Blend Labs handled considerably more mortgage loans in the second quarter, but its revenue did not keep pace.
Funded loans on Blend’s platform increased 14% year over year to approximately 233,000. Mortgage Suite revenue rose half as fast, increasing 7% to $19.2 million.
Economic value per funded loan fell to $79 from $83 in the first quarter. Management attributed the decline largely to fixed-fee customer arrangements, which mechanically lower the per-loan calculation when volume increases.
The gap between production and revenue puts more weight on Blend’s next growth bet: Autopilot, its AI agent for mortgage fulfillment.
Preliminary data from loans processed through Autopilot showed a 10% to 15% improvement in pull-through and a two-to-four-day reduction in cycle time compared with standard workflows, management said during the company’s second-quarter earnings call Thursday.
Blend also estimated that Autopilot automated an average of 4.5 hours of loan fulfillment work per loan.
For lenders, those are the numbers that matter. Fewer manual touches. Faster files. More applications reaching the closing table.
Whether lenders will pay for those gains at scale remains unsettled.
Autopilot Moves From Trial To Paid Contracts
Autopilot became commercially available July 1 after a free preview. Six lenders, including mortgage servicer Onity, had signed contracts as of the earnings call. Blend said 65 lenders activated the product during the preview, processing more than 45,000 loans.
Autopilot reviews borrower documents, checks them against lending guidelines, calculates qualifying income, flags inconsistencies, and requests missing documentation before underwriting.
High-confidence requests can go directly to the borrower. Less-certain issues are sent to the lender’s team for review.
“Autopilot is not a wrapper around a generic model,” Blend co-founder and Head of Blend Nima Ghamsari said during the call.
Blend is positioning the product inside the origination workflow, not as another stand-alone AI tool sitting alongside it. NMP previously reported on Blend’s Autopilot MCP server, which allows authorized AI agents to interact with lending functions including credit, pricing, automated underwriting, compliance, documents, and loan status.
The company is initially selling Autopilot through flat-fee, one-year contracts. Management expects to move toward per-funded-loan pricing over time, tying Blend’s revenue more closely to customer production.
That structure aligns Blend with lender results. It also leaves future Autopilot revenue more exposed to mortgage volume.
Blend did not disclose the value of the first six contracts and cautioned investors against counting on a meaningful near-term revenue contribution. Large financial institutions need time to complete governance reviews of agentic tools, management said.
The delay may work both ways. Governance reviews slow adoption, but they could favor established vendors already connected to lender systems over smaller AI companies trying to enter from the outside.
The Aug. 6 AI deadline was only the beginning. Lenders must now show that their controls work on actual loan files, not simply point to a written policy.
For Autopilot, lender interest may move faster than lender approval.
Churn Raises The Stakes
Blend received more customer churn notices during the quarter. Management said the affected mortgage customers generally planned to move to lower-cost or free point solutions they already had.
The company expects the departures to reduce annual revenue by a low-single-digit percentage.
The churn explains why Autopilot matters beyond its immediate sales. Basic mortgage software is getting cheaper and easier to replace. Blend is betting that automation embedded across the loan workflow will give lenders a reason to stay—and pay.
Autopilot cannot succeed on performance claims alone. Blend must show that better pull-through, shorter cycle times, and less fulfillment work justify another expense in lenders’ already crowded technology budgets.
Operating Loss Narrows
Blend reported total second-quarter revenue of $33.8 million, up 7% from $31.6 million one year earlier and near the high end of its guidance.
Consumer Banking Suite revenue rose 6% to $12.2 million, while professional services revenue increased 12% to $2.4 million.
The company’s GAAP operating loss narrowed to $1.6 million from $4.8 million in the second quarter of 2025. Non-GAAP operating income rose to $7 million from $4.6 million, producing a non-GAAP operating margin of approximately 21%, compared with 15% one year earlier.
“We delivered the quarter we aimed for,” said Jason Ream, Blend’s head of finance and administration.
Free cash flow reached $6.9 million, compared with a $9 million outflow one year earlier. Blend ended the quarter with $44.9 million in cash, cash equivalents, and marketable securities and no debt.
It repurchased 11 million shares for $18.2 million, leaving $13.2 million under its existing share-repurchase authorization.
Blend added or expanded 14 customer relationships during the quarter, including the six Autopilot agreements. Its late-stage pipeline — deals the company aims to close within the next quarter — grew nearly 40% between March and June, management said.
The company is also applying AI internally. Ghamsari said engineering throughput has increased 3.6 times since January with roughly the same headcount.
Softer Volume Ahead
Blend forecast third-quarter revenue of $31.5 million to $33.5 million, representing a year-over-year change ranging from a 4% decline to a 2% increase.
Third-quarter funded-loan volume is expected to total approximately 200,000 to 210,000 loans. Blend projected non-GAAP operating income of $3.5 million to $4.5 million, down from $7 million in the second quarter.
About $1.5 million of the sequential expense increase is tied to Blend Forum, the company’s annual customer event.
For the fourth quarter, Blend expects between 180,000 and 190,000 funded loans, down 10% to 15% from one year earlier.
Autopilot’s first contracts establish that some lenders see value in the product. The harder test comes after the trial: whether its early results hold across more lenders and loan types, and whether lenders cutting technology costs will pay for another platform feature.
