12 February 2024 |

AI-Native Loan Origination

By Alex Johnson

Tiger (1912) by Franz Marc.

3 FINTECH NEWS STORIES

#1: A Mesh of Identity Signals 

What happened?

Mesh, a fintech infrastructure startup focused on B2B identity management, raised a seed round:

Mesh, a B2B identity startup, raised a $5.7 million seed round to increase transparency in marketplaces and e-commerce websites, CEO Diego Asenjo tells Axios exclusively.

Mesh on Wednesday rolled out a multi-factor business identity tool for marketplaces, e-commerce sites, and vendor and contractor compliance.

Mesh brings together verified registration, licenses and insurance coverage information, under a single, universal API, Asenjo says. This enables Mesh to verify, onboard and track legitimate businesses on the company’s marketplaces and e-commerce websites. 

So what?     

Lots of companies have been building cool stuff in the B2B identity space. What makes Mesh different?

From what I can tell, three things:

  1. Mesh is focused on helping B2B marketplaces, in industries such as e-commerce, construction, and property management. The focus on marketplaces, which require a persistent view into their business customers, allows Mesh to design its product around a smaller and (arguably) underserved set of use cases.
  2. The product blends a lot of different signals together in order to create a high level of certainty for the platforms using Mesh. These signals include business verification (name, address, tax identification, etc.), license verification (type, number, issuing state, etc.), insurance verification (policy type, number, insurer, etc.), and ongoing business risk monitoring (tax liens, bankruptcies, etc.)
  3. The company’s founding team and executives have extensive experience in the world of online marketplaces (the CEO and CTO both come from Amazon) and identity verification (the CRO and VP of Partnerships have experience at Mitek, BehavioSec, LexisNexis, Socure, and Dun & Bradstreet).   

Mesh feels like one of those ‘unfair advantage’ seed-stage companies that raises a small round and then, within 6 months, is raising a much bigger follow-on and announcing a slew of brand-name clients.

It’ll be an interesting one to watch.

#2: AI-Native Loan Origination

What happened?

A fintech infrastructure provider focused on small business lending raised a pre-seed round:

Cascading AI … is thrilled to announce the successful completion of its $3.9 million pre-seed funding round to unlock $1 trillion in value of advanced AI for the global banking industry.

Casca, the flagship product of Cascading AI, is revolutionizing the lending landscape as the first AI-native Loan Origination System.

“The real game changer is our AI Loan Assistant, Sarah,” says Lukas Haffer, CEO of Cascading AI. “Imagine a small business owner applying for a loan on a Friday evening and waiting 72 hours for a response from a loan officer. With Casca, they receive an email response within minutes from Sarah – who leads them through the entire application process. Sarah is infinitely patient and infinitely kind; every loan applicant receives the care and attention they deserve.”

Casca is already showing results by achieving nearly 3x higher conversion rates and reducing manual effort in the back office by 90% compared to traditional methods. 

So what?     

I like this one a lot.

The challenge with generative AI is figuring out where to plug it in. Large language models (LLMs) are very capable at tasks that require generalized reasoning (and they’re getting better), and they never get tired or need to take a break. On the other hand, they lack domain-specific experience (think of them as an intern vs a seasoned VP), and they are so eager to please that they will make shit up (hallucination).

What you want to find is a use case that requires low-level human reasoning along with a lot of manual work. The use case should be so overburdened by this manual work that profit margins are thin or non-existent. And the use case should have ample data/rules/processes surrounding it, so that the model can, over time, be trained to perform the job as well (or better) than humans.

Cascading AI’s first customer is Bankwell Bank out of Connecticut. The bank is using Casca to prequalify small business owners for SBA 7(a) loans, which are an inexpensive alternative to traditional small business loans, but also confusing and time-consuming to qualify for.

Bankwell is using Casca to automate that initial qualification process:

In the pilot of Casca, a potential small-business borrower comes to Bankwell Bank’s website and fills out a form, providing basic information to see if they prequalify for a loan. Several tasks need to happen to determine if the business is truly qualified for a loan and eligible for the SBA program.

Cascading AI’s virtual assistant, which is named Sarah, works through those tasks, asking the customer follow-up questions like, “How many years has the company been in business?” and “Tell me why you need a loan.” 

Still, Bankwell is keeping humans in the loop. Humans review every pre-qualified application that Sarah sends them. This, according to Bankwell’s Chief Innovation Officer, helps address the concern over hallucination, which he doesn’t see going away:

“At scale, I think that is the issue … I just don’t see us not having every single thing reviewed by a human.” 

#3: Trust, But Categorize

What happened?

Trustly and MX, two large players in the world of open banking, teamed up:

Open Banking payments provider Trustly has integrated MX Technologies’ data enhancement services as part of its Open Banking product suite.

By using MX Data Enhancement, as part of its Pay with Bank solution, consumer-allowed transaction data is set to be cleansed and categorized to provide comprehensive information for merchants to better understand their users’ needs and requirements. This enables businesses to inform and deliver more tailored marketing offers and loyalty programs, thus supporting their development and expansion.

So what?

This is interesting for a few reasons.

The first is specialization. Trustly has, for quite a while, been laser-focused on ‘pay by bank’ solutions. MX, which has its roots in the world of PFM and transaction categorization, has been doubling and tripling down on its data enrichment capabilities. This partnership is the natural result of this increased focus and a sign that these companies believe the best way to win against larger and deeper-pocketed competitors like Plaid (likely to IPO soon) and Finicity (owned by Mastercard) is to specialize.

Second, I’m curious what the “tailored marketing offers and loyalty programs” that Trustly is trying to enable for merchants (with the help of MX) will look like and how feasible such capabilities will be under the CFPB’s finalized 1033 rule.

My friend Matt Janiga at Trustly gave a good overview on Twitter of what this could look like and the argument for allowing it under the CFPB’s final 1033 rule. Personally, I agree with Matt’s argument, but I’d classify it as a bit of a longshot, given that the CFPB’s proposed 1033 rule strictly limits what third parties can do with consumers’ data (for more on these secondary use restrictions and a whole lot of other 1033 nerdery, check out this essay).


2 FINTECH CONTENT RECOMMENDATIONS

#1: Payfac in 1,000 words (by Matt Brown) 📚

I’m loving these little explainer posts from Matt on various topics in fintech. This latest one – 1,000 words on payfacs, is great.  

#2: JPMorgan Chase Is Adding Branches — The Rest Of The Industry Shouldn’t (by Ron Shevlin, Forbes) 📚

The news that Chase is planning to build 500 new branches in the next three years created a bit of a tizzy on fintech twitter. I myself explained it by comparing Chase to a blue whale. 

Here’s a much less silly and more cogent argument from Ron on why Chase can (and probably should) build new branches, while all other banks shouldn’t.


1 QUESTION TO PONDER

If we created a Hall of Fame for fintech writing – for individual pieces, not publications or authors – which pieces would be first-ballot inductees? Which pieces of fintech writing have stuck with, even years after first reading them?