When AI Meets Real Estate Finance: A Discussion with Faropoint, Brightshore Capital, & More

At IMN CFO/COO East 2026, a session on AI and automation in real estate finance cut through the hype to focus on something more useful: what's actually proven, what's still risky, and what finance teams have learned the hard way. Here's the rundown:

What's Actually Working

Use cases that have graduated from pilot to production tend to succeed in admin-heavy processes where the logic is well understood but the volume of manual effort is high. The catch is that getting these into production takes more documentation and process rigor than teams initially expect. The key is to build workflows robust enough that a human isn't quietly patching gaps behind the scenes.

One notable shift: purpose-built, custom solutions are starting to outpace generic tools (because the value of a lease abstraction tool isn't just extracting data from a lease, it's extracting it into the specific context a given finance team actually operates in).

That context problem shows up everywhere—even with a clear source of truth, the answer to a given question is rarely sitting in one place. The real work is teaching the models an organization's unique context layer and definitions, not just pointing them at a database.

The Architecture Question That Matters Most Right Now

Deterministic versus generative. The strongest AI solutions for finance blend the two rather than leaning entirely on generative outputs. Probabilistic, "best guess" outputs are only trustworthy when a human owns the logic, the calculations, and the code behind them. This becomes especially important when people are tempted to chain multiple AI tools together into a pipeline—each model might be 90% confident, but stack three or four of them and you lose fidelity at every step, with no real audit trail left by the end. For finance, where the numbers need to tie out, that's a real problem, not a theoretical one.

On trust and control, the framing that resonated: AI is predictive and analytical, not infallible. It's a tool that works faster than a human, but it can make the same kinds of mistakes a human would. That means humans still need to own the output. This isn't a reason to avoid automation, but it is a reason to build in the right checkpoints rather than treating AI as a black box that finance teams simply defer to.

Lessons From Stalled/Failed Pilots

Technology is almost never the reason a pilot fails. It's everything around it—translating board-level requirements, legacy debt in existing systems and processes, and plain human inertia on implementation. Teams that treat AI adoption as a technology project rather than a change-management project tend to be the ones that stall out.

1. Measuring ROI is still an unsolved problem for a lot of teams, but the session offered a useful lens: ROI shows up most clearly when you start from a clear use case and a clear cost you're trying to address, rather than working backward from the technology. It's on CFOs and COOs to actually reallocate the freed-up resources and redeploy that value—otherwise the efficiency gain never shows up anywhere real. The benchmarking challenge is that while you can measure time before and after at a macro level, the micro-efficiencies gained by individual people are much harder to quantify, even though they stack up meaningfully over time.
2. For auditors, the process becomes far more straightforward when finance teams can provide well-organized underlying data rather than reconstructing it ad hoc—effectively building a catalog of information that supports the numbers. This is where Predictap's approach fits into the conversation. That's a good proxy for how AI should be deployed responsibly in this space more broadly: help customers simplify their processes without introducing new risk, so they can make more appropriate decisions with the same or less effort.
3. The session pointed to a real gap in the market: the need for a standard AI data addendum that lets finance teams opt in based on their own risk tolerance with vendors—before it even gets to legal—rather than relying on each provider's default terms.

Looking A Year Out

The expectation is that AI becomes a formal, embedded part of finance processes—though notably not the review process itself. Instead, AI will increasingly flag what needs human review, with a human in the loop remaining non-negotiable. Cross-document analysis should also mature, pulling together information that today lives in scattered systems. Underneath all of this sits a persistent theme: how information is shared, and the security and data concerns that come with it, will shape how fast any of this actually gets adopted.