Blend has shared early production results for Autopilot, its agentic system for lending, showing a marked increase in pull-through rates and a reduction in loan-cycle times. Since March 2026, Autopilot's pre-underwriting agent has processed more than 50,000 live production loans across lenders on Blend's Home Lending platform. Blend, founded in 2012 and publicly traded since its 2021 listing, supplies digital origination software to banks, credit unions, and mortgage lenders, so its live production data offers a real-world sample of agentic AI operating inside a regulated consumer-finance workflow.

Results Across Analyzed Cohorts

Across the analyzed cohorts, lenders using Autopilot's pre-underwriting agent recorded:

• Pull-through rates 10% to 15% higher
• Loan-cycle times shortened by 2 to 4 days
• 4.5 hours of loan fulfillment tasks automated on average per loan
• An estimated $600 saved in fulfillment costs per funded loan

Blend emphasized that the early impact comes from live production loans with real borrowers, not pilots or simulations. To measure it, the company compared 24 lender and loan-type cohorts against their own pre-Autopilot performance, a comparison covering more than 175,000 loans. The findings are vendor-reported, drawn from Blend's own platform data and cohort comparisons rather than an independent audit. The stakes for lenders are nonetheless concrete: industry benchmarks such as the Mortgage Bankers Association's lender performance studies have tracked per-loan production costs running into the thousands of dollars in recent years, while depressed origination volumes have kept per-loan economics under sustained pressure.

A Shift Larger Than Efficiency Gains

According to Blend, the results point to a shift larger than efficiency gains. Most AI in lending today accelerates individual steps in a process that still runs sequentially, with each stage waiting on the one before it. Autopilot's agent-first model instead works the file in parallel and in real time, which changes what the origination process costs, how long it takes, and when borrowers get answers. That framing tracks a broader turn across enterprise software, where vendors are moving from AI assistants that speed up individual steps to agents that execute multi-step workflows autonomously.

"I think the biggest mistake people make with AI is trying to be too incremental. Our approach is to ask: What would this industry look like if you designed it from the ground up around agents?" said Nima Ghamsari, Co-founder and Head of Blend. "That's no longer a thought experiment. Autopilot has run on more than 50,000 live production loans, and the economics are already changing."

How Autopilot Works

More than half of borrowers apply for a mortgage loan outside business hours, and more than 90% of borrowers who submit do so within 24 hours of starting, according to Blend network data. Borrower intent is highest in that first day and fades quickly: friction that stalls an application overnight doesn't just delay it, it risks losing the borrower to a competitor. For lenders, that makes the first 24 hours a critical window to keep borrowers moving, even when no one is staffed to respond.

Autopilot operates inside that window. It reviews the application in real time, identifies missing or inconsistent information, completes pre-underwriting work, and follows up with the borrower while they are still engaged. As a result, files reach processing and underwriting cleaner, reducing rework, borrower callbacks, and late-stage conditions.

"Instead of a person taking the first pass and an agent checking the work, Autopilot agents are working behind the scenes before a human ever touches the file. Rules-based automation could never do that, because rules only work on predictable files, and underwriters spend all their time on the ones that aren't," Ghamsari added. "By the time it reaches a loan team, the work is done, which gives them their time back to spend with borrowers instead of paperwork." The scope matters in a heavily regulated process: Autopilot automates the preparation work that precedes human review rather than underwriting decisions themselves, and mortgage origination remains governed by extensive federal consumer-finance rules.

The results set up clear questions for the months ahead: whether the gains hold as Autopilot scales beyond its initial cohorts into a wider mix of lenders and loan types, and how a mortgage-technology landscape long built on rules-based automation responds.

Source: GlobalFinTechSeries