The Signal is a weekly podcast hosted by Chris Grimes and Reuven Gorsht, exploring how AI, automation, and new operating models are reshaping high-stakes industries like mortgage, energy, healthcare, and beyond. Through candid conversations with founders, operators, lenders, and technology leaders, the show digs into what is actually changing on the ground — the workflows being reinvented, the bottlenecks being broken, and the decisions leaders are making as their industries adapt.
Each conversation is designed for practical insight: what is working, what is still hype, and what the next wave of industry transformation looks like.
Site
RSS
Apple
Data updated on 19/10/2026
Recent rankings
Latest chart positions across Apple Podcasts and Spotify rankings.
Shared links between episodes and podcasts
Links found in episode descriptions and other podcasts that share them.
Loan officers building their own customer relationship management systems, or CRMs, with AI was a recurring story at the Digital Mortgage conference in September 2026, celebrated as putting technology in more hands. Sneha Pankaj, Senior Director of Product Management, Mortgage at Cotality, came up through engineering and sees two red flags. She works through them, and the habit that keeps her own assistant useful, with Chris Grimes and Reuven Gorsht.
What context as code means
Sneha noticed she was re-introducing herself to her AI assistant every day: her role, her product, the problem she was trying to solve. So she stopped. She now commits her call transcripts, demos, pitches and roadmap conversations to a repository, the version-controlled file store software engineers keep their code in, and the assistant draws on all of it from a one-line prompt such as a request for talking points before a customer meeting. She calls this context as code. The idea came from BMAD, the Breakthrough Method for Agile AI-Driven Development, an open-source framework that gives each role in building software, from product manager to architect to tester, its own specialized AI agent and feeds it written context. She says the method has held up for six months and she is piloting it with colleagues in product, marketing and sales.
Who checks what the machine did?
In mortgage lending, Reuven Gorsht says, one unchecked AI error could cost hundreds of thousands of dollars plus the lender’s reputation. He has seen how quietly it happens. Sneha Pankaj’s answer is evals: a bank of questions with known answers that checks an AI system’s output the way a quality-assurance reviewer would. She says banks already use them. Chris Grimes calls the same need a governance layer, the check on whether an agent is allowed to act at all. He argues the plumbing is arriving before that check, through tools like Salesforce’s headless CRM and MCP servers, the connectors that let AI tools drive business software. That leaves every lender with a practical question: who approves a loan officer’s home-built CRM before it holds customer data?
What was said
Sneha Pankaj says "loan officers are in the business of building relationships, not building software," and a CRM is a store of personal data that most of its new builders were never taught to protect.
Chris Grimes argues that the systems lenders run on were not built for AI, so nothing yet checks whether an agent is allowed to do what it is doing.
Reuven Gorsht holds that most people in a lender are operators, not builders, and that the job is finding the few who are.
Reuven also holds that you can only catch an assistant's mistake if you have done enough of the work yourself to see the answer is wrong.
Chapters
0:00 Meeting Sneha at HousingWire
1:23 From programmer to product at Cotality
4:45 Context as code with BMAD
9:31 When Claude has too much context
13:32 How much should you trust AI?
15:20 Evals and guardrails
17:44 The AI governance layer
20:03 Loan officers' CRM red flags
22:15 Chris's AI-built loan documents
23:23 Should everyone be a builder?
34:39 Where AI adoption stands
37:54 AI five years from now
Mentioned
Cotality - property data and analytics company, formerly CoreLogic - https://www.cotality.com
BMAD - open-source framework that gives each role in software development its own AI agent - https://github.com/bmad-code-org/BMAD-METHOD
Claude - AI assistant that can work across a folder of files - https://claude.ai
Visual Studio Code - widely used code editor that can host AI assistants - https://code.visualstudio.com
Salesforce - CRM platform with a headless CRM and an MCP server for connecting AI tools - https://www.salesforce.com
Your hosts
Chris Grimes is CEO of FundMore.
Reuven Gorsht runs Deeded and The Variable.
Your guest
Sneha Pankaj is Senior Director, Mortgage Data Solutions at Cotality, formerly CoreLogic — the #1 provider of property data and analytics for real estate, mortgage, and insurance. She spent her early tenure at Cotality leading engineering before moving into product leadership, and today she’s taking a 0-to-1 mortgage retention product from business case to commercialization, in a market where lenders lose up to 80% of their book of business after closing. She’s also been a driving force behind AI adoption at Cotality — leading company-wide AI experimentation efforts and pushing AI fluency beyond engineering into product and sales. She brings a systems thinker’s discipline to product: roadmap thinking, client context, and decisions all committed to the same repository as the code.
Applying for a mortgage inside ChatGPT: what lenders lose
Chris Grimes went looking for a raincoat before a trip to Scotland, typed what he wanted into ChatGPT, and bought one without ever opening a shop's website. Applying for a mortgage inside ChatGPT is the same move with three more zeroes on it, and it is already live. Better now takes a mortgage application without the borrower leaving the chat window, and at the other end of the same business the model companies are pushing into the parts of lending that were supposed to be too complex and too regulated to touch.
WHAT CHANGED WHEN THE MORTGAGE APPLICATION MOVED INSIDE CHATGPT
The enterprise end moved first, and quietly. Google announced a partnership with a large financial institution, Microsoft has done the same, and Anthropic is doing it too. Reuven Gorsht's reading of why is unsentimental. Companies valued in the trillions after four or five years have to show a buyer that a dollar in returns three or five or ten, and the strength of the core model alone no longer does that, because most work gets done perfectly well on the cheaper models. So the labs are hiring the people who already have the relationships. Reuven has watched executives from SAP, Oracle and Salesforce move across to companies like Anthropic, arriving with a Rolodex and a familiar face at the office of the CIO.
The consumer end moved more loudly. Better's application runs inside ChatGPT, so a borrower starting a mortgage there never opens the lender's own interface. Chris expects a dedicated consumer finance surface to follow, with a connector like Plaid pulling the full picture together so the assistant can give advice against real balances rather than guesses. His worry is not that the labs want to own the application layer. It is that they are dissolving it, folding what used to be somebody's product into something the assistant simply knows how to do.
WHY THE ADVICE LAYER IS THE EXPOSED ONE
Reuven's argument is that consumers take the path of least resistance every time, and that the evidence is already in. He watches Amazon vans on his street while the stores in his local plaza close. He bought his wife a Tesla, a purchase in the mid sixty thousands, entirely through an app, and has never met anybody from the company. Keynote speakers at the New England Mortgage Bankers Association put the same point to the industry more bluntly, which is that the trade says borrowers want a relationship when what they actually want is the lowest rate.
The Moat Is Not the Software
Tuesday, September 29, 2026 • Duration 43:33
Scott Larter runs BrokerBot on licensed Canadian property data, and he came to mortgage from the data side, at Opta, Verisk and FCT. Forty-four minutes in, he puts a three to five year clock on his own company's defensibility and names what he thinks is actually holding the door shut. Chris Grimes and Reuven Gorsht push on what is left to defend once a competitive analysis that used to take a team weeks takes twenty minutes and a coffee, and once the same intelligence sits in everybody's pocket. What the three of them land on is narrower than what the industry has been selling, and none of it is the software.
Why AI is not a competitive advantage in mortgage tech
In 2018 Scott Larter put a big orange booth on a trade show floor in Montreal, knew one person in the room, and was there to find out whether property data had any place in mortgage lending at all. Eight years later he runs BrokerBot, and he is arguing about a different question. What counts as a competitive advantage in mortgage tech now that the cost of building the product has fallen to almost nothing?
Scott came to mortgage from the data side, at Opta, Verisk and FCT, which is why he treats a broker's client book as a dataset in an industry that has always treated it as a rolodex. He is talking to Chris Grimes and Reuven Gorsht, who between them have built a lending decisioning platform and a real estate closing business. This page sets out what the three of them worked through and which questions they left open. What they conclude is on the tape.
WHY THERE ARE SUDDENLY HUNDREDS OF VENDORS
Reuven's framing is that the economics of starting a software company have inverted. When he started MoveSnap and when Chris started FundMore, you validated an idea hard first, because you were about to spend at least a million dollars in cash or in sweat equity. Now anybody can build a working prototype in an afternoon, so the constraint has moved.
He brings a figure back from a panel of venture investors he sat in on a few weeks before recording. Their position was that they are still funding roughly the same number of deals a year that they always have, while the denominator underneath that number has grown by something like a hundredfold. Scott's experience points the same way from the other side of the table. In his data years he was approached by ten to twenty startups a year looking for Canadian property data to build on, and he is blunt that cheap Canadian property data does not exist in this country.
The AI Edge: Transforming Non-QM Lending
Tuesday, September 22, 2026 • Duration 26:35
HOW DO YOU AUTOMATE UNDERWRITING WHEN THERE IS NO RULEBOOK
Here is the kind of question Sarah Gonzalez was willing to hand to software. In the example she uses, a matrix calls for an 800 FICO on a million-dollar loan and caps the loan-to-value at 75, and the file in front of the underwriter arrives at 76. That one has an answer before anyone reads the file. Almost nothing else in her book does, because almost everything Logan Finance writes is non-QM, and there is no standard guideline set sitting there waiting to be encoded.
Gonzalez is Chief Innovation Officer at Logan, and she has had to answer the automation question in production rather than on a panel.
CAN YOU AUTOMATE UNDERWRITING WITHOUT AUTOMATING THE CREDIT DECISION
Logan built a tool called Aura. A partner supplies the baseline rules engine, and the rules themselves were written in-house and belong to Logan. That distinction is worth holding onto, because renting an engine and owning the rules inside it is a different asset from renting the whole thing.
What Aura does is narrow on purpose. It reads hundreds of pages of guidelines so that an underwriter does not have to page through matrix after matrix, it reports what sits inside the box and what sits outside it, and then it stops. It does not approve anything and it does not decline anything. Gonzalez is strict about that boundary in a way most vendors selling into this market are not.
If the tool is already reading the guidelines, does the judgment itself move to the machine within a year or two?
WHY SHE DOES NOT START WITH THE ORG CHART
Her panel at the HousingWire AI Summit was about the future org chart, and her answer inside her own organization is that she does not start there. She starts with culture and with the experience she wants the customer or the employee to have, then asks which parts of that work can be automated and which parts require judgment.
WHAT SHE WANTS FROM VENDORS, AND IS NOT GETTING
Stop Training, Start Marketing: AI Adoption in Mortgage Lending
Tuesday, September 15, 2026 • Duration 35:55
WHY WON'T MY LOAN OFFICERS USE THE AI TOOL WE BOUGHT
Every conversation about AI adoption among loan officers eventually arrives at the same complaint. The platform works, the rollout happened, and the login numbers are flat. Lindsey Ricciardelli has a version of that story with a phone call in it.
She had spent almost two years on the platform. The friction was designed out, the assistant worked, and compliance had signed it off. She switched it on for every loan officer, sent an email explaining what it was, and waited.
Then her phone rang. A producer wanted to know who Casey was, and why she was calling his leads.
Casey is an AI assistant. When a borrower's credit gets pulled, or the data shows a refinance opportunity, Casey calls them, and when they are ready she transfers them straight to the loan officer's cellphone. The loan officer on the phone to Lindsey had not read the launch email. Neither, it turned out, had almost anyone else.
That call is the reason this episode exists. Adoption is usually handed to training: book the session, take the attendance, report the completion rate. Lindsey's argument, made on the HousingWire AI Summit stage in Dallas in August, is that training is the wrong department to hand it to. Getting a salesforce onto a new platform is a design problem first and a marketing problem second. What follows is what she built, what she left out, where the resistance came from, and why she still has no adoption number to show you.
DESIGN THE FRICTION OUT BEFORE YOU MARKET ANYTHING
The order matters, and she is direct about it: you cannot market your way past a design flaw. Before any campaign ran, the person building the platform spent months removing the reason a loan officer would bounce off it.
Adopt AI or Take a Compensation Cut
Tuesday, September 8, 2026 • Duration 54:25
Today we have a guest on the show! Abdel Khawatmi is an area manager at Paramount Residential Mortgage Group who started using AI because a note-taker showed up uninvited in a Zoom call and he liked the recording. Two years later he has a custom GPT sitting behind a domain his loan officers can just type, a call-grading project that tells him he talked over someone, and a rule that has nothing to do with any of it: build it, don't buy it.
The part nobody asks him about is the cancelling. He describes the early phase as being a kid in a candy store, saying yes to everything, until a revenue manager asked what all of it was for and he did not have an answer. Four months ago he took a pass through the stack and cut it.
WHAT YOU'LL TAKE AWAY
- Why the fix for a loan officer drowning in tools is usually one fewer tool
- The adoption mechanic that is not a training session, and what it costs him per loan officer
- Where the line sits on a machine speaking as you
- What he would do first on Monday, in the order he would do it
- Why automating the follow-up sent him back to open houses and chambers of commerce
CHAPTERS
(00:00) Open
(00:24) Meeting at the AI Summit
(01:47) Eight years at PRMG
(04:24) The note-taker that started it
(06:22) Efficiency was the aha moment
(09:40) A Tuesday, hour by hour
(14:05) The art of delegation
(18:27) Compliance and the tax return rule
(23:38) Getting a region to adopt
(30:05) Getting past the fear
AI Business Insights from Startup to Success
Tuesday, September 1, 2026 • Duration 37:12
Ninety percent of the room raised their hand. The question was who here is building something, and the room was a Claude meetup in Toronto — three hundred people, with four or five hundred more on a waitlist. Lawyers. Doctors. Everyone building something on a weekend.
Chris Grimes and Reuven Gorsht start there and end up somewhere less comfortable: if anyone can build the product, the product is not the business.
They work through what that means for AI startups, for the app layer sitting between the frontier labs and the customer, and for mortgage lending specifically. Chris rebuilt a top-three fitness app in ninety minutes on a Sunday. A tool set that needed Claude Code four months ago is now a feature of the model provider. Half a trillion dollars went into startups in the first half of 2026, and roughly 43 cents of every dollar went to two companies.
If the product can be built over a weekend, the product was never the business. So what is? That is the question the rest of the episode is trying to answer, and in regulated lending it turns out to have a different answer than it does anywhere else.
What you'll take away
What the frontier labs absorbing the app layer does to the companies funded to sit there
Why lenders do not shop a vendor list, and what actually gets a vendor into a budget
Where the margin turns out to be, and why it is not in the AI business
What makes auditability, rather than capability, the thing that decides who wins in lending AI
Chapters
(00:00) A maxed-out AI week
(01:16) Three hundred people at a Claude meetup
(04:06) A great time and a terrible time to build
(06:08) Why lenders don't shop vendors
(09:29) A product mid-flight, and testing distribution first
(10:46) Half a trillion in funding, and where it went
AI vs. The Human Element: Balancing Trust and Technology
Tuesday, August 25, 2026 • Duration 43:45
After a week of AI conferences, vendor demos, and big promises, Chris Grimes and Reuven Gorsht are asking the questions lenders actually need answered.
In this episode of The Signal Podcast, Chris and Reuven compare notes from the HousingWire AI Mortgage Summit and the TMC event, where AI tools were everywhere — but clear answers were harder to find. Across demos and conversations, three questions kept coming up: which model is powering the tool, how can compliance explain it to an examiner, and can a lender train it on their own data?
The conversation moves from borrower trust and chatbot adoption to model risk, vendor due diligence, and the uncomfortable reality that traditional software assessment may not work when an AI vendor can change models overnight.
In this episode
Why borrower trust in AI may be falling even as AI investment rises
What lenders should ask before signing an AI vendor contract
Why point-in-time vendor questionnaires are breaking down
How compliance teams should think about explainability and model changes
Where AI works best when scoped narrowly
Why convenience often beats cost in consumer adoption
What early adopters risk — and what they may gain
Related Shows Based on Content Similarities
Discover shows related to The Signal, based on actual content similarities. Explore podcasts with similar topics, themes, and formats, backed by real data.
Chris has held a version of this since before he left brokering. His view is that the mortgage was never the thing anybody was buying. The house is. The mortgage is the monthly payment you stop thinking about the week after it funds, which makes the advice wrapped around it, rather than the loan itself, the part that is genuinely exposed. Neither host believes people vanish from a purchase this stressful. Both of them get stuck on the same question, which is where exactly those people are supposed to stand once the friction is gone.
THE GAP BETWEEN AI SPEND AND AI RETURN
The last stretch of the episode is about money, and it is where both hosts turn sceptical. Reuven cites a presentation from one of the world's largest asset managers on the scale of capital expenditure going into AI infrastructure this year, the gigawatts of data centre capacity still to be built, and what a single investor's portfolio companies are spending with Anthropic annually. The figures are striking, and they arrive with a caveat the episode states out loud, which is that the firm presenting them has a great deal of its own capital deployed in exactly what it is describing.
Chris's question is what happens when spending on that scale meets an antitrust regime and a compute shortage at the same time. Reuven's question is simpler and more uncomfortable. He asks what anybody is actually getting back, puts a figure on how much of his own AI output he never uses, and extends it to companies doing the same thing with far larger budgets.
WHAT YOU'LL TAKE AWAY
- The model companies are entering lending from the borrower's end and the capital markets end at once, which puts the pressure on the middle of the stack rather than on any single product.
- What a borrower is buying is the house, so the advice wrapped around the mortgage is the layer most exposed when the friction disappears.
- Price and friction are the two things either host says a consumer genuinely chooses on, and which of the two a lender is worse at.
- Money going into AI is not the same as value coming out, and the honest version of that gap is easier to hear from somebody describing their own usage than from any vendor.
CHAPTERS
(00:00) Every Major AI Lab Has Launched a Finance Tool
(00:59) Why the Model Companies Need an ROI Story
(03:00) Enterprise Partnerships and the App Layer
(07:53) What Your LLM Already Knows About You
(10:16) Buying Through the Chat Window
(14:05) What Borrowers Say They Want Versus What They Pick
(16:59) Amazon, a Tesla, and a Mortgage Renewal
(20:28) Underwriting Agents for Commercial Lending
(22:33) What the Regulators Will Allow
(24:59) When Your Agent Talks to the Lender's Agent
(29:06) AI CapEx and the Compute Shortage
(33:36) Tokens Spent and Value Returned
MENTIONED IN THIS EPISODE
Anthropic - the model company whose finance tooling and commercial lending agent Chris had been reading about the week of the recording.
https://www.anthropic.com
OpenAI - named alongside Anthropic as the other lab pushing into enterprise finance, and the platform Reuven moved his own history across to.
https://openai.com
ChatGPT - OpenAI's assistant, and the window both hosts describe borrowers already living in.
https://chatgpt.com
Claude - Anthropic's assistant, and the one Reuven exported his history out of before switching.
https://claude.ai
Perplexity - named as one of the players that has launched a finance tool this year.
https://www.perplexity.ai
Google - named as announcing a partnership with a large financial institution, and as launching a commercial lending agent of its own.
https://www.google.com
Microsoft - named as doing the same thing on the enterprise side.
https://www.microsoft.com
Better - the lender Chris says now takes a mortgage application without the borrower leaving ChatGPT.
https://better.com
Plaid - the data connector Chris expects a consumer finance assistant to use to pull a full financial picture together.
https://plaid.com
Higgsfield - the video and graphics tool Chris uses, which he found running inside the ChatGPT window rather than on its own site.
https://higgsfield.ai
Reddit - where Reuven had ChatGPT look for a better renewal rate than the one his lender offered him.
https://www.reddit.com
Amazon - Reuven's example of a market where convenience removed the humans from the transaction entirely.
https://www.amazon.com
Tesla - the car Reuven bought for his wife through an app, without meeting anybody.
https://www.tesla.com
SAP - one of the enterprise software companies Reuven says executives are leaving for the model labs, Rolodex in hand.
https://www.sap.com
Oracle - named in the same breath, for the same reason.
https://www.oracle.com
Salesforce - named in the same breath, for the same reason.
https://www.salesforce.com
New England Mortgage Bankers Association - the conference whose keynote speakers argued that borrowers only really want the lowest rate.
Rocket and UWM - named as the large originators most likely to follow Better onto an assistant.
YOUR HOSTS
Chris Grimes is CEO of FundMore.
Reuven Gorsht runs Deeded and The Variable.
LISTEN AND SUBSCRIBE
Captivate: https://the-signal.captivate.fm/
Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259
Chris supplies the uncomfortable part. He argues that low adoption among lenders is not a temporary lag, that it has stayed low, and that a proof of concept which never reaches production is not evidence of anything.
WHAT COUNTS AS A COMPETITIVE ADVANTAGE IN MORTGAGE TECH NOW
Reuven's answer begins by taking something away, in one flat sentence that is the most quotable line on the tape. What he says is left afterwards sits in two places. The first is distribution, meaning a network or a community you can actually put a product in front of. The second is the judgment to know which output is any good, which he ties to twenty-five years of knowing which advertisement is off message and which one will get you in trouble with a regulator.
Scott gives the clearest worked example. He ran a competitive analysis across the Canadian and US markets and turned it into a deck for his team, and he puts a figure on how long that took against the weeks it used to take.
A moat is not a moat if it can be replicated, and that today almost everything can be. You can point a model at a competitor and have it decompose what they are doing in fifteen minutes.
He also raises the problem nobody in the industry enjoys looking at. When you feed your own working into a general purpose tool, you may be sharpening the thing a competitor uses against you next quarter. He puts it as a question, because he does not have the answer either.
THE END OF THE PURE PRODUCT COMPANY
Scott asks the hosts a direct question about their own businesses. If a junior person with some AI ability can now build internally, for a salary, what a vendor charges in licensing, does that change how a lender evaluates products like theirs?
Chris gives ground immediately and says that it will. Then he draws a line, and where he draws it is the argument of this section. He does not think vendors get wiped out. He thinks the pure product company stops existing unless it owns one very specific problem in a very large space, and that what replaces it is a partner who brings most of the way and then works with the customer on the rest. He puts the split in proportions and calls the result a pivot toward services. Whether he is right is the thing to listen for.
Reuven works the same idea through a different industry. He describes replacing an advertising agency brief with a Sunday morning and a model, and is precise about what he did not get, which was thirty years of agency expertise and somebody else carrying the responsibility. His conclusion is that the agencies, the law firms and the accountants who deploy the same tools survive at lower prices, and the ones who do not are in trouble.
WHAT YOU'LL TAKE AWAY
The cost of building a product has collapsed to almost nothing, so the number of companies chasing each niche has grown far faster than the number that ever get funded.
Rolling out AI is not like rolling out a loan origination system. A loan origination system arrives as a set of defined screens and nobody has to decide what it is for, and that difference is the whole reason a thousand licences can change nothing.
A moat built on user experience is not a moat, because a competitor can point a model at it and have it decomposed in an afternoon.
The scarce thing is no longer the intelligence. It is knowing which of its outputs is any good, and that comes from years nobody can prompt for.
CHAPTERS
(00:00) How the three of them met
(07:03) Hundreds of companies chasing the same niches
(08:28) When the cost of development goes to nothing
(12:23) Why lender adoption stays low
(15:55) Why a thousand licences change nothing
(16:25) Redesigning the job instead of bolting AI on
(19:36) The end of the pure product company
(20:54) What Claude did to an ad agency brief
(25:24) A competitive analysis in one sitting
(28:56) The value of intelligence
(31:41) The three reasons anyone buys anything
(44:12) What actually protects BrokerBot
MENTIONED IN THIS EPISODE
BrokerBot Inc. - the Canadian platform Scott runs, built on licensed property data and a broker's existing client book.
https://www.thebrokerbot.ca
Opta - the Canadian property intelligence business where Scott built an AVM, now Verisk Canada.
https://optaintel.ca/
Verisk Analytics - the analytics group that acquired Opta, and part of the data background Scott describes.
https://www.verisk.com
FCT - the Canadian title and property intelligence company Scott names alongside Opta and Verisk.
https://fct.ca
Claude - the model all three describe using, for competitive analysis, advertising strategy and prototyping.
https://claude.ai
Arcads - the video generation tool Reuven used to produce a batch of advertisements.
https://www.arcads.ai/
FundMore - Chris's company, named when he and Reuven compare what it used to cost to start a software business.
https://fundmore.ai
House Sigma - named by Scott in a passage about disruption of the Canadian real estate data market. The claim he attaches is his own.
https://housesigma.com
REALTOR.ca - named by Scott in the same passage, alongside RBC, Pinch Financial and TD Insurance. Same caveat.
https://www.realtor.ca
Mortgage Professionals Canada - runs the national conference where the three of them met.
Deeded - Reuven's company, named in the cost comparison and when he describes rebuilding its advertising.
MoveSnap - the company Reuven left in 2019, named in the origin story.
YOUR HOSTS
Chris Grimes is CEO of FundMore.
Reuven Gorsht runs Deeded and The Variable.
Scott Larter runs BrokerBot, a Canadian platform built around the client book a mortgage broker already has rather than the next application. He came to mortgage from data, at Opta, Verisk and FCT, and argues the next competitive advantage in the industry is not faster submissions but sustained client relationships.
LISTEN AND SUBSCRIBE
Captivate: https://the-signal.captivate.fm/
Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259
The sharpest stretch of the conversation is not about her own build at all. It is about everyone selling to her.
She arrived at the recording from a week of vendor demos. Her question at the end of each one was not about the model, it was about the implementation timeline and the implementation cost, and the answers kept coming back at six months and a price she thought the work no longer justified.
WHAT YOU'LL TAKE AWAY
- The automatable surface in non-QM is eligibility, not decisioning, and the two are worth separating in writing before anyone builds anything
- A rules engine you rent with rules you own is a different asset from a platform you rent entirely
- The honest return on an eligibility engine is underwriter throughput, and hers is 20 to 25%
- Eliminating a task and eliminating a person are two separate decisions, and conflating them is a leadership failure rather than a technology one
- If a vendor cannot tell you why implementation takes six months, that is the question to keep asking
CHAPTERS
(00:00) What AI can't do in non-QM lending
(01:35) An operator handed the innovation job
(03:29) Taking a technology role without a technology background
(06:25) Automating a lender with no rulebook
(09:22) Will judgment ever move to the machine
(11:29) Why she doesn't start with the org chart
(12:24) Eliminating task, not judgment
(14:01) What the manager's job becomes
(16:27) Why vendor AI still takes six months
(21:18) Inside Aura, eligibility not decisioning
(22:48) A 20 to 25% lift for underwriters
(25:07) The right partner at the right cost
MENTIONED IN THIS EPISODE
Logan Finance, the non-QM lender where Sarah Gonzalez is Chief Innovation Officer
Aura, Logan's eligibility engine, built on a partner's rules engine with rules Logan wrote and owns
HousingWire AI Summit, the Dallas event in August 2026 where the hosts met her
The Mortgage Collaborative, whose summit she had attended the week before recording
IDP and OCR, the document technologies she argues get folded into the word AI
Claude, named as an example of what per-seat AI costs a mid-size lender without change management around it
RE/MAX, Redfin and Rocket, named by Reuven as evidence the ground is shifting upstream and downstream
ABOUT THE GUEST
Sarah Gonzalez is Chief Innovation Officer at Logan Finance. She has spent thirty years in mortgage, including chief operating and president roles at earlier companies, and now works on where technology genuinely changes how lending gets done rather than where it only adds a step.
The specific decision worth copying is what he did about prompting. The industry's answer to a salesforce that cannot get useful output from a model is to teach prompting, which means a course, a deck and a Saturday. Lindsey's team went the other way and pre-loaded the prompts a loan officer would actually need, so the interface became a plain-language question. You ask it the way you would ask a colleague, and the output arrives already shaped, because the work sits behind the box rather than in front of it. It helped that the engineer had been a loan officer himself and knew which questions were real.
The word she uses for the target is one she brought from agency work rather than from software. In advertising, the aha moment is the point where a customer insight lands and the consumer goes "oh, I get it" with a small laugh. That is the thing being engineered. Not comprehension, not competence, recognition. Her claim is that a rollout succeeds or fails on whether a person reaches that moment on their own inside the first minute, and that everything upstream of it is design work.
WHY AI ADOPTION FOR LOAN OFFICERS IS A MARKETING JOB
Once the platform stopped being the obstacle, the obstacle was attention. Loan officers are inundated, and Lindsey's assessment of internal email is blunt: people do not read it, including when a marketer wrote it. That is not a communications failure to be fixed with a better subject line. It is a channel that does not work, and the campaign has to move.
So it moved to where the salesforce already was. Her team had spent years building an audience of loan officers who follow the company online, and the platform messaging went out through that channel instead of the inbox. Alongside it ran the mechanics any marketer would recognise: testimonials solicited from producers, a gift card for the trouble, and success stories pushed out repeatedly rather than announced once. A top producer who built something over a weekend was put on stage at the sales rally, and it was among the best-rated sessions they ran, because the person demonstrating it closed loans for a living.
The most-quoted piece of evidence in her HousingWire talk was a loan officer who pulled six refinances in under 24 hours using the platform. She is careful about how that works as a marketing asset: the result is the incentive, and her job was to make sure everyone else heard about it.
NAMING THE AGENTS, AND WHAT THEY ARE NOT ALLOWED TO DO
The fix for the unread email was not another email. Casey got a face, a voice and a commercial, and a line telling loan officers to save her number because when she calls, money is calling. People started talking about her, which is the point at which an internal launch stops being an announcement and becomes something with a shape people can hold.
The compliance work behind it is the part most rollout stories leave out. Casey has been through review and is trained on the fact that she is not a licensed originator: pushed toward anything she cannot answer compliantly, she offers a transfer instead. Lindsey's description of the testing is that they did a lot to try to break her. A loan officer who does not want her can switch her off, and the cadence, the script and the voicemails get reviewed as consumer tolerance shifts.
WHAT YOU'LL TAKE AWAY
- Prompt training is a symptom. If a salesforce needs a course to get output from your platform, the prompts belong in the platform, not in the curriculum
- Internal email is not a channel for a launch. The salesforce already follows the company somewhere else, and that is where the campaign goes
- Naming an agent is a positioning decision, not a gimmick. A named colleague lands differently from a new module on a stack people already resent
- A rollout two years in can still have no adoption number, and the honest version of that is more useful than a percentage
CHAPTERS
(00:00) Open
(01:14) A month into the merged company
(02:22) Two years building the platform
(05:13) Design the friction away
(08:25) Incentives, testimonials and six refis
(10:51) Early adopters and laggards
(14:14) Adoption when the salesforce keeps growing
(18:47) Casey, and the call she got about her
(21:48) What Casey is not allowed to do
(24:21) Friction versus value
(27:51) The borrower who came back six months later
(29:21) What she would do differently
MENTIONED IN THIS EPISODE
American Pacific Mortgage, the lender where Lindsey now leads marketing and business development
Synergy One Lending, where the AI enterprise platform was built with an outside AI partner and which merged into APM in June 2026
Casey, the AI assistant that calls borrowers on a credit-pull or refinance trigger and transfers them to their loan officer
Alex, the AI texting assistant that nurtures borrowers who say they are not ready yet
Steve Majerus, now president of American Pacific Mortgage
HousingWire AI Summit, the Dallas event in August 2026 where Lindsey made the case that adoption is a marketing problem
The Mortgage Collaborative, whose event the hosts went to afterwards and whose June survey of lender AI deployment comes up late in the conversation
Encompass, the loan origination system whose training session Lindsey watched from the back of the room
Claude, which her marketing team uses to build borrower personas out of social posts and Reddit threads
ABOUT THE GUEST
Lindsey Ricciardelli is SVP of Marketing and Business Development at American Pacific Mortgage, where she supports AI adoption across the salesforce following Synergy One Lending's merger into the company. Her argument, made on the HousingWire AI Summit stage in Dallas this August, is that getting people to use a new platform is a design and marketing problem rather than a training problem: you engineer one moment that lands, instead of scheduling another session nobody attends.
Abdel Khawatmi is Area Manager at Paramount Residential Mortgage Group (PRMG), running offices in New Jersey and lending across several states. He was headed for law school before mortgage, holds a master's in risk and compliance, and has built his book on the files other originators pass on — self-employed borrowers, thin credit, layered income and unusual properties. He was named a HousingWire Rising Star in 2026.
