Can AI Predict the Next Big IPO? Crunchbase Thinks So.

Crunchbase, the firm best known for its startup financing data, is using artificial intelligence to predict when those startups will raise funding, get acquired or hit the public markets.

But in an era where chatbots are taking over traditional web and data searches, its biggest challenge is predicting its own future.

The startup was founded in 2007 as a crowdsourced database. In 2010 AOL acquired Crunchbase through its purchase of tech blog TechCrunch, but in 2015 it spun it out.

Yet since launching its premium database services roughly a decade ago, Crunchbase hasn’t dramatically innovated on its core product: a comprehensive data set of startup financing that has served as a go-to for startup founders, venture investors and Silicon Valley watchers.

---

---

That’s changing with the launch of its AI-based prediction engine Wednesday, which uses Crunchbase’s 17 years of startup data to predict where Silicon Valley’s hottest startups are heading, the firm’s chief executive, Jager McConnell, said in an interview.

A reckoning

Crunchbase’s new direction was one of necessity—when OpenAI’s ChatGPT launched in late 2022, the moment sparked a crisis: What is the role of a database company like Crunchbase when chatbots can trawl the web for similar data and provide similar answers?

“If you deal with historical facts, once the AI absorbs it, it doesn’t need you anymore,” McConnell said. “Historical data companies are already dead.”

Like other startups that have been challenged by the emergence of generative AI, Crunchbase needed to pivot. At a leadership meeting convened to discuss its future, McConnell said the team decided it could make predictions using its most valuable asset: the proprietary data generated by the startup’s 80 million users.

That data isn’t the public information on a company’s profile. It’s all of the data Crunchbase doesn’t expose to its users, such as when company profiles have been edited, who edits them and what they’re editing. 

If a startup’s employee makes edits to a company’s profile and is searching for investor profiles, plus there’s a spike in investor interest in that startup’s profile, those are the kinds of signals Crunchbase’s AI uses to indicate the startup is about to raise funding, McConnell said.

Each startup has thousands of potential signals like those pointing to whether they might be about to fundraise, be acquired or make a play for an initial public offering, McConnell said. Public AI platforms don’t have access to that customer usage data, he added, making it the “crown jewels” of Crunchbase’s assets.

Peering into the crystal ball

Based on Crunchbase’s own testing, its fundraising predictions are up to 95% accurate.

Last year, the company predicted that AI startup Anthropic had a 74% probability of raising cash. Anthropic raised $2 billion in January. Last October, Crunchbase predicted that the startup Coda, maker of a productivity platform, would be acquired with a 93% probability. The company was bought by Grammarly, maker of an AI-based writing assistant, two months later.

What’s much harder to predict is when startups shut down, McConnell said. The company’s accuracy rate for startups that fold is below 50%, he said, because companies can survive for a long time, even with slowed growth.

The company uses open-source machine learning tools to train its AI prediction models, as well as OpenAI’s AI assistant and tools to help create content and process data, McConnell said.

News Corp, owner of The Wall Street Journal, has a content-licensing partnership with OpenAI. 

Crunchbase’s own pivot, which began about two years ago, included some bloodshed. It laid off one-third of its 200-some employees, mostly in sales and marketing, in order to reinvest in data science and engineering, McConnell said. Sixty percent of its roughly $50 million in revenue currently comes from premium subscriptions for the company’s data, and the other 40% from access to its behind-the-scenes user data, he said.

Now, with a stronger technical team in place, the company is also aiming to harness AI to make it easier for customers to use its products. Crunchbase’s AI agent, named Scout, is like a chatbot that users can prompt with questions, and helps them search and analyze startup data.

---

---

It’s not clear how useful Crunchbase’s AI-based insights and agent will be. Large venture-capital firms tend to build out their own data science teams, with prediction engines of their own. The goal is for Crunchbase to be a source of data and prediction for them, McConnell said, rather than replace their own work.

There are also other sources of data on startups and financing, including firms like PitchBook and CB Insights, which compete directly with Crunchbase. Like Crunchbase, PitchBook has a tool called VC Exit Predictor, which aims to use AI to forecast which companies will be acquired, go public or shutter.

Trusting what’s ahead

The challenge for startups like Crunchbase is staying ahead of AI-based tools, while the challenge for AI tools is getting access to proprietary data, said Mike Gualtieri, a vice president and principal analyst at Forrester Research.

But new reasoning models like those from OpenAI and its Chinese competitor DeepSeek might be capable of making the kind of predictions that Crunchbase and PitchBook are making, because they can work through more complex problems, Gualtieri said.

“It’s going to come down to, whose predictions do I trust?” Gualtieri added.

That’s a question Crunchbase is grappling with too. Does it trust its own prediction regarding its future?

“It says we’re very likely to get acquired,” McConnell said. “That is a probable outcome for us.”

Write to Belle Lin at belle.lin@wsj.com