AI in mortgage lending: how lenders use AI agents to close more loans
Many mortgage loans are lost long before underwriting: a lead that waits a day for a callback, an application that stops at the income section, or a borrower who never sends the last bank statement. AI agents help lenders close those gaps by responding to every borrower right away and following up until the loan funds or the borrower moves on.
Where mortgage lenders lose loans
- Slow follow-up on new leads. A Harvard Business Review study of 1.25 million sales leads found that the first hour matters most. Companies that contacted a lead within an hour were nearly 7 times as likely to have a real sales conversation with that person as companies that waited longer. Compared with companies that waited a day or more, they were more than 60 times as likely.
- Stalled applications. Borrowers start an application, then stop when they have a question, need to talk to a co-borrower, or aren't sure about the rate.
- Missing documents. Pay stubs, bank statements, and letters of explanation can take weeks to collect when each request is a separate email or call.
- Missed refinance windows. When rates drop, lenders need to reach past borrowers quickly, before another lender does.
Each lost application adds to the cost of the loans that do close. The Mortgage Bankers Association reported that loan production expenses for independent mortgage banks averaged $11,094 per loan in 2025, and named lower application pull-through as one of the reasons.
How AI agents help across the loan lifecycle
An AI agent can work with each borrower from first inquiry through closing and beyond, over chat, text, and phone:
| Stage | What the agent does |
|---|---|
| Inquiry | Responds within minutes at any hour, answers questions about loan programs, and asks qualifying questions |
| Prequalification and preapproval | Estimates what the borrower can afford from their income, debts, and down payment, then pulls credit with consent and presents the rate and loan options your rules allow |
| Application | Helps the borrower finish the application and follows up over days or weeks if it stalls |
| Underwriting support | Gathers what underwriting needs, tells the borrower about new conditions, and keeps them updated until a decision is made |
| Documents and conditions | Tells the borrower exactly what's needed, accepts uploads, and verifies income and assets through permissioned bank data |
| Processing and closing | Sends status updates, schedules appraisals and closings, and answers "where is my loan?" questions |
| After closing | Handles payment, servicing, and collections questions, and reaches out about refinancing or a home equity line when it fits the borrower |
When a borrower wants to talk to a loan officer, or a step requires one, the agent can pass along the conversation history, so the borrower doesn't have to start over.
Example: restarting a stalled application
Here's how an agent might recover an application that stopped partway through:
- A borrower starts a purchase application on Sunday night and stops at the income section.
- On Monday morning, the agent texts to ask if they'd like help finishing. The borrower is self-employed and isn't sure which documents count as proof of income.
- The agent explains the options and offers to verify income by securely connecting a bank account, with the borrower's permission. The borrower connects their account, and the agent completes the application.
- The agent books a call with a loan officer, who can see the full conversation before the call.
- Over the next few weeks, the agent collects the remaining conditions and keeps the borrower updated until closing.
This works because the agent remembers each borrower's situation and keeps working toward the same goal across many conversations, instead of starting fresh each time.
Results lenders are seeing
Rocket Mortgage built its Digital Assistant on Sierra. It works across chat and voice to collect information, pull credit, present personalized rates, customize loan options, and help with payments and loan status. Clients who start with the Digital Assistant close at three times the rate of those who don't. When clients use both AI chat and a banker, conversion is four times higher for both refinance (lead to application) and purchase (lead to preapproval letter). The assistant handles more than 400,000 chat conversations and over one million outbound dials each month.
How to measure AI in mortgage lending
Track each step from lead to funded loan, so you can see where borrowers drop off:
| Metric | What it measures |
|---|---|
| Speed to lead | Time from a new lead to the first contact |
| Contact rate | Share of leads who had a conversation with the agent or a loan officer |
| Lead-to-application rate | Share of leads who started an application |
| Application completion rate | Share of started applications that were submitted |
| Pull-through rate | Share of applications that funded |
| Time to close | Days from application to closing |
| Cost per funded loan | Total production cost divided by funded loans |
| Handoff rate | Share of conversations passed to a loan officer, and why |
Pull-through rate and cost per funded loan matter most. Faster responses only count if more of those borrowers end up with a funded loan.
Compliance requirements for AI in mortgage lending
Mortgage lending is one of the most regulated parts of financial services, so the agent has to follow the same rules as your loan officers and processors:
- Fair lending. The Equal Credit Opportunity Act and the Fair Housing Act prohibit discrimination against borrowers. Test the agent's responses across borrower profiles and keep records you can review.
- Disclosures. Under TILA-RESPA rules, lenders must send a Loan Estimate within three business days of receiving an application. Make sure the agent's actions trigger your disclosure workflows on time.
- Licensing. The SAFE Act and state laws define which activities require a licensed loan originator. Work with your compliance team to decide which steps the agent handles and which go to licensed staff.
- Consent for calls and texts. Contact borrowers only through channels and numbers they've agreed to, and honor opt-outs right away.
- Data privacy. Verify the borrower's identity before discussing loan details, and use permissioned access for financial data.
- Audit trails. Keep a record of every message, disclosure, and action.
Sierra's Compliance Console lets lenders set these rules across channels, use cases, and states. Borrowers can also connect their bank accounts securely inside the conversation, through Sierra's partnership with Plaid.
Getting started
- Pick one point where you lose borrowers, such as response time on web leads or stalled applications.
- Connect the agent to your CRM, loan origination system (LOS), and pricing engine, so it can answer questions and take action in the conversation.
- Set disclosure, licensing, and consent rules with your compliance team before launch.
- Define how the agent hands off to loan officers, including what context it passes along.
- Track lead-to-application and pull-through rates, then expand to document collection, servicing, and refinance outreach.
AI agents built on Sierra's Horizon can follow up with borrowers over days, weeks, or months, so every lead and application gets attention until the loan funds.
