TRAC has developed an AI model that predicts which startups are most likely to become unicorns. The company has updated its list to reveal the 30 new startups for 2025 identified by the model. TRAC states that there is a 1 in 5 chance that an identified company will become a unicorn. unicorn.
Venture capitalists invest in technology, but they have traditionally selected early-stage investments that are decidedly low-tech, based primarily on intuition, the founder’s background, and personal relationships.
TRAC, an early-stage venture firm based in San Francisco co-founded in 2020 by Fred Campbell, Joseph Aaron, Scott Pine, Steve Marek, and Dick Fredericks, wants to change that. That’s what I think.
The company has developed a proprietary model that uses AI to predict which early-stage startups are most likely to become unicorns (companies valued at $1 billion or more). In 2023, for the first time, TRAC revealed exclusively to Insider the 30 startups its model identified, along with its methodology. This year, the company agreed to provide an updated list.
There are some surprises about TRAC’s model. The model is based on more than 30 public and private data sources, which Aaron calls “venture capital’s moneyball.”
First, the company says it’s far more effective to focus on which startups are less likely to succeed than to pick winners.
“Our algorithm isn’t really picking the needle in the haystack; it’s removing all the hay,” Aaron explained. “Our AI eliminates approximately 99% of all early-stage companies from consideration, because our data predicts that these companies are more likely to fail.”
Another surprising thing about TRAC’s model is that it does not evaluate founders as predictive. Instead, the 291,000 investors in its database, especially the only 247 top angel investors and companies the company calls “superforecasters,” are far more useful in determining a startup’s success. I found out something.
“These extraordinary investors are making money on two-thirds of their positions, and one in five investments has a return of more than 10x,” Aaron explained.
TRAC declined to share the full list of SuperForecasters, but shared a random sample of 30 of them with BI last year.
Since less than 2% of all startups attract SuperForecaster, more than 98% of all startups are excluded from the TRAC formula.
How accurate is the TRAC formula? Venture investments, like all early-stage investing, are typically judged after a decade or more, so it can take a long time to know who’s really good at the job. Masu.
The company says there is a 1 in 5 chance that a company it identifies will become a unicorn, and that it is particularly good at eliminating false positives, or failed investments.
“Most early-stage companies fail within 18 months of raising capital,” Aaron said. “An early-stage VC of a similar vintage would have had more than 20% of their portfolio with false positives in their first few years.”
In the 2023 list, some listed companies have already achieved or are close to unicorn status. Legal AI startup Harvey AI was valued at $1.5 billion in 2024. AI startup Hebbia raised $130 million in 2024 at a $700 million valuation.
TRAC has made 61 seed investments to date, and claims that none of them suffered a loss. ”
“This is the only statistic we’re proud of,” Aaron said.
This is the latest alphabetical list of 30 companies identified as the next unicorns in the 2025 TRAC model. All companies are valued at less than $270 million. The average valuation is $149 million.