Stephanie Song, who used to work at Coinbase on the business development and investments team, got irritated by how many due diligence jobs she and her team had to do every day.
In an email exchange, Song stated that analysts complete hundreds of hours of work at midnight that no one wants to do. “At the same time, funds are putting less money into projects and looking for ways to make their teams more productive while cutting costs.”
Song wanted to find a better way, so he joined forces with Brian Fernandez and Anand Chaturvedi, two former coworkers at Coinbase, to create Dili (not to be confused with the capital of East Timor), a platform that uses AI to try to automate important steps in investment due diligence and portfolio management for private equity and venture capital firms.
So far, Allianz Strategic Investments, Rebel Fund, Singularity Capital, Corenest, Decacorn, Pioneer Fund, NVO Capital, Amino Capital, Rocketship VC, Hi2 Ventures, Gaingels, and Hyper Ventures have given Dili $3.6 million in venture capital. He is a product of Y Combinator.
“AI changes everything in an investment fund,” Song said. “It changes everything from analysts to partners to back-office tasks.” “Fund investment professionals want a unique way to make decisions. They can now use their huge amounts of data to combine what they know about the deal with how it fits into the funds.” In a tough economic situation, Dili has a one-of-a-kind chance to become a place where people can go to get money.
Funds are looking for an edge, and Song is right about that. Investing risk can also be reduced in new and interesting ways. VCs are said to have $311 billion in cash that they haven’t spent yet. Last year, they raised the least amount of money—$67 billion—in seven years as they became less confident in early-stage businesses.
Dili isn’t the first company to use AI in the due diligence process. According to Gartner, early-stage investors and venture capitalists will use AI and data analytics in more than 75% of executive reviews by 2025.
Several companies, both new and old, are already using AI to look through financial documents and huge amounts of data to make market comparisons and reports. These companies include Wokelo (whose clients are private equity and venture capital funds, like Dili’s), Ansarada, AlphaSense, and Thomson Reuters (through its Clear Adverse Media unit).
But Song is adamant that Dili has “first-of-its-kind” tech.
She also said, “We can deliver very high accuracy on certain tasks, like pulling financial metrics from large, unstructured documents.” “To give [our AI] models high-quality context, we’ve built custom indexing and retrieval pipelines that are tuned for certain documents.”
Dili uses GenAI, especially sizable language models like OpenAI’s ChatGPT, to improve the effectiveness of investment processes.
The platform first creates a knowledge base that stores a fund’s past investment decisions and financial data. It then uses the above models to automate tasks like reading private company data databases, managing due diligence request lists, and searching the web for unknown people.
On top of that, Dili now lets you do automatic comparison research and industry benchmarking on a company’s stack of deals. Once funds enter their deal data, they can use one space to compare old and new business possibilities.
Song said, “Imagine getting an email about a new investment opportunity or an update on a portfolio company and having a platform make AI-generated deal red flags, competitive analysis, industry benchmarking, and a preliminary summary or memo based on your fund’s past investing patterns right away.”
It’s important to know if you can trust Dili’s AI, or any AI, to handle your business.
After all, AI isn’t always known for sticking to facts. Fast Company tried ChatGPT’s ability to describe articles and found that it often got things wrong, left out important parts, or made up details that weren’t in the articles it outlined. It’s easy to see how this could become a real issue in due diligence work, where correctness is very important.
When AI makes decisions, it can also bring bias into the process. There was an experiment by Harvard Business Review a few years ago that showed an algorithm that was trained to suggest investments in new businesses would pick white entrepreneurs over people of color and would prefer to invest in businesses with male owners. That’s because the public data that the algorithm was trained on showed that women and founders from minority groups tend to have a harder time getting funding and end up raising less venture capital.
Also, some companies might not feel safe putting their private, personal information through a third-party model.
According to a poll by Bloomberg Law, 30% of deal lawyers said they wouldn’t use AI as it is now at any point in the due diligence process. They said they were worried about things like breaking deal secrecy agreements by putting third-party information into AI software.
To calm people down, Song said that Dili is still tweaking its models—many of which are open source—to cut down on hallucinations and make the system more accurate overall. She also made it clear that private customer data is not used to train Dili’s models. Instead, she said, Dili plans to give funds a way to train their own models using private, offline fund data.
“Hedge funds and public markets have put a lot of money into tech, but private market data has a lot of untapped potential that Dili could help firms find,” Song said.
Dili had its first test run last year with 400 experts and users from different banks and funds. But as the company grows its team and adds new features, it wants to move into new uses. Song says the final goal is to become an “end-to-end” solution for investors through research and portfolio management.
“We think this core technology we’re building can be used in all parts of the asset allocation process in the long run,” she said.
Dili wants AI to do the due research for her.
