Commercial real estate lender Greystone is using AI to push the bounds of creativity and recreate workflows in addition to enhancing operational efficiency.  

New York-based Greystone issued more than $2 billion in multifamily loans in 2025, ranking ninth in the United States, according to Fannie Mae. It is also a prominent lender in the healthcare, affordable housing and senior housing sectors, with $75 billion in total originations over the past five years, according to the company.  

Greystone is working to quantify the efficiency gains resulting from its AI projects, but the company is most excited about what those gains can enable, Zac Rosenberg, executive vice president of IT strategy and engineering, told FinAi News.  

“If you can compress a 60-day underwriting process down to 60 minutes, that’s not incremental efficiency,” he said. “That’s unlocking entirely new ways of doing business — for instance, daily client portfolio underwriting, proactive client engagement, the ability to handle deal types and complexity we previously couldn’t.” 

Greystone expects quantifiable and significant returns from AI deployment in 2026 due to the maturity of the AI ecosystem, investment in its engineering team and collaboration between its IT and business teams, Rosenberg said. 

Rosenberg recently sat down with FinAi News to discuss Greystone’s AI strategies and broader AI trends shaping the real estate lending industry. What follows is an edited version of that conversation: 

FinAi News: How are AI advancements affecting real estate lending strategies, and what are the implications for the real estate industry overall?  

Zac Rosenberg: I think many lenders initially approached AI as a cost-savings tool — a way to automate existing processes and improve efficiency. That’s understandable, but it’s actually too narrow a lens. What we’re seeing with this wave of AI is something much broader. It’s unlocking fundamentally new ways of working.  

Of course we want efficiency gains. But if you simply take your existing workflows and ask, “How do I automate this,” you’re missing the real opportunity. AI isn’t just good at automating tasks. It’s creative. It can think strategically alongside you. It can help you identify new products, new services, new ways to serve clients.  

At Greystone, this means shifting from a reactive posture to a proactive one. Historically, clients came to us with a problem or opportunity, and we responded. Now, armed with portfolio data and market intelligence, we can reach out to clients and say, “We noticed your maintenance costs are unusually high. Let’s dig into why and explore whether there’s an opportunity here — maybe a supplemental, maybe operational improvements that unlock new value.”  

That’s the difference AI makes. It’s not just about doing things faster. It’s about seeing things you couldn’t see before, and having conversations you couldn’t have before.  

FinAi: What strategies is Greystone deploying to capitalize on new technologies? 

ZR: Our strategy really comes down to three things: 

1. We’ve been intentionally building an engineering team youwouldn’t typically find in a commercial real estate lending company. We’re not trying to out-spreadsheet our competitors or out-cold-call them. We’re going to out-engineer them. That requires investing in really talented, product-minded people who can build sophisticated systems and tools.  
2. We’ve broken down the traditional “IT-as-support” model. Our engineers sit right next to our underwriters, servicers and loan originators — not remotely, but physically close. Great ideas are more likely to be sparked at the lunch table than the conference table, low-pressure places where an engineer can turn to an underwriter and say, “Walk me through this process. How does this actually work?” That proximity drives better solutions.
3. We’re being deliberate about modularity in our architecture. The AI landscape is changing incredibly fast. With new models, capabilities and best practices emerging constantly, we need systems that aren’t locked into one tool or approach. We’re exploring multiagent orchestration using sophisticated frameworks, which lets us compose specialized agents that can reason about specific domains. This modularity means we can swap underlying approaches or update our methodology without rebuilding everything from scratch. It allows us to pivot quickly as the landscape evolves.  

The real leverage isn’t any single tool. It’s the combination of talent, proximity and flexible system design.  

FinAi: What areas of operations are best suited for AI integration and why?  

ZR: Many people use AI and automation interchangeably — but I think that is fundamentally incorrect. The key distinction is automation is great for deterministic processes, where you know exactly the steps, the decision points, the if-then logic. But AI shines in areas where you can’t automate because the work requires dynamic thinking, creativity and reasoning that changes based on context.  

Underwriting is a perfect example. Historically, groups have tried to automate it by building complicated nested logic: if this, then this, then that. Real estate is too nuanced for that. Properties have unique characteristics. Markets shift. You encounter situations you’ve never quite seen before. The more edge cases you try to code in, the more fragile the system becomes.  

What AI agents do differently is handle this dynamically. You’re examining a property. The first layer looks good. You go deeper and discover something unexpected — environmental concerns from adjacent parcels, for example. Instead of having a predetermined path for that scenario, an AI agent can ask: What do we need to understand here? Should we get an environmental assessment? Have we seen similar situations? What does the data tell us? It can reason through the situation in real time.  

AI creates the most value in areas where reasoning is complex, edge cases are numerous, and you need adaptive thinking. It’s powerful in underwriting. AI is valuable in portfolio analysis, where you’re surfacing insights across hundreds of properties and bringing to light patterns you couldn’t see before. And it’s also useful in risk management, where you’re thinking through second- and third-order effects of different scenarios.  

FinAi: What risks do real estate lenders face when implementing new AI tools, and what are the keys to managing these risks?  

ZR: AI is powerful, and the more resources and access you give it, the more capable it becomes. But capability without guardrails is dangerous. AI agents are like brilliant but reckless interns: They have great ideas and can execute them, but sometimes they get tunnel vision and miss the practical context.  

The solution isn’t to lock things down. It’s to build the right infrastructure.  

First, sandboxes. We create development environments with realistic but completely synthetic data. Synthetic properties, synthetic financials, synthetic personas that look real enough to build against but pose zero risk to actual client information. Engineers can experiment, crash things, try new approaches without ever touching production systems or real data.  

Second, testing and governance. This is where a lot of teams see guardrails as obstacles to innovation, but that’s backward thinking. Rigorous testing, data governance and access controls actually free you to innovate.  

Until you have proper tests in place and sandbox environments, you’re always scared of what you’re building. You have to be cautious because you could really break something. On the other hand, invest in the right infrastructure, and you can experiment boldly because you’re protected.  

The real risk isn’t the technology — it’s rushing to production without the fundamentals.  

FinAi: How can real estate lenders customize AI solutions to best serve specific sectors?  

ZR: The worst approach is to throw your entire knowledge base at a model and say, “Solve this.” That’s how you get unpredictable results. Instead, you want agents working on highly specific tasks with access to highly relevant information and tools.  

You can use the same underlying model for different purposes and change the context you give it. That’s where multiagent orchestration comes in. Instead of one general-purpose underwriting agent trying to handle everything from A to Z, you have specialized sub-agents, each focused on a specific transaction type or problem, with access to the specific information and tools they need.  

Take affordable housing. An agent built for these deals would have access to materials on Low-Income Housing Tax Credits, HUD Housing Assistance Payments contracts and sponsor-initiated affordability programs. It would understand the constraints these programs put on the property — what you can and can’t do with units, how rents are determined, what expenses are allowed. It would know how to model the different subsidy structures and how they affect timeline and cash flow.  

That’s fundamentally different from a standard multifamily agent, which would focus on market rents, stabilized (net operating income) and traditional financing structures.  

The same goes for other sectors. The architecture is consistent — modular, composable agents with access to relevant context, context that is customized. An agent analyzing student housing looks at different metrics. An agent working on industrial properties thinks about different risk factors. The framework scales; the inputs change.  

You don’t need radically different infrastructure for each property type. That would become a nightmare to maintain. 

FinAi: How do you expect AI to reshape real estate financing over the next three to five years?  

ZR: I expect the industry will eventually realize that AI is a much stronger thought partner and strategic adviser. It’s not just a tool to make existing work faster; it’s a collaborator that can challenge your thinking and help you see things differently.  

Some people will hand over the keys and let AI drive. It might work, it might not. The better approach is active partnership. When I’m working through something with AI building an application, working through a problem, I make sure we have checkpoints where it explains what we’re doing and why. I want it to challenge me, disagree with me, point out weaknesses in my reasoning. We debate. Then we figure out solutions together.  

I use voice mode on my drive to work for exactly this reason. I’ll walk through a presentation or an idea, and I’m having the AI stress-test my thinking in real time.  

That same principle applies to how organizations implement AI. You can race ahead and impose changes on your teams, allowing AI to take the wheel, but that’s not typically how real progress is made. Progress is made when the team is actively engaged and working toward a shared goal. AI does not replace team engagement or the importance of having clear, established goals. To the contrary, the teams and institutions that will win are the ones that use AI to help them communicate openly, iterate rapidly, engage holistically and resist the temptation to just let it run on autopilot.  

The key is to ride the wave together, not get consumed by it.  

Register here for the inaugural FinAi Banking Summit, taking place March 2-3 in Denver. View the full event agenda here.