the future of capital markets and artificial intelligence
for the last four and a half years, i have worked in debt capital markets building data analytics tools for consumer credit issuers and investors (including consumer unsecured loans and mortgages). i have gotten exposure to the largest credit firms in the industry, from originators, to asset management firms, to securitizations issuers. lately, you hear a lot about "private credit" being the hot thing because of where rates are and changing capital requirements for banks, creating opportunities for new entrants in the market. exposure to their workflows, combined with my interest in technology, has made me think a lot about how software and artificial intelligence are going to impact capital markets over the next decade or two.
a common impact discussed with artificial intelligence is the trend towards leaner software companies. llm coding assistants and agents will allow companies to employ fewer people, and folks like sam altman have spoken about how a "one person software company with $100m in revenue" will be possible with the large language models. maybe that is an exaggeration, but i definitely think software companies a decade from now will be significantly leaner, and ai tools will improve the margins even further. i think you are going to see something similar credit markets, where the asset management and investment firms that embrace technology will operate with better agility than those who do not embrace the power of software, data engineering, and large language models.
i've seen a thesis that ai agents will eventually replace a lot of junior analyst level work. this is probably true. on the more extreme end, though, i've seen claims that ai agents will make investment decisions. this is maybe true. certainly, ai agents will propose investment decisions. i think its debatable whether those investment decisions will be better than average, which is really what being a good investor is about. the mistakes i think a lot of technologists have when thinking about financial markets are 1) underestimating how deeply adversarial it is, 2) not realizing how difficult it is to actually outperform, and 3) misunderstanding what work actually goes into making an investment decision. from talking to people in asset management and investing, the only two things that really matter in finance are 1) is my insight better than the market and 2) how quickly can i execute on my insights.
for ai to really impact capital markets, it has to help at least one of those two things: ai either has to help you generate more unique insights that are good (by good, i mean leading to better returns) or execute transactions and deals more efficiently. you need to understand how ai can enable those two things.
ai to generate unique insights
generating unique insights, that are correct, in financial markets is really difficult. with the growth of retail trading in equities, people think it is easy, but that belies the scale at which a lot of firms, including credit firms, need to operate and compete on. you are managing hundreds of millions of dollars, if not billions, on behalf of very powerful institutions and people in an environment where there are dozens of well funded, intelligent, and capable competitors. these competitors are not some distant organization- they are competing directly for the same limited partners, talent, and deals that you are.
in credit, i've found there are three drivers for unique insights: 1) relationships- you build relationships with originators that give you insight into their collateral and business, which opens up investment opportunities, 2) economic conditions- you have some thesis on how economic conditions will change and how that change will impact prepayments and defaults on the credit you are investing in, and 3) in depth, large scale data analysis of credit performance that allows you to predict loan performance better than your competitors.
artificial intelligence will probably impact relationship management and development to some degree (emails will be written using large language models, investor relations and reports materials will be ai generated, etc), but i don't think there will be a significant impact in that realm. relationships will always be personal, and ultimately the "dealmakers" will want to know the person they are sitting across. i don't expect ai to significantly impact the way a credit firm operates or is structured in this regard; firms will still require senior people to go network, build relationships, and identify opportunities for business.
analyzing economic conditions is about how much information you can process. ai will have a big impact here. right now, analysts are flooded with information; if you've ever seen a finance person in action, they usually have bloomberg or cnbc on the tv, they're looking at the bloomberg terminal and getting news notifications live, they are reading in depth research reports, and they are responding to emails and messages about their market. i think in the future, large language models will process all of this information, summarize it, and dynamically present it to an analyst depending on what is relevant when. analysts will then be able to create more streamlined pictures of how they expect economic conditions to trend.
the biggest impact, however, will be on large scale data analytics to generate unique insights. this is also what i am most familiar with in my role at dv01, since we build the tools that are used for this right now. i've previously written about generative experiences within consumer software, but the same will apply to financial services. generative ai is going to transform the application layer, and that will apply to the applications that credit analysts use. the first component relevant to analysts is the data they need to identify opportunities. in credit markets, this is often loan data, which includes the entire payment history of a collection of loans and some additional characteristics about the borrower.
data comes from different sources and different formats; my current company builds data pipelines that standardize and normalize loan data. i believe that in the long run, however, analysts will be able to use large language models to build their own data pipelines to process data. more and more data processing is moving to sql, instead of technologies like spark, and large language models are becoming more and more proficient at translating natural language into sql. as large language models become more precise and capable of identifying the necessary context without human input, they will be able to generate the necessary sql to transform data for credit analysts.
the last step will be abstracting away the infrastructure that data pipelines need to run. this is not an easy step. i think the solution will looks similar to databricks and the data lake architecture, where each organization or desk has a massive data warehouse that holds everything they need, from unstructured to structured data. the analyst's interaction will be abstracted behind natural language; they will describe the data they want to access and how they want to process it through some agentic large language model system that lives on top of their warehouse.
the second component of applications for analysts, beyond the data, is well, the actual application. this includes tables, visualizations, and aggregations calculated using the loan data supplied. generative ui will completely take over the analytics workflows for analysts, where long lived applications will have custom charts and tables constructed by analysts using natural language. these applications will live on top of the data pipelines and transformations that the analyst sets up themselves using natural language. the additional necessary component is an adapter between the large language model, the software being generated by the large language model, and some form of financial calculator capable of aggregating data, implementing core financial concepts that the large language model can use via tool calling.
when it comes time to analyze data to generate insights, analysts will be able to quickly query and transform data into their needs and generate real time analytics software on top of it, allowing them to quickly explore investment ideas all through natural language. notably, this dramatically will reduce the operating overhead of an investment firm and allow them to move faster because analysts will be able self-serve analytics for themselves.
ai to execute on insights more efficiently
the second big driver of success in finance is executing a transaction and insight faster than your competitors. usually, there is a pretty limited window before other's notice your insight and it becomes priced in. that means the speed at which you can move is critical. this is another area where i believe artificial intelligence will be transformational to how firms operate.
to execute on an insight, there is a natural split between front office execution and back office operations. front office execution requires a combination of investment proposal, evaluation, and trade execution. generative ai will allow analysts to prepare necessary materials and reports faster than ever, with natural language. beyond just generating the tables and visualizations, eventually i believe large language models will be instrumental in writing the commentary associated with an investment deal.
pdfs, spreadsheets, and manual steps by operations teams reconciling bids with accounting books and cashflow could be streamlined with agentic information retrieval. there are some more foundational technology improvements that will be made here, but in credit, a lot of the loan level diligence as a part of a credit transaction will be automated, including document review and compliance checks. this means that funds will flow faster, meaning cash can be deployed more quickly into return generating assets.
there are a ton of repetitive back office tasks that i think large language models will automate, though they will need a set of quantitative tools to do so in order to reliably tie out numbers, reconcile accounts, and ensure compliance. a lot of operations work is stitching together data and communication between necessary parties within a firm, and i think a lot of this communication could be automated; cross referencing transactions for compliance with investment mandates, identifying potential trade compliance issues in communication, aml/kyc operations, sharing transaction information with accountants, banks, and lawyers are all things that ai agents could take over in the future.
by and large, i think there is a significant operational improvement that financial firms will benefit from if they can effectively introduce large language models and ai into the heart of their operations.
what it all amounts to
so where does all of this lead us? what does an asset management firm look like in 2035?
i think the best firms will be even leaner, less reliant on manual work from humans, and execute faster. technology first firms, who rethink their entire operations from front to back to leverage artificial intelligence will have an edge not because ai systems are making investment decisions, but because they allow people to think and execute faster. agents capable of multi-step data retrieval, aggregation, and interpretation lend themselves well to the kinds of analysis and operations that asset management firms have to do. sifting through and processing large amounts of low quality and disparate data, quickly aggregating it into easy to understand packages, and then building assumptions based off how you think the market will perform will be easier and faster with large language models and agentic systems that act as an extension of an analyst's brain.