The AI Employee for Banks

Author: YC Root Access · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: The AI Employee for Banks · Channel: YC Root Access · 11:26 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, Markets and economy

Summary

Castle AI builds an AI employee to automate bank operations, initially focusing on mortgage servicing and now serving multiple lending products. Its founders chose to augment banks’ existing systems rather than require risky, lengthy core-system replacements, and say the company has processed $2 billion in transactions using AI agents. They recount abandoning two earlier startup ideas after customer research pointed them toward handling mortgage payments by phone. Looking ahead, they aim to support operations across financial institutions while hiring across applied AI, product, and infrastructure.

Key points

Castle AI describes its product as an AI employee that automates operations inside banks.
The company began with mortgage servicing and works with 10 of the top 25 servicers, including Newrez.
The founders favored integrating AI into existing bank operations over replacing core systems, which can take years to change.
Castle says it has processed $2 billion in transactions using AI agents, including payment processing.
The founders initially pursued real estate and AI sales-development ideas, but restarted after deciding those ideas were not essential enough to customers.
Customer interviews and conference visits revealed that millions of Americans make mortgage payments over the phone, often handled by human contact-center agents.
A live demo failed twice before working during a pitch competition; the event generated inbound interest and helped the founders land their first customer.
Castle moved its team to Phoenix and worked from Newrez’s offices to get the servicer live.
The company is expanding beyond mortgages into auto loans, credit cards, personal loans, HELOCs, and loan origination.
Castle says it is hiring across applied AI, product, and infrastructure teams.

Chapters

Start	Chapter	Summary
00:00	What Castle AI does	The founders introduce Castle’s AI employee for bank operations and describe its growth in mortgage servicing and payment processing.
01:01	Why build an AI employee for banks?	They explain why banks need operational automation that works with existing core systems rather than replacing them.
02:36	Pivoting toward mortgage servicing	The founders recount their earlier startup ideas, why they abandoned them, and how customer interviews led them to mortgage payments by phone.
06:45	The first customer and going live	They describe a malfunctioning pitch demo that still brought in leads, their first customer, and moving to Phoenix to launch with Newrez.
08:48	Expansion plans and hiring	The founders discuss applying Castle across lending and bank operations, the role of human judgment, and their hiring plans.

Quotes

Castle is an AI employee that automates all the operations inside a bank. — 00:00

we recently crossed $2 billion in transactions processed using AI agents. — 00:31

we couldn’t see ourselves like building a really large company on that idea. — 04:10

10 million Americans making their mortgage payment every single month over the phone. — 05:44

automating the high-volume labor-intensive work and freeing up time for complex workflows to be handled by human agents — 09:52

Concepts and entities

Castle AI, Nitesh, Rishi, Insight Partners, Newrez, Devin, Voice AI, YC

Full transcript (original language)

00:00 [music] >> All right. So, I’m excited today to welcome Nitesh and Rishi, the founders of Castle AI, who just closed a $24 million Series A led by Insight Partners. Congrats, guys. >> Thank you. >> What does Castle do? >> Yeah. No, thanks for having us, Dana. Excited to be here. So, Castle is an AI employee that automates all the operations inside a bank. We started 2 years ago,

00:31 uh starting with mortgage servicing operations, and today we work with 10 of the top 25 services like Newrez and help them automate their customer service and collections operations across the business. >> And in terms of volume, where are you guys at now in terms of processing payments? >> Uh so, we recently crossed uh $2 billion in transactions processed using AI agents. Yeah, it’s crazy to think that same time last year we were doing about 10 million every 2 months, and now we do that every day. >> Yeah. >> The interesting approach you have is

01:01 that you are building a company as an AI employee. Why why is that? >> Yeah. So, this is actually one of the uh interesting points that we noticed is you know, I feel like most of the problems that banks face today is that they in order to grow that grow revenue, grow uh loans, grow deposits, they also need to grow the back office. And one of the interesting things that we saw is that uh banks are also forced are really bad at change management.

01:32 It’s very hard for such large regulated industries to completely replace their core systems. And so, as we built Castle, we had an opportunity to either, “Hey, do we go and execute on building a new system of record, or do we help them How do we help banks sort of adopt AI in a reliable way and deliver value quickly uh without them having to go through this entire process of you know changing their core systems which might take 5-10 years. We were seeing a lot of um you

02:05 know rapid improvements happen in like say for example coding where you know we use tools like Devin >> Yeah. >> and they’ve helped us supercharge our workflows. And we we found that the you know banks really needed something similar in order to like supercharge their own operating workflows but there was like no good solution out there that was helping them do that. Uh it was either you change the existing system of record or you had this like one point solution. So that’s when we decided that

02:36 hey we have an opportunity here to like really help our customers reliably deploy AI across their operations and realize that value that other industries are seeing without having to go through that rip and replace of their existing core systems. >> Now let’s go back to two years ago. You were here at YC going through the batch in summer 24 and you guys actually applied with a very different idea. >> Yeah. >> Do you remember what it was? >> I think it was real estate assistants for to help real estate

03:07 agents close more loan to close more kind of deals. So AI for home buying essentially. I think within week one we proved out that we had a pilot we lost that and then we realized that this is not going anywhere and then we started talking to a bunch of more people went into the AI SDR route where we were helping essentially other companies message other businesses to source more business. >> I think it was a AI SDR for specifically home buying. I think you were helping

03:39 all these brokers find more leads to buy homes and you actually made progress on it. You were working on that for most of the batch. >> Yeah. >> It was around the summer. I think you were working on that through June, July and beginning of August and you got to about 4,000 in MRR. Just like, “Okay.” But something happened and you decided to kill the idea 1 month before demo day and went back again to zero. This is the second time you went back to zero. >> Yeah, I know. I clearly remember we were actually in Phoenix and we were driving

04:10 you know on the way to the airport in SF. And Nitesh and I were like, “What are we What are we doing? Like we we don’t really see we couldn’t see ourselves like building a really large company on that idea.” And most of all like we we we didn’t feel that our customers this was like existential for our customers. That like if we disappeared tomorrow that they would absolutely be you know disappointed that like they no longer have a castle. It felt like if we were

04:42 like a nice-to-have for them and Nitesh and I were like we you know we we really cared about like building something for the long term and we wanted to do like go back to the drawing board no matter how painful that was and restart over to build something that like our customers like couldn’t live without. >> So you were sprinting. This is only 30 days before demo day and you went on this sprint on doing a lot of user interviews. You’re flying all around the country to meet people. You went to conferences for 2 weeks.

05:14 And tell us about how you eventually landed in this idea of around agent home mortgages. >> We went all to basically every conference that we could find. It was a lot of awkward conversations of standing outside conferences as people walked out and trying to like cold meet them and talk about their problems. But we basically did that and we actually got pretty good at it. >> And got less awkward. >> And less awkward at it. And so it was actually at one of the conferences here in San Francisco where we

05:44 learned about 10 million Americans making their mortgage payment every single month over the phone. >> You told me that. I was shocked. It’s like there’s people that call over the phone to pay or be reminded for the mortgages. It’s like >> Every single month. >> I was like, what? >> Yeah, and and all of these calls were actually being handled by a human agent in the contact center because, you know, mortgage lenders just don’t have as much They have very highly regulated. It’s very risky for them to outsource these

06:14 operations to other outside countries. So, all these calls were being taken in the United States in place like Phoenix, Dallas, and is all the all that like the mortgage lenders could talk about that if someone could solve this problem for us like that would be a game-changing experience. And no one in Voice AI was like relatively like new at that time and no one was really doing it to like process payments as for such a regulated industry like mortgage. And that’s that’s how we got our first break. >> Do you remember how you got your first

06:45 customer? Because by that time you were running out of time. There was only 2 weeks until demo day. You got this funny insight. And >> Yeah. >> What what happened? I think it took another 2 weeks to get your customers. >> Yeah. Yeah, so we we were like, okay, how do we speed run this? So, we went to another conference where it was actually we found out about this pitch competition. And it was in San Diego. Before our time on the stage, the teacher was in the background. You want to talk about that? >> Rishi was still going on the conference trying to meet people and I had found a

07:17 corner of the auditorium and was just kind of finish our demo in time. And we went up to the stage. We started the demo. It was a phone call and just to show whether we can take a payment. First time he called in, it didn’t work. >> On stage. >> On stage. >> In front of hundreds of people. >> So, we blamed it on his phone. We said, okay, it’ll work with my phone instead. So, I gave him my phone. Second time we called in, it didn’t work. And then I said, oh, I forgot to turn it on. And then [laughter] the third time we called in, it actually ended up working. And we we were really scared that, you know, this

07:48 didn’t go as well, but uh turns out we actually won that competition. And that created a ton of inbound for us and helped us land our first customer. >> In spite of the demo malfunctioning, that’s a that’s a cool story. So, you got a bunch of inbound from that. And I remember you guys just closed on Nica Time, your first customer just few days before demo day. >> Yeah. >> Cool. Then you went off to the races. And then after that, you guys went into build and close one of the largest servicers. Couple months later?

08:18 >> Yeah, so we had a few early customers that we were prototyping with. And then we got in touch with Newrez, which is the third largest servicer in the country. This is the time where we, you know, they had a similar problem. We knew that they had this problem. So, to actually make them live, we moved the entire company to Phoenix. And we moved in to their offices and, you know, I didn’t come back until they were live. >> So, you were cranking it until you went live. >> Yeah.

08:48 >> Now, fast forward. Now that you’re live and now processing 2 billion in transactions, where does Castle go in the next 5 to 10 years? >> What we’ve been able to do is prove that applied AI autonomous agents can work in one of the most regulated operational environments uh in financial services, which is mortgage servicing. And as a result of that, we’ve actually built a lot of technology underneath it that allows these agents to be highly

09:21 effective in real-time settings. We’ve basically taken that operating model and that technology and now applying across the consumer lending life cycle and across like a bank’s operations. So, we initially while we initially started with mortgages, we now have large banks using Castle’s AI employee to service auto loans, credit cards, personal loans, HELOCs, but also like originate loans. We think 5 years from now that, you know, Castle can be the the number

09:52 one AI employee that works as a co-worker across every operational financial institution there is doing the work alongside human agents and automating the the high-volume labor-intensive work and freeing up time for complex workflows to be handled by human agents where they where they need the judgment. And this is essentially like to say that like there is a ton of demand for financial products and banks want to be innovative and lend to their

10:24 community, build better experiences for their customers. The only thing that stops them from doing that is them not having the operational bandwidth or resource in order to, you know, provide personalized experiences for each of their customers. And Castle’s AI employees unlock that. >> So, one of the other things you’ve done is basically you are enabling your customers to basically prompt the agents, but inherently these agents run LLMs which are non-deterministic, but you’re working in a regulated industry. So,

10:54 somehow you built a hardness that makes your customer changes be deterministic and be compliant. That’s pretty hard to do. And you guys are growing a bunch, so you’re hiring a bunch of roles. Tell us about what are some of the roles that you’re hiring for. >> Yeah. So, we’re hiring across the stack with the Series A, we’re now starting to work with, you know, some of the largest banks across the country. So, we’re investing a lot in like applying AI in the real world and building these systems around it. So, we’re growing our

11:26 applied AI team, growing our product teams as we get into like multiple different products. And and also scaling our infrastructure teams. >> All right, guys. Thank you so much for coming and congrats again on the Series A. >> Yeah. Thanks so much, Anna. Appreciate it. >> [music]