When we started Nectar in 2021, our vision was ambitious, but our resources were limited. We wanted to provide flexible capital to real estate operators across the country, but traditional underwriting processes would have made this impossible at our scale. Today, we’ve originated over 150 loans across the USA with a lean team of 10 full-time employees. The secret? We didn’t just adopt AI; we made it the backbone of our entire operation.

Here’s exactly how we did it, and the lessons we learned along the way.

Building a Culture of AI Discovery

The first thing we did was make AI exploration part of our DNA. For the past year, every team member hasn’t just been encouraged to adopt AI tools—they’ve been required to. More importantly, we instituted weekly showcases where team members presented new AI findings, tools, or applications they’d discovered.

This isn’t just about staying current with technology. It’s about creating a mindset where everyone is constantly looking for ways to work smarter. When your loan officer discovers a new document processing technique or your operations manager finds a better way to automate client communication, that knowledge immediately spreads across the entire team.

The compound effect has been incredible. What started as individual productivity gains quickly became systematic improvements that scaled across our entire platform.

Our Technical Foundation Was Already AI-Ready

Here’s where we got lucky—or maybe prescient. Since inception, we’ve built our underwriting technology stack internally. When AI capabilities started accelerating, we didn’t have to rip and replace legacy systems or navigate vendor relationships. We had the ultimate flexibility to integrate new AI capabilities directly into our existing workflows.

This gave us a massive advantage. While competitors were trying to figure out how to retrofit AI into their processes, we were already experimenting with direct integrations. When a new model dropped that could improve our document processing accuracy by 15%, we could have it implemented and tested within days, not months.

The lesson here isn’t that everyone needs to build their own tech stack. It’s that having control over your core processes—whether through internal development or the right partnerships—is crucial when you want to move fast with emerging technologies.

One Size Definitely Doesn’t Fit All

This might be the most important lesson we learned: different AI models excel at various tasks, and trying to force one model to do everything will limit your results.

Take PDF processing, for example. It sounds simple, but the difference in accuracy across models is stark. In underwriting, a mistake on a financial document can cause massive legal or financial issues. We have zero tolerance for errors in this area, so we spent considerable time testing different models until we found the ones that consistently delivered the accuracy we needed.

For document assembly, we leverage models with larger context windows—like Gemini’s latest releases—to create comprehensive loan packages. Claude’s models are unmatched at delivering natural-feeling text for marketing content. We’ve even started experimenting with Google’s Veo 2 for video content, and we’re excited to test Veo 3.

The key is being willing to seamlessly integrate the best models into your workflow. Maybe in the future, one model will be able to do everything perfectly. But right now, using the right tool for each specific job is what separates good results from extraordinary ones.

Train Your AI Like You’d Train an Employee

Here’s something most people get wrong: you cannot plug and play AI models and expect optimal results. Pre-training is mandatory for almost every application.

We approach AI training exactly like we’d approach training a new employee. We’re very specific about what our business does, what the AI’s functions will be, and what we want the results to look like. The better we can teach the model about our business, give it clear processes to follow, and set proper expectations, the better our results will be.

For example, our underwriting AI doesn’t just analyze cash flow statements—it understands our specific lending criteria, knows what red flags to watch for in each particular underwriting document, and can flag potential issues that would matter specifically to our investment thesis.

The more autonomous we can make our AI systems through proper training, the more they behave like highly trained employees who can work independently while maintaining our standards.

Staying Ahead of Rapid Development
The AI space is developing at breakneck speed. Models are gaining new capabilities monthly, and what’s 87% accurate this quarter might jump to 94% next month. Staying on top of these developments isn’t optional—it’s a competitive necessity.

Honestly, it’s a lot of work. But by baking this discovery process into our culture, we’ve distributed the load across our entire team. Everyone is constantly evaluating new tools and capabilities, which means we catch improvements and opportunities faster than we would if it were just one person’s job.

This vigilance has paid off repeatedly. When a new model is developed that can improve our processing speed or accuracy, we're often among the first to implement it, giving us weeks or months of competitive advantage, especially since most of our competitors aren't even using pre-AI technology.

Looking to the Future

What excites me most about the future isn't any single AI breakthrough—it's the compounding effect of continuous innovation. Every month, new models emerge with capabilities we couldn't have imagined just weeks before. Each advancement doesn't just add to our toolkit; it multiplies what's possible. We're building an ecosystem of specialized AI agents, each trained on best-in-class models for their specific domain. As these models evolve, so does our entire business—not incrementally, but exponentially. This isn't about keeping pace with technology; it's about harnessing each wave of innovation to redefine what a lean, efficient financial services company can achieve.

Our goal remains the same: fund real estate operators faster than the competition while thoroughly underwriting enough to deliver the consistent double-digit returns our investors expect with the lowest possible risk.

The 20x capacity increase wasn’t just about processing more loans—it was about maintaining our quality standards while scaling. AI didn’t replace human judgment in our underwriting; it amplified our team’s ability to make better decisions faster.

For any business considering serious AI implementation, my advice is simple: start with culture, invest in flexibility, embrace specialization, train deliberately, and stay current. The technology is powerful, but how you implement it will determine whether you 20x your business or just add expensive complexity.

The storm coming to the economy will separate the companies that can adapt quickly from those that can’t. AI isn’t just helping us weather that storm—it’s helping us thrive in it.

Derrick Barker is the CEO and Co-Founder of Nectar, a fintech platform providing flexible capital to commercial real estate operators. He began investing in real estate while studying at Harvard University and continued to build his portfolio while trading bonds at Goldman Sachs. Over nearly a decade, he grew his real estate holdings to $450 million in value before launching Nectar in 2021 to solve the capital access challenges he experienced firsthand as a real estate entrepreneur.