenue growth. Some of them have incredible revenue growth rates. There are some of those in our own portfolio. There are many, obviously, in the private-company ecosystem where revenues are growing rapidly. Revenue growth is not the end of it. Obviously, you have to look at gross margins and all these other aspects.
