At Blend Forum 2025, our annual executive gathering where more than 100 leaders from the nation’s top banks, credit unions, and IMBs came together to talk about the future of lending, one theme stood out. In his opening keynote, Nima Ghamsari, Co-Founder and Head of Blend, put it plainly: the speed of technology adoption is now the defining advantage for lenders.
Institutions that move quickly from pilots to practice, with use cases that lower costs and strengthen relationships, will set the pace for the industry. The urgency of that message carried through every discussion. Lenders don’t need more bolt-on tools that add complexity. They need intelligent systems that sit at the core of origination and actually execute the work.
Beyond digitization
The industry has spent the last decade investing heavily in digitization. Online applications, e-signatures, borrower portals, and automated verifications have transformed the front end of the borrower experience. Those investments paid off in higher pull-through rates, shorter cycle times, and better engagement.
But the core economics haven’t changed. Origination still costs $10–12K per loan, cycle times average 20–30 days, and exceptions still send files back to human hands. What digitization modernized were the touchpoints, not the process itself. Files sit in queues, documents are checked manually, and quality control happens after the fact.
The result is an industry weighed down by rigid workflows while consumer expectations and market pressures accelerate.
Why agentic AI is a step change
The opportunity for AI in mortgage lending isn’t just about making existing steps faster. It’s about rethinking how the process moves altogether. Traditional rules-based automation can pass a file along, but it breaks down in the gray areas where most lending actually happens.
Agentic AI changes that equation. These systems interpret information, reconcile inconsistencies, and act on their own while knowing when to bring in a human for oversight. Documents aren’t just digitized, they’re understood. Conditions aren’t just flagged, they’re resolved. Origination becomes less about processing and more about managing outcomes.
This represents a true step change: from static workflows to dynamic, continuously executing systems. It’s the difference between an assembly line that halts whenever something doesn’t fit and a system that adapts instantly to keep production moving.
Early pilots point to what’s next
The shift from theory to practice is already underway. Forward-looking lenders are piloting agentic AI capabilities that move beyond surface-level automation and into the execution layer of origination. Blend is testing applied use cases within its platform to show how AI can handle more of the heavy lifting across the lifecycle.
Document intelligence now classifies and verifies files in seconds, pulling out critical data and flagging discrepancies that once required hours of review. Conversational intelligence is helping loan officers by summarizing calls, surfacing intent signals, and providing real-time coaching that strengthens both compliance and conversion.
Another promising area is quality control. Manual audits of hundreds of documents and thousands of checks have long been a drag on productivity and a source of costly risk. Early pilots show that AI can perform this review dynamically, producing a transparent quality score in minutes. The outcome is not just efficiency but stronger loan quality and greater investor confidence.
Together, these pilots illustrate what the next chapter of origination could look like: a system where AI is not a side feature but an active participant in moving loans forward.
The competitive imperative
These examples show what’s possible, but they also highlight a widening gap between experimentation and enterprise value. According to recent studies, 80% of institutions are experimenting with AI, yet fewer than 5% have taken those efforts into production. Too many initiatives remain siloed, disconnected from workflows, and ultimately fail to deliver measurable outcomes.
That tension surfaced clearly at our AI Roundtable. Some lenders are just beginning, testing AI in narrow use cases like document review. Others are piloting broader applications such as internal copilots or knowledge repositories, but struggle with scaling governance, data quality, and adoption across the enterprise. In many cases, individual employees are experimenting faster than corporate programs can keep up, creating a patchwork of adoption levels inside the same organization.
For lenders, the challenge is no longer whether AI works in theory. It’s about moving from scattered pilots to systems that materially impact cost, certainty, and growth. Institutions that embed intelligence at the core of origination, rather than bolting it on at the edges, will pull ahead in both efficiency and borrower experience
Looking ahead
The future of lending belongs to those who adopt systems that don’t just digitize processes but actually think and act on their own. By moving beyond experimentation and embedding intelligence into the execution layer, lenders can create a fundamentally different operating model — one defined by speed, certainty, and trust.
Click Here
