Figure Says AI Agents Lift Funded Home-Equity Conversions 143% When Paired With Loan Officers

Share this article!

Figure Technology Solutions is testing a version of mortgage automation that looks less like a chatbot and more like a junior loan-production employee.

The home-equity lender and capital-markets platform said Sept. 10 it has integrated Sierra’s new Horizon platform into its loan-origination workflow to re-engage borrowers who start a home-equity application and then stall. According to Figure, borrowers who interacted with the AI agent and a loan officer funded loans at a rate 143% higher than borrowers handled by loan officers alone.

That is an eye-catching number in an industry where lenders have spent heavily on point-of-sale technology, automated verification and borrower communications without eliminating one of the oldest problems in consumer lending: people simply stop responding.

Figure’s pilot is aimed directly at that gap. The agent can contact stalled applicants by voice and text message, remain active over multiple days, help borrowers through routine steps such as identity verification and permission for credit checks, and then hand the file back to a human loan officer when the borrower reaches a point that requires live assistance.

Figure is positioning the agent as support for loan officers rather than a substitute for them. The software handles repetitive follow-up at scale, while the originator remains responsible for conversations that depend on judgment, persuasion and borrower confidence.

The conversion numbers deserve attention — and scrutiny

Figure reported several results from a July 2026 internal performance analysis. Stalled applicants who interacted with the agent advanced through individual friction points at rates 30% to 52% higher than borrowers who did not. Borrowers who engaged with the agent, with or without loan-officer assistance, funded 67% more loan volume. The largest reported lift — 143% — came when the agent and a loan officer worked together.

Those figures are company-reported pilot results, not an independent industry study. Figure did not disclose in the announcement the total number of borrowers in the test, the control-group design, the time period over which the 143% comparison was measured, or whether borrower characteristics differed between groups. That does not make the numbers unimportant, but it does limit how far lenders should extrapolate from them.

The underlying problem is real. Figure cited the Mortgage Bankers Association’s 2025 Home Equity Lending Study, which found average HELOC closing pull-through of 49% in 2024. The company also pointed to roughly $35 trillion of homeowner equity and said nationwide HELOC volume reached $271 billion in 2025.

In practical terms, a lender can spend money to generate a lead, obtain an application, verify part of the file and assign a loan officer — only to lose the borrower before closing. That makes abandoned applications a revenue problem, not merely a customer-service problem.

Mortgage AI is moving closer to production work

The Sierra partnership is notable because the agent is being placed inside Figure’s loan-origination system rather than sitting outside the mortgage workflow as a general-purpose assistant. Figure says the agent was trained to work natively in its LOS and can act on a file over an extended period instead of answering a single question and ending the interaction.

Sierra describes Horizon as a platform for so-called long-horizon agents — software designed to keep working across multi-step tasks instead of responding to one prompt at a time. Figure is Sierra’s first U.S. integration of Horizon, according to the companies.

That distinction matters for mortgage companies. Much of the industry’s first wave of generative AI has focused on call summaries, email drafting, marketing copy and internal knowledge search. Those uses can save time, but they sit around the perimeter of origination. Re-engaging an applicant, obtaining missing permissions and moving a borrower from an incomplete file toward funding puts AI much closer to revenue production.

It also raises the stakes. Every automated borrower interaction touches compliance, recordkeeping, consent, fair-lending controls and reputational risk. A system that simply summarizes a call can be reviewed after the fact. A system that is actively moving consumers through a financial transaction requires stronger supervision, testing and escalation rules.

Early results favor a hybrid model

Figure CEO Michael Tannenbaum characterized the model as a combination of humans and agents. The early data, if it holds up outside the pilot, supports that framing more than the familiar prediction that AI will replace mortgage originators.

Mortgage lending remains a high-friction transaction. Borrowers disappear because they are confused, busy, shopping competitors, worried about documentation or uncertain whether proceeding makes financial sense. An automated system can follow up indefinitely and consistently. It cannot assume that every stalled borrower should be pushed toward a closing.

That is where loan officers remain important. The most useful application of this technology may be triage: automation keeps routine files moving, identifies the exact point of friction and brings a human into the conversation when persuasion or judgment is actually required.

For lenders, the economic question will be whether the conversion lift survives outside Figure’s own platform and whether the added funded volume outweighs integration, compliance and oversight costs. Figure said it plans to make the integration available to partners in its network over the coming months, which should create a broader test than the initial pilot.

If those deployments produce similar results, the competitive benchmark for mortgage technology could shift. Lenders may stop asking whether their AI can answer borrower questions and start asking whether it can measurably move a loan from application to funding without weakening compliance or the customer relationship.