This content is from: Corner Office
Short Sellers and Hedge Funds Sign On to Fintech Company Whose Algos Sniff Out Corporate Fraud
Start-up accelerator Y Combinator is backing Toronto’s Bedrock AI, which uses artificial intelligence to deliver an edge to investors.
Short sellers and hedge funds haven’t wasted any time signing up for a new source of information on corporate fraud from a young Canadian financial technology company. In fact, Toronto-based Bedrock AI, which uses artificial intelligence to identify corporate fraud, has just received backing from Y Combinator, a start-up accelerator firm that has invested in such tech giants as Airbnb, payments company Stripe, and DoorDash.
Bedrock AI officially launched in April and has already lined up institutional hedge fund clients managing a total of more than $25 billion in assets under management.
“What we do is find impactful information in Securities and Exchange Commission filings that our models have learned are associated with downside outcomes specifically earnings quality risk, governance risk, and malfeasance,” Kris Bennatti, a CPA and programmer who founded the company, told Institutional Investor.
Clients pay annual subscription fees to access the firm’s research using a web-based dashboard.
Prior to starting the firm, 30-year old Bennatti worked for two years as a data scientist at Hansell McLaughlin Advisory Group in Toronto. Earlier in her career, she was a senior accountant at KPMG Canada.
Being both a programmer and an auditor gave her the insight that led to the creation of Bedrock AI.
“There is a big gap between the people who prepare corporate disclosure and the people consuming it,” she said. “The people preparing it have no incentive to make it clear and easy to read,” which means that while technically there is a lot of public information about companies, “little of it is being used.”
“People can’t find the relevant information. They don’t have time to read through a 10K.”
Sensing the tech world was missing an “enormous opportunity” to address the problem, she quit her job and in November of 2019 started building Bedrock AI. She was soon joined by Suhas Pai, a natural language processing researcher, who is the firm’s chief technology officer. A third partner, Heather Judd, came on in March as the chief operating officer.
Bedrock’s model has processed over 80,000 securities filings, dating back to 2019, and it is updating with new ones as they are submitted, according to Bennatti. The model looks at quarterly and annual filings, as well as 8-Ks, reports of unscheduled material events or other corporate changes.
Bedrock AI also is engaged in pilot projects with four securities regulators.
It is a particularly useful tool for short sellers, said Chris Drose, a Bedrock AI client, who is launching a new activist short hedge fund called Bleecker Street Capital.
“The most important information is usually buried in the hardest-to-reach places,” he said. “Bedrock AI has done an amazing job highlighting the disclosures and bits of information that companies would rather stay buried deep in their filings.”
Bedrock AI has done little marketing and its clients have mostly learned about the firm through Twitter or “word of mouth,” said Bennatti.
The start-up’s current clients are primarily long-short equity hedge funds, but she has recently received requests from pension funds as well.
Even if these investors aren’t engaged in shorting stocks, they care about the harm to their reputations that could come from having a large investment in a company that is clearly a fraud, said Bennatti.
Bedrock AI got a boost of public recognition last December when tech entrepreneur Alex Danco interviewed Bennatti on the topic of fraud for his popular Substack financial newsletter.
Danco, a veteran of Chamath Palihapitiya’s Social Capital, now works for Shopify in addition to writing the newsletter.
Bedrock AI began discussions with Y Combinator in May, a month after its product went live. The accelerator has invested $125,000 in the firm for a 7 percent stake. And while that amount is small, it’s a stamp of approval that should make raising seed capital easier, Bennatti said.
The company's product is different from other services that track securities filings, such as Michelle Leder’s Footnoted. Leder told II her product uses a combination of human and machine learning, with more than 300 filters scanning SEC filings.
She welcomes the newcomer.
“I’ve been helping Kris and Heather on an informal basis, trying to beta test their product,” she said. “I don’t really see it as a competitor to footnoted and, in theory, could help me get to more filings.”
“I will never say that AI outperforms human beings,” Bennatti said. “But one of the things AI does really well is quantifying risk in multidimensional space.”
For example, she noted that Bedrock’s model found that 10 different risk factors were disclosed in Credit Suisse’s 2019 financials. While that might not have predicted the Archegos and Greensill disasters, it did reveal that “Credit Suisse is much higher risk than their competitors,” she explained.
Bedrock AI's model also has revealed problems at companies currently under investigation or targeted by short sellers, such as Canoo, EHang Holdings, an autonomous aerial vehicle company, Life MD, and Lordstown Motors Corp.