By Alex Johnson
Kasisto, maker of banking chatbots, is launching its own large language model:
Kasisto today announced the launch of KAI-GPT, the world’s first banking-specific large language model (LLM), designed to address the industry’s unique needs for accuracy, transparency, trustworthiness and customization. Powered by KAI-GPT, KAI Answers is Kasisto’s first generative AI application to use the LLM. For bank employees on the front line of customer care, it provides the right answers quickly, via a contextual, human-like conversational experience.
It’s not just pre-seed startups that are repositioning themselves to capture some of the reflected allure of generative AI, it’s everyone.
[Ross Geller voice]: PIVOT!
Kasisto was a logical company to try and claim the mantel of “world’s first banking-specific large language model,” as it has been working with banks to launch AI-powered chatbots and virtual assistants for a while now.
A couple of interesting things about KAI-GPT:
- The first application being built on KAI-GPT is KAI Answers, which is essentially a natural-language interface built on top of banks’ employee knowledge management systems. Choosing to go this route rather than building a better version of the customer-facing chatbots that Kasisto is better known for suggests to me that it didn’t feel like the tech was ready for customer-facing use cases.
- Kasisto claims that by training its own Generative Pre-trained Transformer model exclusively on banking data and for banking use cases, it can increase the accuracy of the model’s responses. Consider me mildly skeptical. The essential nature of generative AI is that it’s really good at guessing what a response to a specific prompt would be likely to sound like. A little hallucination is par for the course. I’m guessing this concern is one of the factors that pushed Kasisto to focus on an employee-facing use case first.
- It’s not clear from the press release exactly what data KAI-GPT was trained on (it mentions that it is focused on being “trusted by financial institutions for how their proprietary data is used”), but this seems like it will be a significant source of questions moving forward – what data sets are these models trained on? How effectively can these models be refined and customized using a specific company’s proprietary data? How tempting will it be for larger companies to build their own LLMs rather than build on top of general ones like OpenAI or industry-specific ones like Kasisto?
#2: Stop Misusing Credit Scoring as a Wedge
Another fintech company focused on improving credit scores through the reporting of rental data just raised a seed round:
Rent reporting platform Boom has raised $4.5 million in a seed funding round joined by Plaid co-founders William Hockey and Zach Perret.
Starting Line led the round, with participation from Clocktower Ventures, Company Ventures and Gilgamesh Ventures.
Launched in late 2021, the Boom allows renters in the US to build credit using their rent payment.
The company reports to all three credit bureaus – Experian, Equifax, and TransUnion – and claims that users have seen an average increase of 28 points in their credit scores within just two weeks of using the app.
We now have a well-established playbook in this area. It goes something like this:
- Build a small B2C business focused on helping renters improve their credit scores by reporting their on-time rent payments to the credit bureaus.
- Leverage your traction on the consumer side to sign up property management companies as customers or partners (help your tenants improve their credit scores!)
- Build out a broader suite of solutions for both renters and property management companies.
It sounds like Boom is well down the path with this playbook already:
Boom says it has built a strong subscriber base for its $2-a-month offering over the last year, with a growth rate of over 450%. It is approaching $1 million in revenue run rate an inked partnerships with industry players such as Progressive, Apartment List, and national property management companies.
Rob Whiting, CEO, Boom, says: “We invested heavily into building our end-to-end software solution for rent reporting and consumer app layer and now we’re focused on revolutionizing the entire Renter Experience, from both the renter and the apartment operator perspective.”
As a company-building strategy, this is probably smart.
But I don’t like it.
This strategy relies on exploiting consumers’ desire to improve their credit scores, which, in the short term, can be used to generate revenue (Boom costs $2 per month + a $10 enrollment fee) and, in the long term, can be used to attract property management companies and other commercial partners.
I use the term “exploiting” intentionally because none of the fintech companies using this playbook seem to understand or be willing to acknowledge that the first version of the FICO Score that takes rental data into consideration is FICO 9, but the vast majority of lenders use FICO 8.
The promise of access to credit is a powerful B2C wedge. I would like to see it used more responsibly.
#3: What Do We Want?
It’s proving difficult for Apple savings account customers to get their money back out:
Min-Jae Lee was curious to try out the Apple account and intrigued by its high interest rate. She deposited $100,000 in April, but soon decided she would rather have her money elsewhere. On May 1, she tried to transfer it out.
It took more than three weeks for her to get it.
Lee, a lawyer, said Goldman told her to contact JPMorgan Chase, where she was trying to move the money. Then, on Goldman’s advice, she tried sending the money to her Vanguard account. The $100,000 moved there before going back to Apple the same day, she said.
Goldman then called her, she said, and told her that the money could be transferred only to the account from where she had sent it. She initiated a transfer to Ally on May 16.
A few days later, Goldman told her that her account was under a security review.
Her Apple account showed a zero balance, but the money wasn’t in her Ally account either. It finally showed up there on May 25. She isn’t entirely sure why.
The specific issue here seems to be that Goldman Sachs’ compliance and fraud management policies for newly-opened accounts are inhibiting customers’ ability to shuttle their money around. You can transfer money into your Apple Savings account, and you can move money back out into that same external account, but anything else is triggering delays and account security reviews.
More broadly, I think this is a great example of colliding regulatory priorities.
What do we want?
- An open ecosystem where consumers can quickly and easily move their money around to the best providers!
- A stable banking system where deposits are sticky and banks feel confident to make fixed, long-term loans!
- A flourishing competitive environment!
- A banking-as-a-service system that is well-regulated and not overrun with risk!
- A limit on the power of large tech companies!
When do we want all of this?
2 FINTECH CONTENT RECOMMENDATIONS
#1: The Impact Of ChatGPT And Open Banking Cannot Be Underestimated (by David Birch, Forbes) 📚
This a good reminder that many of the concerns about the combination of generative AI and financial services (including from me!) may be missing the forest for the trees. In this article, David uses the early days of aviation as an analogy for how AI-powered banking may evolve in ways that are difficult to predict.
#2: Capital One: Buffett’s Latest Banking Pick (by Marc Rubinstein, Net Interest) 📚
Y’all should be reading more of Marc’s stuff. This is a good example of why.
1 QUESTION TO PONDER
The recent consolidation of the BaaS platform space has me wondering – which companies will be the big winners from the coming fintech infrastructure shakeout?
Bonus points if you suggest companies other than FIS, Fiserv, and Jack Henry.