Mortgage AI adoption works best when leaders build a repeatable process before adding more tools. In this August 3 episode of Settling Scores, cohosts Abdel Khawatmi of PRMG Got Mortgages and Eric Levin, cofounder of Model Match, argue for infrastructure, clear instructions and practical next steps—not another collection of disconnected experiments. AI should give originators speed and clarity while preserving the personal brand and judgment that make their relationships valuable.
Episode originally published August 3, 2026. Archive recap added October 24, 2026.
Watch Settling Scores Episode 11: AI Adoption & Housing Wire AI Summit Preview on YouTube.
Key takeaways: build the process before expanding the stack
- Define the business objective within sales, service or support before choosing a tool.
- Make useful data accessible in the workspace an originator already uses.
- Pair access with a standard operating procedure, a responsible teammate and a next action.
- Describe the audience, purpose, format and tone in the prompt.
- Save effective instructions and expand a successful workflow incrementally.
- Use automation to support an authentic local presence, not replace it with interchangeable content.
A summit preview focused on the producing originator
The conversation looks ahead to the HousingWire AI Summit in Dallas. Eric identifies Tuesday, August 11, as Abdel’s anticipated stage appearance. Abdel describes a ground-up AI discussion that would connect a producing originator’s perspective with marketing leadership and executives. The emphasis is on what people can actually adopt after hearing about the possibilities.
Abdel expects his own changing experience with AI to shape that discussion. His understanding of the opportunity has evolved through experimentation rather than a single permanent breakthrough. That matters for leaders: a workflow that seemed advanced earlier may become a starting point later, but users still need a clear path into it.
His intended message is not simply that companies should impose more restrictions or distribute more software. Governance matters, he says, but leaders also need to understand how an originator thinks and works. He hopes attendees will leave with actionable planning and a clearer view of how AI can support their businesses. These are expectations for the upcoming session, not a report of what occurred there.
Infrastructure first: turn a recording into a usable handoff
Abdel’s recording-device example makes his infrastructure-first approach concrete. He describes connecting the device through Zapier so his team can receive notifications, and discusses connections with ChatGPT and Claude. The important feature for his workflow is not merely possessing a transcript. It is arranging what happens after the conversation.
He distinguishes summary styles by purpose. A referral-partner call needs a different summary from a client consultation. A useful system should reflect that distinction instead of producing an identical block of notes for every interaction. Leaders can make the process easier to adopt by deciding the desired output before handing the tool to originators.
A suggested implementation is to define the summary’s audience, the commitments it should identify and who reviews it before any follow-up. Begin with a sanitized example and confirm that the notification reaches the intended teammate. Recording consent, approved access and retention rules need separate review; a connected workflow is not itself evidence that every privacy or compliance requirement has been satisfied.
This is the blueprint Abdel wants leaders to provide: establish the structure, make the handoff understandable and put guardrails around execution. A tool becomes more useful when the team knows what information it should produce and what to do with that information.
Model Match access and the paradox of more options
Eric describes Model Match opening its API and releasing an MCP connection. His practical explanation is that information can become accessible inside a user’s preferred workspace rather than requiring every task to happen within the product’s own interface. Later, he specifically discusses users connecting through Claude or GPT.
That decision creates a tension he addresses directly. More access can reduce friction, but more options can also overwhelm people. Someone may repeatedly try an interesting tool without adopting it deeply enough to create a useful result. Opening another connection does not automatically solve that behavior.
Eric therefore pairs technical access with consultation: help people select an option, commit to it, adopt it and act on it. Abdel reinforces the accessibility point by describing less friction in reaching Model Match information and a newer interface that makes common search directions easier to understand. He also discusses a CRM feature as an alternative to circulating exported spreadsheets.
The spreadsheet objection is operational, not cosmetic. When administration, business development and growth work happen in separate copies, the team can lose a shared view of the work. A suggested improvement is to identify where the current record lives, who updates it and how the next person knows an action is complete. Neither an API nor a CRM can substitute for those decisions.
Choose the business model before asking for a list
Abdel organizes the starting question around sales, service and support. Even within sales, he asks whether the business is consumer-direct, referral-based or a combination. Different models need different information and different next steps. A retention-oriented strategy should not be assumed to fit a branch built around referral relationships.
His referral-research example begins with filters such as closings, buyer-side or seller-side activity, and social presence. He then describes evaluating the resulting prospects against a defined ideal profile: whether they brand themselves, publish online, remain accessible and present consistent contact information across sources.
The important shift comes after the list exists. Abdel asks what the team will actually do with it. Data has to support a decision about fit, further research or an appropriate conversation. An online inconsistency may call for verification; limited social presence may reveal an opportunity to help rather than a reason to dismiss someone automatically.
A suggested local application for an Ocean or Monmouth County team is to define the referral relationship it can genuinely support before running research. Specify the communities served, the type of educational support available and the relevant business focus. Review suggested contacts for current details and practical fit. A large list is not the same thing as a thoughtful relationship strategy.
Write the SOP that makes delegation possible
Abdel argues that teams often connect tools without providing step-by-step instructions. He wants an originator to have a procedure covering the route from logging in to applying filters and producing the final usable output. He also calls for something that tracks whether the process is working.
That changes adoption from an individual technical challenge into a team capability. An originator who enjoys experimentation can work directly with the system. Someone else can delegate research to a trained teammate, provided both understand the criteria and the expected deliverable.
For a suggested first SOP, describe the purpose, approved data source, required filters, verification steps and destination for the result. Include what to do when the output does not fit the request. Finish with an explicit handoff: who receives the work and what decision or conversation follows. Instructions should be usable by someone who did not build the automation.
Keep local content personal, not boilerplate
Abdel describes content work that combines Canva, a specific originator’s brand and Buffer distribution. His emphasis is on adapting material to a niche and a local market, including recognizable places, rather than repeating generic designs. The technology supports distribution, but the usefulness comes from the specificity of the message.
For a suggested local educational piece, start with a question the team can answer accurately for its audience. Choose relevant community context, verify references and keep the originator’s own voice. Scheduling a post is the end of production, not proof that it offers something distinctive. Review the published version as carefully as the draft.
Eric’s broader point is that a polished AI-generated asset can be an entry point. The next question is what useful step can be added without overwhelming the person using it. Expansion should follow a working process rather than become another reason to abandon one.
Better prompts specify purpose, audience and output
Abdel’s detailed prompting example explains why “make a flyer” is too vague. He specifies what he wants to communicate, why it matters, who should receive it and how the result should look. His example includes a two-column layout, a call to action at the bottom and a tone that combines authority with empathy toward referral partners.
Both hosts favor dictation as a way to capture that context. The goal is not a longer prompt for its own sake. It is to give the assistant information that would otherwise remain in the originator’s head. Eric notes that some difficulties with the Model Match connector arose from prompting rather than the connection itself.
They also identify an output that needs independent checking: QR codes. Their practical advice is to obtain a working code separately instead of trusting a generated image. A suggested review includes scanning the final code and checking its destination before distribution. Attractive formatting should never substitute for functional testing.
Eric’s closing prompt technique is especially reusable. Once a result is satisfactory, ask the assistant for the prompt that would have produced it in one run, then save it. Review that instruction before reuse, update the changing inputs and preserve the constraints that made the output useful. This turns experimentation into a teachable process.
Suggested actions: make the next adoption step small
- Select one responsibility. Choose a recurring sales, service or support task with a clear recipient.
- Define the output. State the audience, required information, format and next action before connecting tools.
- Document and test. Have a teammate follow the SOP with approved, non-sensitive examples.
- Review actual execution. Check whether the output was accurate, reached the right person and supported the intended follow-up.
- Expand deliberately. Add the next connection only when the current process is understandable and repeatable.
This suggested sequence applies the episode’s central distinction: AI can improve speed and clarity, but the originator still supplies purpose, judgment and an authentic relationship. A smaller workflow that people use is a stronger foundation than an ambitious system they cannot explain.
Frequently asked questions
Who hosts the August 3 AI summit preview?
Abdel Khawatmi of PRMG Got Mortgages and Eric Levin, cofounder of Model Match, cohost this Settling Scores discussion about practical mortgage AI adoption.
Does this episode recap the HousingWire AI Summit?
No. It previews the Dallas summit and Abdel’s anticipated August 11 appearance. His proposed themes and hoped-for takeaways should not be read as confirmed event outcomes.
What does infrastructure-first AI adoption mean?
Define the workflow, handoffs and guardrails before distributing tools. Abdel wants originators to receive a usable process rather than build every connection themselves.
Which Model Match capabilities are discussed?
The hosts discuss MCP access, an open API, a newer interface and a CRM feature. Eric also describes connecting Model Match information through Claude or GPT.
Why is an SOP important for adoption?
It makes the work repeatable and delegable. Someone other than the builder should understand the filters, expected output, review requirements and next action.
What should an effective AI prompt include?
State the task, purpose, audience, format and tone. Add relevant constraints, review the result and save revised instructions when they produce useful work.
Should an AI-generated QR code be trusted?
Not without testing. The hosts recommend obtaining a working code separately; scan the final published version and verify that it reaches the intended destination.
How can a team avoid generic automated marketing?
Use a specific audience, an authentic voice and verified local context. Distribution tools can schedule material, but they do not supply the judgment that makes it relevant.
How can I contact Got Mortgages?
Call 973-536-0276 or visit GotMortgages.com to begin a conversation with PRMG Got Mortgages.
Make speed and clarity serve the relationship
To talk with PRMG Got Mortgages, call 973-536-0276, visit GotMortgages.com, contact the team or build your mortgage plan.
Explore more Mortgage Intel 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.