Transforming Mortgage Lending: How Marcus Arkan's Lorien AI is Revolutionizing the Industry
Marcus Arkan, the founder and CEO of Lorien, is a visionary entrepreneur at the intersection of finance and technology. With nearly two decades of experience in mortgage lending and a passion for coding that dates back to his teenage years, Arkan has positioned himself as a trailblazer in the fintech industry. His journey from loan origination to creating AI-powered solutions for the lending workflow showcases a unique blend of industry knowledge and technical prowess.
In an exclusive interview with AI World Today, Arkan shares insights into his entrepreneurial journey, the challenges faced in the mortgage industry, and how Lorien AI is transforming lending processes. From his early days as an OpenAI beta tester to developing specialized AI agents that handle complex mortgage tasks, Arkan’s story is one of innovation, perseverance, and a deep understanding of the financial sector’s needs.
1. Walk me through your journey to founding an AI company.
I’ve been coding since I was a teenager, but my professional career started in finance. Over the years, I’ve built custom applications for my own companies. Tools that became the backbone of our operations. This perspective of understanding both the financial world and having the technical ability to build solutions is really what led me to create Lorien AI.
2. How does your experience in both coding and finance contribute to your unique perspective in the fintech industry?
My background is specifically in mortgage lending, and that’s where the dual perspective really matters. In mortgage operations, I experienced firsthand how complex these processes need to be. Every checkbox, every verification step exists for a reason. But I also saw how this necessary complexity was crushing productivity across the entire industry.
My coding background changes how I view these same mortgage operations. I see which tasks drain the most time, how to automate them, and which success metrics really matter. ROI is everything. Every dollar on tech needs to cut more than a dollar from operational costs. Now, obviously in mortgage lending, we can’t ignore compliance. It’s critical. But compliance without ROI is just overhead. In today’s lending environment, margins are too thin for anything that doesn’t add to the bottom line.
When you’ve built systems while being accountable for business results, you develop solutions that deliver on both fronts. You’re not just moving fast and breaking things. You’re automating within a framework that protects everyone involved.
3. What specific pain points in the mortgage and lending industries did you identify that motivated you to create Lorien AI?
At CIBC’s President’s Choice Financial division, I saw high-volume mortgage lending firsthand. Everything moved through slow manual workflows. The real pain point was how it all compounded. Processors spending large portions of their day on document management. Loan officers waiting hours for fee updates. IT teams constantly patching automations. Onboarding new staff meant weeks of training on company systems and policies.
So when I launched my own brokerage, I built RPA systems to address these inefficiencies. Even with basic automation, we scaled to over 600 loan officers across Canada. But that success revealed deeper problems. Every time a form changed or a new regulation came in, we’d have to rebuild parts of the system. The technology wasn’t there yet. Then in 2020, I became an early beta tester for OpenAI and suddenly, everything clicked. This was the missing piece.
4. Could you explain how Lorien AI’s specialized AI agents work and how they differ from traditional automation solutions?
We’ve built AI agents that handle mortgage processing tasks that normally require experienced professionals. Our agents work within predefined workflows, just like experienced employees follow procedures. But with hundreds of document types, varying lender requirements, and different borrower scenarios, they need flexibility. So they’re equipped with various tools and can make decisions based on context.
When reviewing a loan file, they don’t just pull data. They connect information from different documents, check everything matches up, and make sense of the full picture.
Here’s a cool example: A foreign investor submitted Costa Rican bank statements that looked like an amortization schedule in Spanish. Nothing like US formats, but Lorien handled it perfectly. It found the account numbers, dates, balances, and transactions, then updated everything directly in our loan system.
That adaptability is very different from traditional RPA and NLP solutions that break when forms change or layouts shift.
5. How did your experience as an OpenAI beta tester in 2020 shape your vision for Lorien AI?
Back then, I was experimenting with models most people haven’t heard of. The best ones were Davinci, Curie, and Ada. They’re nothing compared to the models today, but the fact that AI could understand what I prompted and continue my train of thought until the max tokens reached was very impressive.
I’d been using Google’s Dialogflow since 2017 (basically a fancy decision tree), so this was a big step forward. Once I understood how transformer architecture worked, I realized something: these models would eventually be able to handle entire workflows. Not just understand text, but execute complex multi-step processes from start to finish.
That’s when I saw what Lorien could become. AI that takes on the whole mortgage workflow. Documents, validation, conditions, everything. Not replacing processors, but giving them superpowers. Finally, we would have the ability to do what RPA couldn’t.
6. Can you elaborate on the challenges you faced when implementing earlier technologies like RPA systems and NLP models in the mortgage industry?
When we implemented RPA in our mortgage operations, we hit predictable roadblocks. Every regulatory change, every form update meant rebuilding parts of the system. We’d trade loan processors for technical staff just to keep the systems running. It let us handle more volume, but the savings were not everything we hoped for.
What became clear was that you need a mixed approach. RPA, traditional NLP, and modern AI each solve different problems. The answer wasn’t picking one solution. It was combining them.
7. You claim that your solution is 70% faster and 3x more productive. Can you provide specific examples of how this is achieved?
Our biggest improvements target the work that happens after applications come in. Consider what processors actually do all day: collecting documents, analyzing each one, updating systems as borrower details change, plus constant follow-ups with title companies, insurers, and HOAs via email, text, and phone. Every step requires verification to avoid costly mistakes for lenders and borrowers alike.
Our technology changes this process. While human processors handle one task at a time on a single file, our AI agents work across multiple tasks and files simultaneously. What typically takes hours of manual work gets completed in minutes, ready for review. This parallel processing versus sequential manual work is how we achieve 70% reductions in processing time and triple the productivity in our pilot tests.
8. How does Lorien AI ensure that professionals remain in control while AI handles mundane tasks?
We built Lorien to keep humans in control. Every task the AI performs requires human approval, denial, or revision. Our interface streamlines this review process so professionals can validate hours of AI work in minutes. The goal is to automate repetitive work so professionals can spend even more time on what matters: complex problem-solving and client relationships.
As specific tasks reach near-perfect accuracy, we’ll gradually remove human checkpoints. Document classification hits 99.9% accuracy? That can run autonomously. Complex underwriting decisions? Those stay with humans longer.
9. How do you envision Lorien AI democratizing access to advanced AI capabilities for smaller lenders?
Traditional mortgage software gatekeeps through volume minimums and upwards of five-figure setup fees. If smaller lenders wanted to build internally, hiring software engineers with ML experience is prohibitively expensive. And without technical knowledge, you’re relying on trust and luck: trust that they’re writing quality code, and luck that they understand your business needs and can execute effectively.
While we have volume-based tiers, they’re designed to be accessible from day one. No five or six-figure setup fees. Our specialized models deliver enterprise-level capabilities while remaining computationally efficient to keep costs reasonable across all tiers.
Small lenders shouldn’t need a tech team to use AI. We’ve worked hard to bring over 100 integrations so the technology adapts to their existing systems. The goal is for teams to be able to start using it immediately - not after months of implementation. When smaller lenders can compete on service instead of just processing speed, everyone wins.
10. What are the primary challenges you face in implementing AI solutions in an industry as regulated as mortgage and lending?
We’re dealing with three main challenges right now, and I see a fourth coming.
First is the regulatory landscape. You’re not just dealing with federal regulations like TRID and RESPA, but also state-specific rules that can add regulatory overlays or contradict each other. We’ve had to build extensive knowledge graphs to understand how regulations interact and apply in different scenarios. What’s compliant in Texas might not be in Colorado.
Second is data. There’s two parts here: AI needs accurate information to function, and we need to protect sensitive borrower data. Human validation gives us the first, anonymization techniques and on-site data centers handle the second.
Third is explainability. When AI calculates income for a self-employed borrower, it’s processing K-1s, 1099s, bank statements, P&Ls. Dozens of documents with complex rules. Regulators need to see exactly how we reached that number. We log every decision: why 23 months of statements instead of 24, which deposits counted as revenue, which got excluded. Every citation to Fannie, Freddie, or FHA guidelines. Total transparency.
The fourth challenge hasn’t hit yet but I can see it coming: AI regulations. When AI becomes standard in mortgage operations, there’ll be rules about everything from model testing to performance monitoring.
11. What’s coming next in AI that has you particularly interested?
Speech-to-speech models represent the next major shift. Current voice AI sounds synthetic because it maintains constant pacing and tone. The new generation adjusts speaking patterns dynamically. Proper emphasis, natural pauses, appropriate emotional context.
Tool integration is what’s holding everything back. Voice models can’t pull data directly yet. They convert to text first, process the request, then convert back. This creates noticeable lag or latency. Once that’s solved, we’ll have AI that conducts complex phone conversations without any awkward delays.
For mortgage processing, this is transformative. Processors spend a lot of time on phone calls. Following up on documents, verifying employment, checking insurance quotes. These aren’t high-judgment tasks, but they require natural conversation. Voice AI removes the ceiling on how many of these calls you can make.
We’re actively developing voice capabilities for our platform. I can’t reveal specifics yet, but this technology will really change loan processing. The daily life for processors and loan officers is about to look very different.
12. Finally, what advice would you give to other entrepreneurs looking to innovate in the fintech space using AI?
Start by identifying a real problem in the industry. If you’re technical but new to an industry, partner with someone who knows it inside out. If you’re from the industry but not technical, find a technical co-founder who gets excited about solving real problems, not just playing with AI.
The critical step is breaking down existing processes into smaller, manageable pieces. Figuring out what’s slowing everything down and what really requires human intervention. If you don’t have years of experience to already know this, then you need curiosity and access to people doing the work.
At this point you need to identify where AI actually improves things. Do your homework and find out if anyone successfully solved this problem already. I mean really solved it, using it daily in production. The gap between AI that promises to work in presentations, versus AI that handles real financial data and systems reliably separates real solutions from vaporware.
Build proof-of-concepts and get them into production quickly. Real-world testing teaches you what actually works, not what looks good in theory. Then comes the harder challenge - taking what works in your environment and making it enterprise-ready. Making something work reliably for everyone, at scale, is completely different from making it work for yourself.
Marcus Arkan’s vision for Lorien AI represents a significant leap forward in the mortgage lending industry. By combining his extensive experience in finance with cutting-edge AI technology, Arkan is not only addressing long-standing inefficiencies but also democratizing access to advanced AI capabilities for lenders of all sizes. As the fintech landscape continues to evolve, Lorien AI stands at the forefront, promising to reshape the future of mortgage processing and lending operations. With ongoing developments in voice AI and a commitment to regulatory compliance, Arkan and his team are poised to drive innovation that could redefine the mortgage industry for years to come.