The most useful starting point for mortgage AI is a recurring task that gets in the way of serving people—not another software subscription. On The Signal Podcast, Abdel Khawatmi, area manager at PRMG, joins Chris Grimes and Reuven Gorsht to explain how he uses meeting summaries, automated reminders, guideline tools and call coaching to support his origination business. His central lesson is that technology should create more room for personal conversations. Just as importantly, his experience shows why adopting AI eventually requires deciding what to stop paying for.
Conversation originally published September 8, 2026. Recap added October 24, 2026.
Watch the original conversation on YouTube.
Key Takeaways: Automate the Follow-Through, Not the Relationship
Abdel’s first useful AI application was modest: a meeting note-taker that helped him remember conversations and prepare follow-up. That starting point matters because it addressed an existing responsibility. He already reserved time to respond to people after meetings; AI made the information behind those responses easier to retrieve.
The broader system combines AI, conventional automation and human administration. Calendar reminders do not need generative AI to be valuable. A meeting summary does not become useful until someone reviews it and delivers the next step. A guideline assistant does not remove the need for mortgage judgment.
Three practical lessons run through the conversation: build around a specific operational problem, make tools easy for the team to find, and measure whether the workflow deserves to stay. Abdel also describes a deliberate reversal from collecting subscriptions to consolidating them. More technology was not automatically producing a better business.
Start With Meeting Follow-Up That Survives a Busy Day
Before adopting AI note-taking, Abdel depended on an evening time block to follow up with everyone he had met that day. The weakness was not a lack of intent. It was the distance between the conversation and the response, with urgent loan issues competing for his attention in between.
His early approach was to place a meeting summary into a chat project and specify the audience, his communication style and a short output length. For a real estate agent, that meant a concise message that sounded like him rather than a lengthy formal recap. The value came from preserving the substance of the discussion without reconstructing it from memory.
He later describes a more organized handoff: Zapier moves captured call information into a team communication channel, and the team uses the structured output for follow-up. The summary separates pain points, mortgage options and responsibilities for the borrower, agent and lender. That is more useful than an undifferentiated transcript because each participant can understand what comes next.
For another mortgage team, the practical application is to define the follow-up format before choosing the tool. Include the purpose of the conversation, unresolved questions, assigned actions and deadlines. Establish recording consent, approved storage and recipient rules before capturing calls. Internal coaching notes and sensitive borrower details should not travel automatically with an external recap.
Use Calendar Automation to Make the Human Team More Reliable
One host notices that Abdel’s system sent a meeting reminder even though the hosts had initiated the invitation. Abdel explains that calendar-connected automation supports reminders around scheduled meetings, while his administrators also receive the information they need to coordinate personal confirmation calls.
This is an important distinction in an AI discussion: some of his useful infrastructure predates his AI adoption. The benefit comes from connecting a calendar event to a dependable action, not from making every step intelligent. A reminder gives the recipient an opportunity to confirm, reschedule or resolve confusion before a time slot is wasted.
The administrative team remains part of that system. For a client consultation, a person may call to confirm who the borrower will meet and when. The automation supplies consistency; the team supplies context and reassurance. That combination can be especially helpful when the borrower has not yet developed a relationship with the loan officer.
An effective implementation should define which calendar events trigger reminders, who receives internal notifications and what happens after cancellation. It should also prevent duplicate or contradictory messages. The operational test is straightforward: does the process make appointments clearer while requiring less manual coordination?
Build Guideline Access Around Repeated Questions
Abdel describes a custom GPT supplied with mortgage reference materials, including the FHA handbook and Fannie Mae and Freddie Mac guides. He makes it accessible through a memorable domain so loan officers do not have to remember where the underlying tool lives. His aim is to reduce repeated questions that otherwise interrupt processors and managers.
The same idea extends to referral partners who may hesitate to ask a basic financing question. A familiar access point can make preliminary education easier. The strongest part of this approach is not the chatbot itself; it is identifying the questions people repeatedly need answered and giving them a consistent place to start.
A responsible application needs more structure than uploading documents once. Assign someone to maintain the reference library, identify document versions and test answers against the source material. Require the tool to distinguish agency guidance from lender overlays or investor requirements. A fluent response is not evidence that a scenario qualifies.
Guideline retrieval and income analysis also carry different risks. Finding a relevant policy passage can support a conversation, while calculating qualifying income can affect the file itself. Start with bounded research questions and clear escalation paths. Treat uncertain answers, missing facts and conflicting requirements as reasons to involve the appropriate mortgage professional.
Keep Borrower Data Boundaries Ahead of Convenience
The conversation directly addresses a tempting shortcut: uploading a borrower’s tax return into a general-purpose AI tool to obtain a quick calculation. Abdel warns his team against uploading identifiable client documents and distinguishes informal tools from enterprise systems adopted through the company.
He also discusses using a limited document excerpt without identifying information. The safer operational lesson is data minimization, not the assumption that cropping a document makes it safe. Financial information can remain sensitive even when a name is removed, and an excerpt may omit facts needed for an accurate analysis.
Before using borrower material, establish which systems the company permits, what information they may receive, how access is controlled and who reviews the result. A calculation should remain reproducible and subject to the lender’s normal verification process. An AI-generated explanation should clarify documented facts, not create facts or substitute for required third-party confirmation.
Similar boundaries apply to automated outreach. Abdel emphasizes using enterprise-level systems and company-approved tools when dealing with consumer contact and document processing to maintain compliance. Screening alone is not a complete compliance program. The channel, consent, message content and company policies all matter, particularly when automation moves from an internal productivity tool to direct consumer contact.
Use AI Preparation to Improve Referral Conversations
Abdel describes using automation to handle mundane tasks so he can prioritize 'belly-to-belly' facetoface conversations with the local community and referral partners. He notes that embracing AI systems allowed him to be tighter with partners and more available for in-person activities like open houses.
His referral-support work includes a branded AI audit that examines a partner’s online presence and produces strengths, weaknesses and action steps. He describes an immediate seven-day plan followed by longer checkpoints and performance tracking. This gives a meeting a practical focus beyond a generic request to exchange business.
For a loan officer applying the idea, a useful audit might identify inconsistent business information, an unclear contact path or educational content that is difficult to find. Those observations should be checked before being presented. An AI-generated score is an organizing device, not an independent measure of a partner’s business quality.
Preparation should help the loan officer ask better questions, not manufacture familiarity. Relevant public professional information and details a partner has willingly shared are a sounder foundation than assumptions about private circumstances. The meeting should ultimately explore the partner’s goals, client needs and service gaps.
Abdel’s rejection of an AI video avatar reinforces the point. He tried the approach but felt its movement and delivery did not represent him. His lesson is selective adoption: a tool can be capable and still be wrong for a relationship-driven brand.
Make Adoption Visible Through Coaching and Simple Access
Abdel does not describe team adoption as sending everyone a list of subscriptions. He uses his own production workflow as the demonstration. Loan officers can see how calls become summaries, how the team handles follow-up and how a tool fits the work they already understand.
He also subjects his own calls to AI coaching through a Claude project. The feedback flags behaviors such as interrupting, talking too much or missing an opportunity to handle the discussion differently. This positions AI as a review aid for the originator, not only as a mechanism for evaluating other employees.
A practical coaching program should use a defined rubric: whether the borrower’s objective was established, whether options were explained clearly and whether responsibilities were confirmed. Review the underlying call before accepting the model’s criticism. Consistent criteria make feedback more useful than a score with no explanation.
Access and incentives matter too. Abdel describes consolidating resources into a central location and rewarding a production milestone with a recording tool. His larger adoption argument centers on financial, time and relationship benefits. He also frames additional manual support as a business cost that must be funded. That is his management perspective, not evidence that AI necessarily reduces compensation or eliminates a particular role.
Build Versus Buy: Own the Workflow, Question the Subscription
Abdel strongly prefers building workflows around how his team operates rather than accepting an all-in-one platform by default. His concern is dependence: when too many functions live in one vendor’s environment, changing providers can become difficult. He also rejects tools that are too complicated for practical adoption.
Yet his position is more nuanced than building everything himself. He values integrations that reduce manual movement of information, including connections to the loan origination system. He acknowledges that buying can make sense when a vendor has solved a problem that would require engineering resources his team does not have.
His cancellation of Gamma illustrates the distinction. Once he had established a repeatable presentation style, he moved that work into an existing Claude project. The benefit was not only removing a subscription. It also meant one less platform for administrators and new hires to learn.
The broader lesson is to compare total workflow cost rather than subscription price alone. Count setup, training, maintenance, duplicate entry, review and exception handling. A homemade tool is not free if it depends on hours of troubleshooting, and a purchased platform is not efficient if the team avoids it.
Keep useful consolidation separate from unchecked automation. A system that generates client-facing financial information still needs appropriate controls for accuracy and distribution. Faster delivery is only valuable when the content and the audience are right.
A Monday-Morning Checklist for One Measurable AI Workflow
First, inspect your calendar and a recent week of work. Identify one recurring responsibility that slips, takes too long or repeatedly interrupts someone else. Abdel’s suggested starting points include past-client contact, internal guideline support and scheduled social content. Choose the one that addresses your actual bottleneck.
Second, write down the workflow before buying anything. State what triggers it, which inputs it needs, what output should appear and who owns the next action. Separate work that can be automated from work that needs a processor, assistant, manager or loan officer.
Third, set the boundaries. Confirm approved tools, permitted data, review requirements and a fallback when the automation fails. For past-client outreach, decide how replies reach a person. For a guideline assistant, define when the question must be escalated. For scheduled content, assign editorial approval.
Fourth, establish a baseline and run a limited pilot. Track the relevant outcome: completed follow-ups, time spent preparing recaps, unanswered questions or appointments requiring manual correction. Include rework and mistakes so that apparent speed does not conceal a new burden.
Finally, keep, revise or remove the workflow before adding another. Cancel redundant tools and document the process clearly enough for someone else to operate it. Reserve the recovered time for consultations, partner meetings or community activity. That closes the loop in Abdel’s approach: simplify the machinery so the people get more attention.
Frequently asked questions
What is a practical first AI project for a loan officer?
Choose a recurring task with a visible gap, such as preparing meeting follow-ups or keeping in touch with past clients. Abdel recommends examining your calendar and current processes first. Pilot one workflow, measure its results and expand only after it works reliably.
How can AI help write mortgage meeting follow-ups?
An approved note-taking tool can capture the discussion, and a structured prompt can organize the recap into objectives, open questions and next steps. Specify the audience and desired length. Review the message and remove information the recipient should not receive before sending it.
Do automated appointment reminders require AI?
No. Abdel describes calendar-connected reminders that were useful before his broader AI adoption. Conventional automation can trigger reminders and notify administrators, while people handle rescheduling, clarification and personal confirmation calls.
What should a mortgage guideline GPT include?
It should use approved, maintained reference materials with identifiable versions and a clear scope. Ask it to point to the relevant source, distinguish agency guidance from overlays and flag missing information. It should support research rather than issue final eligibility decisions.
Is removing a borrower’s name enough before uploading a tax document?
No. Other details can identify the borrower, and financial information may remain sensitive without a name. Use only company-approved systems and permitted data. Any resulting calculation still needs review under the lender’s normal process.
How does Abdel use AI to coach his own calls?
He describes a Claude project that reviews calls for behaviors such as interrupting, speaking too much and missing opportunities to improve the conversation. A team applying this idea should establish a consistent rubric and check the recording before accepting the feedback.
When should a mortgage team buy software instead of building a workflow?
Buying can make sense when the product offers needed integrations, controls or technical capabilities the team cannot reasonably maintain. Compare the full implementation burden with the cost of building, including training, upkeep and manual work between systems.
How can an AI brand audit help a referral partner meeting?
It can organize observations about the partner’s public business presence into a few concrete improvements and follow-up checkpoints. Verify those observations and use them to start a discussion about business goals. Do not present an AI-generated score as an objective valuation of the partner’s performance.
To talk with PRMG Got Mortgages, call 973-536-0276, visit GotMortgages.com, contact the team or build your mortgage plan.
Explore more Got Mortgages Insights and Abdel Khawatmi’s profile. Explore more conversations at the Got Mortgages Podcast website.
Author: Abdel Khawatmi with PRMG Got Mortgages | 973-536-0276
This article is general mortgage information only. It is not a commitment to lend, a rate quote, or legal or tax advice. Loan approval, terms and pricing depend on a full application, credit review and property details, and market conditions change. Talk with a licensed loan officer and your own legal or tax professional about your situation. NMLS #1712023 | PRMG NMLS #75243.