Amnon Shashua, founder and CEO of Mobileye, is attentive to complex problems that he believes can be solved with AI, and that AI itself can be solved to become more reliable. Alongside creating and running his self-driving car technology company – which he took public, then sold to Intel, then spun off again – he has hatched a number of other ideas.
Now, one of them is to raise money and build significant momentum.
One Zero, a fintech aimed at using AI in retail banking, is raising at least $100 million, TechCrunch has learned.
Despite being co-founded by one of Israel’s most prominent and successful founders, One Zero has received surprisingly little attention until now outside of its home market. But the company has raised about $242 million so far, and in 2023 it was valued at $320 million, according to PitchBook data. Our sources say the valuation will be significantly higher in the next round.
It’s unclear who the investors are, but the company’s previous backers include Tencent, OurCrowd, and SBI Ventures (the now-independent company that was formerly part of SoftBank).
One Zero’s momentum comes amid a frenetic pace of activity for Shashua, who serves in a non-executive role at the company, with Gal Bar Dea as CEO. Over the past few years, Shashua has founded or co-founded startups working on humanoid robotics (Mentee); alternative approaches to large language models for generative AI (AI21); and, launched just a few weeks ago, AA-I Technologies (pronounced “double AI”), which Shashua describes as his effort to build an “AI scientist.” He is also a professor of computer science at the Hebrew University of Jerusalem.
One Zero’s equally ambitious mission is to “make private banking accessible to as many people as possible,” he said in an interview. It’s about democratizing the kind of high-touch advisory services that wealthy individuals get when banking, in a market where not only does the average person not get that kind of service today, but is also able to do so. looking at a future where there may not be a physical bank, nor humans to help.
It meets this ambition through a dual commercial orientation. In Israel, where OneZero is based, the startup acquired a banking license and built a full-stack retail bank. At the same time, One Zero uses the knowledge gained from this retail sector – which Shashua described in an interview as a “sandbox” – to train its models and refine its technology in order to license this technology to banks operating elsewhere.
Retail now has about 110,000 customers, Shashua told TechCrunch, and while it has yet to announce any licensing deals, the company says it has received a number of inbound inquiries from major banks as of this year. sense.
So far, the company’s cornerstone — and what it plans to invest its funding on — is a chatbot called Ella, which aims to be better than current chatbots while providing services that human bankers could not.
According to Shashua, while there have been many efforts to integrate AI into retail banking, for example around functions such as expense management, their possibilities are limited.
“We don’t see banks deploying artificial intelligence to a level where they replace a banker,” he said.
As an example, he said, take automated communication. You can ask a banking chatbot very basic questions, such as “how much money is in my account?” “, or information about recent transactions, and he can usually respond. But it’s a different story if you’re asking something with calculations, such as “how much money will I have in my deposit account at the end of the year, based on activity so far?” ”, or “what is the best way for me to purchase.” a car based on my financial profile? Not only are chatbots unable to answer such questions, but most personal bankers cannot either.
“There is an opportunity here, where generative AI can apparently do this,” he said. “This goes way beyond tracking expenses.”
One Zero’s approach to building such AI, as Shashua described it, is very ambitious and seems as tricky as autonomous driving. It focuses on the use of several major language models. Some models may be optimized for different tasks, he said, but running tasks through multiple LLMs can also provide a diversity of answers, which are then put through a verification process to understand when the answers are misleading or erroneous.
And if those answers don’t turn out to be helpful or correct, the AI doesn’t try to say anything anyway, he said. “It’s normal (for him) to say I can’t solve your problem. I can’t answer your question,” he said. “Humans can’t answer every question either, right? So that’s okay. It’s not okay to say , here is an answer to your question, and the answer is completely wrong, completely wrong.
The system starts with more basic tasks such as expense management and the plan is to add more features over time to help advise customers on financing large purchases or saving smarter.