Insights
- Despite digital advances, the average time to close a mortgage loan is still 38 to 42 days, largely due to complex workflows, document overload, and multiple handoffs.
- Agentic AI can autonomously make decisions aligned with business goals through adaptive workflows and continuous learning rather than just following preset rules.
- Rather than replacing platforms like Fannie Mae’s Desktop Underwriter (DU) or Freddie Mac’s Loan Product Advisor (LPA), agentic artificial intelligence (AI) can augment them by automating document collection, prevalidating inputs, and triaging exceptions before underwriting.
- Agentic AI can transform application capture, document processing, underwriting, and appraisals, thus reducing manual effort, minimizing errors, and accelerating cycle times for both borrowers and lenders.
- Successful adoption will involve orchestrating internal workflows, coordinating with external partners, ensuring compliance and security, and maintaining human oversight for critical decisions.
The mortgage industry still struggles with speed
Digital adoption in mortgages has risen over the years, yet the average time to close a mortgage loan still lingers around 38 to 42 days, the same as in 2018. The delay in closing a loan adversely impacts the borrower experience and has financial implications in terms of closing costs. Despite advancements in technology that automate repeatable tasks, derive insights from unstructured data to predict loan defaults or generate content like correspondent lending packages, the time to close a mortgage hasn’t reduced.
The challenges lie in the complexity of mortgage process
While loan origination system transformation and technology adoption have helped the mortgage industry, the mortgage loan process still involves multiple handoffs and external dependencies. The process is characterized by myriad system user roles and permissions, multiple integration points, and many stakeholders. The key friction points in the process include application capture, document processing, underwriting, and appraisal.
Application capture requires borrowers to supply vast amounts of information from a wide range of sources. The mortgage industry also deals with hundreds of document types of multiple pages that range from clean PDFs to handwritten documents and nonstandard forms. This means underwriters must make decisions based on a huge amount of information parsed from these multiple documents.
This raft of information in a wide range of formats means any attempt to fix this process struggles in the face of document overload.
The promise of agentic AI
While traditional AI typically executes predefined tasks based on rules, agentic AI is designed to make autonomous decisions aligned with specific business goals, rather than waiting for prompts and directions. It uses dynamic, adaptive workflows and continuously learns from user interactions and outcomes, rather than being limited to its initial programming.
Instead of a simple linear process, agentic AI follows an iterative cycle: It senses the environment, plans, acts, and then reasons from the results. The result should be less friction and better experience for both lender and borrower.
Here is an illustration of how agentic AI could work. The tech is not designed to replace existing systems like the digital platforms from Fannie Mae’s DU or Freddie Mac’s LPA that help assess a borrower’s credit risk and eligibility standards. Instead, it augments these platforms by automating document collection, prevalidating borrower inputs, and triaging exceptions before they reach underwriting. By acting as a connective tissue between loan origination systems, third-party providers, and human underwriters, agentic AI elevates the performance of existing technology.

Agentic AI can transform mortgage lending
- Working with the borrower to apply for a mortgage: A borrower’s loan journey begins with the Uniform Residential Loan Application (URLA), a grueling 236-field form that overwhelms first-time borrowers and creates opportunities for errors. It is the most significant source of friction — intimidating, manual, and error-prone.
With the introduction of agentic AI, the experience changes dramatically. The agentic system converses with borrowers, allowing them to provide income or property details in plain language, while the system transforms this into structured data that auto populates the correct sections. The agentic system extracts the data from uploaded documents to validate these entries instantly, and credit report data flows in automatically to complete liabilities. If additional information is required, the system doesn’t wait for the borrower to log in but autonomously reaches out based on the kind of data required. If there are delays in getting the required information, the system proactively sends out reminders at intervals as appropriate for the customer type. The borrower feels as if they are working alongside a mortgage application assistant, and by the time the borrower reviews the application, most fields are already verified and completed.
The impact is two-fold: not only does the borrower feel relief from the burden of complex forms, but lenders benefit from higher-quality, error-free applications entering the pipeline. - Mortgage document processing: Traditionally, incomplete or outdated borrower documents often go unnoticed until processors have time to review them. This creates bottlenecks, with days lost between submission, review, and borrower follow-up.
Agentic AI can remove these bottlenecks. As soon as the application lands, documents can be quickly classified and the necessary data extracted. It recognizes whether pay stubs are outdated, tax returns are illegible, or employment letters fail lender criteria. Instead of waiting for human review, the agentic system proactively notifies borrowers and sends reminders to ensure corrections happen without delay. The system learns to continuously improve document classification and extraction, which in turn reduces inaccurate data capture. As this is also being continuously refined, the system becomes smoother and better as time goes on. - Mortgage loan underwriting: While automated underwriting systems can render instant findings, human underwriters must manually validate data, documents, and compliance for final approval. This step is not only the most critical but often the most time-consuming.
The agentic system begins its work the moment documents are uploaded rather than waiting for the loan to reach the underwriting queue. It continuously monitors borrower files, flags inconsistencies, and prefills conditions based on underwritten guidelines. Once cleared, those conditions automatically update, often before the underwriter even touches the file.
The system continues to learn some of the common inconsistencies in loan applications and ensures that they are taken care ahead of time. In the event of loan conditions, it autonomously routes the loan to the processor if needed or creates a note for the processor to address. This is a skill that the agentic system continues to learn so that it knows when the loan should be sent back into the workflow, when it’s just an informational action, and when it requires loan processor intervention.
The system also learns to perform loan services trigger at the right time, with full context and available data. If a service call fails due to integration errors, the system interprets the error code and either retries automatically or reaches out for corrective information. Upon completion, results are validated and appended directly into the loan origination system, ready for underwriting.
The net effect is that human underwriters can focus solely on high-value judgment calls and final review, instead of routine checks. This can reduce underwriting cycle time from weeks to days, while improving quality and auditability. - Mortgage appraisals: Perhaps the most notorious delay in mortgages in the US is appraisal scheduling and completion. Initiatives such as Freddie Mac’s automated collateral evaluation (ACE) and ACE+ PDR (property data report) or Fannie Mae’s Value Acceptance + Property Data program are an attempt to bring down the issuance cycle, but inspection and property data collection of over 100 data points including 40 to 60 photos remain human-intensive and time-consuming.
With agentic AI, the appraisal process takes a data-driven approach. Once instigated, the agent contacts borrowers, validates property details, and schedules appraisals using appraiser availability, property characteristics, and borrower preferences.
This speeds up the process: at present, the current state typically takes 48 hours for an appraisal company to be assigned and sometimes a week or more to schedule an appraisal. Once scheduled, the agentic system determines whether an in-person appraisal, or a remote appraisal, or a 3D scan inspection capturing a property’s detailed 3D digital model, is best suited for the case. With continuous learning and feedback, agentic AI can make decisions around the type of appraisals, complete the respective appraisal forms, perform communication to stakeholders as needed.
Agentic AI can also prefill several of the required property data points such as the property dimensions, layout, number of rooms, by comparing it with historical appraisals on comparable properties reducing manual human inspection effort. What normally takes a week or more to schedule can be cut to hours, directly translating to a shorter cycle time and faster closing. - Mortgage preapprovals: Homebuyers seek multiple preapprovals, but only about half convert to actual loans. While preapproval signals serious intent, simply collecting minimal information isn’t enough to improve conversion.
The agentic system triggers when a borrower requests preapproval and goes beyond document collection. It verifies borrower credit and existing mortgages, calculates weighted interest rates based on loan balances and payments, and assesses overall creditworthiness. By taking a broader, data-driven view, the system ensures the preapproval decision reflects borrower risk and affordability. The agentic AI system doesn’t wait at every step but keeps on connecting with borrowers as needed, so that the borrower feels better engagement throughout the process. The system learns about the potential conversion based on past data and tailors the engagement with the borrower accordingly.
The business case for agentic AI is compelling. According to the Mortgage Bankers Association, the average cost to originate a loan exceeds $11,000. A large share of this cost comes from manual labor, rework due to errors, and time delays. Faster closings improve borrower satisfaction and increase loan pull-through rates, directly impacting revenue. For lenders, these savings represent a meaningful competitive advantage.
“The future of mortgage lending depends on combining innovation with responsibility. Agentic AI allows us to transform both the origination and servicing business by leveraging specialized AI agents across the process. This provides a faster, efficient and self-serviced approach for funding and servicing loans. By integrating technology with human oversight, we not only drive efficiency but also uphold fairness and trust — delivering long-term value for both borrowers and the broader financial system.”
A practical guide for mortgage lenders to adopt agentic AI
Lenders only control a fraction of the loan journey, often depending on third parties such as closing agents. Most loan origination systems (LOS) are legacy platforms, and they are critical and too complex to replace. For mortgage issuers to deliver real value, here is a practical guide to integrate agentic AI:
- Orchestrate internally: Focus on the parts you own. Deploy orchestrators on top of existing LOS systems to enhance current workflows instead of trying to replace the workflows.
- Coordinate externally: For everything outside lender’s direct control, use APIs and standard protocols to connect your orchestrators with external partners. Build bridges to the ecosystem so LOS-bound processes work together with third parties.
- Design for partial autonomy: Agentic AI should manage end-to-end tasks when possible, but recognize when coordination must pass outside the lender’s walls. Know when to direct, and when to collaborate.
This adoption isn’t just about plugging in new technology. A smart lender takes a measured approach, aligning the technology with business strategy and internal workflows, and putting the right controls in place.
Here are the core considerations:
- Compliance and regulation: Lenders must pay close attention to the risk of bias, especially with requirements like the Home Mortgage Disclosure Act. Regular audits are essential, and bringing on responsible AI experts helps ensure the system is fair and ethically sound. The explainability of agentic AI decisions, supported by audit trails and transparency, provides regulators and investors with confidence that lending practices are still fair, unbiased, and ethically sound.
- Technology landscape: Most lenders run on third-party LOS platforms. As vendors work on integrating agents into these systems, rollout will take time. Consider partnering with service providers that can build agents to work with your existing infrastructure. Above all, take a holistic view — don’t just chase the next use case; ensure your tech stack is future-ready.
- Security and privacy: Security must be central when building agentic AI. Assess frameworks and build guardrails for data handling at every step.
- Human in the loop: Autonomous does not mean unchecked. Limits are necessary — define where agents need approval for high-risk actions or access to sensitive information. Keep qualified people involved in critical decisions.
Adopting agentic AI in mortgages is about working with what you have, connecting responsibly with partners, and putting safeguards in place. Build for reality. Enhance what’s proven. Mitigate risk smartly. That’s how you deliver measurable results and avoid costly missteps.