CHAPTERS
- 0:05 – 0:42
Kastle AI: an “AI employee” automating bank operations at scale
Diana welcomes Kastle AI founders Nitish and Rishi and frames the company’s momentum after a $24M Series A. The founders define Kastle as an AI employee automating bank operations, beginning with mortgage servicing and expanding across major servicers.
- •Kastle positions itself as an AI employee for bank operations
- •Started in mortgage servicing; now used by 10 of the top 25 servicers
- •Recent $24M Series A led by Insight Partners
- •Focus on automating customer service and collections
- 0:42 – 1:00
Traction metrics: from $10M every two months to $10M per day
Rishi shares how quickly usage scaled, culminating in over $2B in transactions processed by AI agents. The conversation highlights the acceleration from early volume to daily throughput.
- •Crossed $2B in transactions processed by AI agents
- •A year prior: ~$10M every two months
- •Now: ~$10M processed every day
- •Signals strong product adoption in real banking workflows
- 1:00 – 2:50
Why banks need AI employees: growth without “rip-and-replace” change management
The founders explain the core thesis: banks can’t scale revenue without scaling back-office headcount, and they struggle to replace core systems. Kastle aims to deliver reliable AI-driven workflow automation without requiring multi-year core-system migrations.
- •Bank growth typically forces proportional back-office growth
- •Regulated institutions are slow at change management
- •Replacing core systems can take 5–10 years
- •Kastle integrates as an automation layer vs. building a new system of record
- •Goal: fast, reliable AI deployment across operations
- 2:50 – 3:46
Rewinding to YC: the original idea and early pivot attempts
Diana revisits Kastle’s YC origins, when the team applied with a different product concept. Nitish explains they explored an AI assistant for real estate/home buying and then an AI SDR angle before realizing the direction wasn’t right.
- •Initial concept: AI for home buying/real estate assistance
- •Early pilot failed; team reassessed quickly
- •Shifted to an AI SDR approach for home buying lead-gen
- •Made progress but still questioned long-term potential
- 3:46 – 5:00
Killing a working product: choosing mission-critical over ‘nice-to-have’
Despite reaching about $4K MRR, the founders decided to reset one month before Demo Day. They describe the hard criteria: they wanted a problem customers couldn’t live without, not a helpful but non-essential tool.
- •Reached ~$4K MRR during the batch
- •Decision to kill the idea ~30 days before Demo Day
- •Key insight: product felt ‘nice-to-have,’ not existential
- •Commitment to building a large, durable company
- •Willingness to restart even under extreme time pressure
- 5:00 – 5:37
The 30-day sprint: intensive user interviews and conference cold outreach
With little time left, the team launched a rapid discovery sprint—traveling, attending conferences, and starting conversations wherever possible. They refined their pitch and learned how to uncover urgent, operational pain.
- •High-tempo user research under Demo Day deadline
- •Two weeks of conferences and constant customer conversations
- •Cold outreach tactics (approaching people outside sessions)
- •Iterative learning: getting ‘less awkward’ and more effective
- •Goal: find a high-pain, high-volume operational problem
- 5:37 – 6:43
Breakthrough insight: 10M Americans pay mortgages by phone—handled by humans
At a conference in San Francisco, they learned an unexpectedly large volume of mortgage payments happen over the phone monthly. This operational burden, combined with regulatory constraints on outsourcing, made voice automation a compelling wedge.
- •Discovery: ~10M Americans make mortgage payments by phone monthly
- •Calls handled by human contact-center agents
- •Regulation and risk make offshoring/outsourcing difficult
- •Voice AI was new; few were processing payments in regulated mortgage servicing
- •Clear ‘game-changing’ demand signal from industry conversations
- 6:43 – 8:06
Landing the first customer via a high-stakes pitch competition (and a broken demo)
With only weeks left, the founders ‘speed-ran’ customer acquisition by entering a pitch competition in San Diego. Their live demo failed twice on stage before working on the third try—but they still won, generating inbound that helped close their first customer just before Demo Day.
- •Entered a pitch competition to accelerate credibility and exposure
- •Live voice-payment demo failed twice in front of hundreds
- •Improvised explanations and fixed the issue on the third attempt
- •Won the competition, creating significant inbound interest
- •Closed first customers days before Demo Day
- 8:06 – 8:48
Going all-in on deployment: moving into Newrez’s office to go live
After early prototypes, Kastle targeted Newrez, the third-largest servicer, to prove the model at serious scale. The entire team relocated to Phoenix and worked from the customer’s office until the system was live.
- •Early customers used for rapid prototyping and iteration
- •Newrez: third-largest mortgage servicer became a key scaling milestone
- •Team moved the company to Phoenix to ensure successful launch
- •Embedded execution until production ‘go-live’
- •Demonstrates high-touch enterprise deployment strategy
- 8:48 – 10:40
The 5–10 year vision: AI coworkers across the full consumer lending lifecycle
Rishi outlines expansion beyond mortgage servicing into broader bank operations, including servicing multiple lending products and even loan origination. The long-term goal is an AI coworker that automates high-volume tasks while humans handle judgment-heavy exceptions.
- •Proved autonomous agents can operate in highly regulated environments
- •Applying the same operating model across bank operations
- •Now supports auto loans, credit cards, personal loans, HELOCs, and origination
- •Vision: AI employee as a ubiquitous coworker in financial institutions
- •Automation frees human staff for complex, judgment-driven workflows
- 10:40 – 11:09
Compliance and reliability: making non-deterministic LLMs work in regulated banking
Diana highlights a core technical challenge: banks need deterministic, compliant behavior even though LLMs are inherently probabilistic. The discussion emphasizes building guardrails and systems that allow customers to configure agents safely.
- •Customers need to ‘prompt’/configure agents for workflows
- •LLMs are non-deterministic, but banking requires predictable compliance
- •Implied need for guardrails, auditing, and controlled behavior
- •Reliability is central to adoption in regulated environments
- 11:09 – 11:47
Post–Series A scaling: hiring across applied AI, product, and infrastructure
The conversation closes with how Kastle plans to scale after funding, especially as they work with larger banks and expand product breadth. Rishi lists key hiring priorities across engineering and product to support real-world AI systems.
- •Hiring ‘across the stack’ following Series A
- •Expanding applied AI team for real-world deployments
- •Growing product teams to support multiple product lines
- •Scaling infrastructure to handle enterprise-grade needs
- •Focus on supporting larger banks nationwide
