In This Article
- What a Loan-Guideline Agent Does (and Does Not Do)
- How the Agent Works Inside Your Microsoft 365 Tenant
- Why Lenders and Credit Unions Are Building These Agents Now
- ABT MortgageGuide: A Purpose-Built Example
- The Three Roles That Benefit Most
- The Governance Configuration That Makes This Work
- What Examiners Will Ask (and How This Helps You Answer)
- Frequently Asked Questions
Loan officers do not spend most of their day talking to borrowers. A large portion of it goes to hunting. They hunt through Fannie Mae selling guides, Freddie Mac bulletins, internal credit policy documents, and investor overlay memos to answer one question: does this loan fit?
When a processor pulls up a file on a non-warrantable condo and the answer depends on which overlay applies to which investor, the search can take 15 minutes. On a busy day, that happens a dozen times. Nothing about the search requires professional judgment. The judgment happens after the answer is found. The search itself is a bottleneck.
A loan-guideline agent inside Microsoft 365 Copilot closes that bottleneck. It knows your document corpus, returns an answer with a citation in seconds, and every query it handles is logged automatically in your Microsoft Purview audit trail. Your loan officers get more done. Your compliance team has a record. Your examiners have something to look at.
ABT builds these agents for mortgage lenders and credit unions as part of the Microsoft 365 tenants we manage. ABT MortgageGuide is our purpose-built example. This article explains what a loan-guideline agent actually does, how it works inside your existing Microsoft 365 environment, and what the governance story looks like for a CFPB- and GSE-regulated shop.
What a Loan-Guideline Agent Does (and Does Not Do)
A loan-guideline agent is not a decision-maker. It does not approve or deny a loan. It does not replace underwriter judgment. What it does is retrieve the right guideline text, from the right source, and hand it to the staff member who needs it, fast enough to matter.
Here is the before-and-after in concrete terms.
A loan processor working a jumbo file needs to confirm whether a gift fund can cover the entire down payment for a specific investor product. She opens the investor overlay PDF, searches for "gift funds," reads through four pages, checks whether the client's product code matches, and confirms the answer. Total time: 12 minutes. She also makes a judgment call about whether she found the right version of the overlay.
She asks the Copilot agent: "Can a gift fund cover 100% of the down payment on a jumbo purchase for Investor X?" The agent returns the relevant passage from the current overlay document, with the document name and section number. Total time: 30 seconds. The answer is grounded in a document she already has access to. The citation is right there.
The agent does not make the credit decision. It finds the rule. That is the job.
How the Agent Works Inside Your Microsoft 365 Tenant
Copilot agents built with Microsoft Copilot Studio run inside your Microsoft 365 tenant. They do not call external AI services with your document content. They access only the files and data that the user querying the agent already has permission to see.
That matters for regulated lenders. A processor who does not have access to a restricted internal policy document cannot get information from that document through the agent. The agent respects the same permissions your SharePoint and OneDrive already enforce. Sensitivity labels on your policy documents control what the agent surfaces. If a document is labeled confidential, the agent treats it that way.
How the Query-to-Answer Flow Works
The document corpus. Your agency guidelines (Fannie Mae, Freddie Mac), investor overlays, internal credit policy, and underwriting memos live in SharePoint. The agent is grounded on those documents. When it answers a question, it draws from your own files, not from a generic training dataset.
The query. A loan officer or processor asks a natural-language question. "What is the maximum DTI for a conventional conforming loan with a credit score between 680 and 699?" The agent searches the corpus, finds the relevant section, and returns the answer with a citation.
The citation. The agent returns the document name, section, and the specific passage it used. The loan officer can click through and verify. An underwriter reviewing the file later can see what source the processor consulted.
The audit log. Every query the agent handles is captured in Microsoft Purview Audit. If a GSE or examiner asks how a guideline determination was made on a specific loan, the interaction record exists.
Why Lenders and Credit Unions Are Building These Agents Now
The document problem in mortgage is not getting simpler. Agency guidelines change. Investor overlays change. Internal credit policy changes. Keeping staff current on all of it is a training challenge. Keeping guideline lookups accurate is an operational challenge.
Microsoft Copilot Studio is the platform that lets lenders build agents grounded on their own document corpus. First West Credit Union deployed Microsoft 365 Copilot agents for retail lending and mortgage renewal workflows, reaching 93% adoption and reporting more personalized member service with less time spent searching documents.
For U.S. mortgage lenders operating under CFPB oversight and Fannie Mae and Freddie Mac requirements, the governance configuration matters as much as the speed. The agent is only useful if the compliance team can stand behind what it surfaced. ABT's positioning on this is straightforward: the technology works. Getting the governance right before staff uses it on borrower files is what ABT brings to the deployment.
ABT MortgageGuide: A Purpose-Built Example
ABT MortgageGuide is an agent built to demonstrate what a production loan-guideline tool looks like for mortgage lenders running on Microsoft 365.
The outcomes MortgageGuide is built to deliver:
- Faster guideline lookups for loan officers and processors, with citations tied to your actual document corpus
- Reduced back-and-forth between loan officers and underwriters over guideline interpretation
- A Microsoft Purview audit trail on every query, so you have documentation of how guideline determinations were made
ABT manages the Microsoft 365 tenants for the lenders we work with. That means the Copilot configuration, the SharePoint document library structure, the sensitivity label setup, and the Microsoft Purview audit configuration are all part of the same engagement. The agent does not get handed over and forgotten. Before any of this is useful, your tenant needs a baseline configuration in place. Our guide on what Microsoft 365 Copilot Chat personalization settings actually do covers the foundational setup every deployment starts from.
The agent does not get handed over and forgotten. The Microsoft 365 configuration that makes it work is part of the same engagement ABT already manages for your tenant.
The Three Roles That Benefit Most
Loan Officers
Loan officers need fast, confident answers on product fit. Is this borrower eligible? Does the property qualify? Can they use a gift fund? A loan-guideline agent reduces the time between question and answer from minutes to seconds. More importantly, it reduces the cognitive load of remembering which overlay applies to which product. The agent knows.
Processors
Processors handle loan conditions. A lot of those conditions are guideline-driven: verify that the documentation meets investor requirements, confirm that the property appraisal meets agency thresholds, check that the co-borrower's income is calculated the way the investor expects. Each of those checks is a document lookup. A processor with a loan-guideline agent spends more time building the file and less time searching for the rule.
Underwriters
Underwriters benefit differently. They are not the primary users of the lookup function because they already know the guidelines. Their benefit is in review time. When a processor has used a guideline agent and the answer is cited in the file, the underwriter does not have to re-verify the interpretation. The source is already documented. That cuts review time and reduces back-and-forth conditions.
The Governance Configuration That Makes This Work
Speed matters. The governance configuration is what makes the speed usable in a regulated environment. Four things need to be in place before a loan-guideline agent handles its first live query on a borrower file.
Every query and response is captured. This is not automatic at the right scope for agent interactions. ABT configures this as part of the deployment.
Your policy documents and investor overlays should already carry Microsoft Purview sensitivity labels. The agent inherits those labels. If your sensitivity label configuration has gaps, those gaps show up in the agent's behavior.
The agent sees what the querying user can see. If you want processors to see investor overlays but not underwriting policy exceptions, the SharePoint permissions need to reflect that before the agent goes live.
The agent is only as current as the documents in the corpus. If you add a new investor overlay to SharePoint and forget to retire the previous version, the agent might return conflicting information from two versions of the same document.
None of these four requirements is specific to AI. They are standard Microsoft 365 governance practices. The agent makes them more visible because the stakes of getting them wrong are higher when staff are using the output on live borrower files.
What Examiners Will Ask (and How This Helps You Answer)
CFPB examiners and GSE quality control reviewers are already asking questions about AI tools in mortgage operations. The questions they are asking are operational, not theoretical. "What AI tools are your staff using?" "How do you ensure that guideline lookups are accurate?" "What documentation exists of how compliance determinations were made?"
A loan-guideline agent built inside Microsoft 365 Copilot gives you concrete answers to those questions.
What You Tell Your Examiner
- "Our staff uses a Copilot agent grounded on our investor overlays and Fannie Mae and Freddie Mac guidelines."
- "The agent runs inside our Microsoft 365 tenant, configured and managed by ABT."
- "Every interaction is logged in Microsoft Purview Audit. We can pull that log for any loan."
- "The agent does not make credit decisions. It retrieves guideline text with citations. Staff judgment closes every file."
The FFIEC has not published a specific AI examination playbook as of this writing. The U.S. Treasury's Financial Services AI Risk Management Framework, released in February 2026, contains 230 control objectives and is non-binding. But those controls may inform supervisory expectations as regulators develop formal AI examination procedures. Institutions that document their AI governance posture now, before examiners develop formal procedures, are in a stronger position than institutions waiting for the playbook to arrive. The Purview audit configuration, eDiscovery, and retention policies that apply to all Copilot interactions, including agent queries, are covered in detail in our article on Copilot memory and data governance gaps that banks and credit unions need to address now.
If you want to understand where your current Microsoft 365 configuration stands before adding any AI capability, start with ABT's AI Readiness Assessment at getmygrade.myabt.com. It surfaces the gaps in your Purview, Entra ID, and SharePoint configuration before those gaps show up in an agent deployment or an examination.
Ready to See Whether a Loan-Guideline Agent Fits Your Operation?
ABT manages Microsoft 365 tenants for more than 750 financial institutions, including mortgage lenders regulated by CFPB and the GSEs. Talk to an ABT expert to understand whether a loan-guideline agent makes sense for your document corpus and compliance posture.
Frequently Asked Questions
No. The agent retrieves guideline text and returns it with a citation. The credit judgment is still made by your underwriters. The agent speeds up the lookup so underwriters spend their time on the judgment, not on searching for the rule.
The agent returns results from whatever documents are in your SharePoint corpus. If you have an outdated overlay in SharePoint alongside the current version, the agent might return either one. The solution is a clear version-control process on your source documents. ABT configures the initial corpus, but the institution is responsible for keeping those documents current, the same as any other SharePoint library.
Microsoft Purview Audit captures the interaction records. Your IT administrator or compliance officer can search those records using Microsoft Purview eDiscovery. The institution decides how to manage and retain those audit logs based on its own record retention policies and applicable regulatory requirements.
You can build your own agent using Microsoft Copilot Studio. ABT MortgageGuide is an agent ABT built and can deploy for lenders we work with. For institutions that want a purpose-built starting point and do not want to build from scratch, MortgageGuide is the faster path. For institutions with a development team and specific workflow requirements, ABT can help design and deploy a custom agent instead.