Eglae Recchia, CEO, Keyway

Guest: Eglae Recchia, CEO, Keyway

91% of CRE Firms Know AI. 9% Act on It.

Keyway CEO Eglae Recchia on agentic infrastructure, deal underwriting, and the 9% problem in CRE AI.

 

Commercial real estate (CRE) AI adoption remains critically low despite near-universal awareness: a Keyway survey found that while most commercial real estate firms believe AI will transform their business, only 9% have implemented it in any meaningful way. Keyway - a purpose-built agentic infrastructure platform for commercial real estate - addresses this gap by aggregating 50,000 normalized data points per asset, running proprietary agent orchestration across multiple LLMs, and delivering comp sets, deal narratives, and offering memorandum analysis in seconds rather than hours. The platform serves owners, operators, lenders, and investment sales professionals across the full deal life cycle, from acquisition screening to asset management.

 

Key Takeaways

 

• Only 9% of CRE firms have acted on AI. According to a Keyway survey, the overwhelming majority of commercial real estate companies believe AI will change their business - but awareness has not translated to operational deployment.
• The build-vs-buy debate has a clear answer. A CTO Recchia spoke with recently spent $100,000 in API tokens in a single month by giving his entire team unrestricted access to general-purpose LLMs. Focused, purpose-built platforms consistently outperform internally assembled stacks on both cost and accuracy.
• Horizontal LLMs hallucinate in fiduciary environments. General-purpose models produce variable answers to identical inputs. Keyway's proprietary agent layer - trained on thousands of CRE deals and tens of thousands of OMs - enforces consistency. Hallucination is not an acceptable failure mode in underwriting.
• Agentic infrastructure is not a point solution. Keyway's architecture routes tasks to the best available model for each job rather than locking clients to a single LLM. As underlying models improve, so does the platform - without client-side engineering work.
• The workflow problem is structural, not motivational. CRE professionals understand AI exists. The barrier is identifying which specific workflows to hand off. Recchia's diagnostic approach - asking what frustrated a professional most in their last underwriting session - cuts through abstraction to actionable deployment.
• Comp generation moved from 5 hours to under 60 seconds. Client case studies on the Keyway platform document this compression. The platform scores similarity, benchmarks against live market conditions, and surfaces the delta between an OM's claimed comps and current actuals.
• Data security is the real enterprise gating factor. Keyway is SOC 2 certified, operates with full client data isolation, and does not use client information to train or inform any other client's outputs. For regulated lenders and institutional operators, this architecture is a prerequisite for adoption.

 

Adam Gower has spent more than 30 years in commercial real estate, closing over $1.5 billion in transactions across multiple market cycles. His clients today collectively manage more than $45 billion in AUM. That vantage point - across GFC, rate cycles, and now the AI transition - shapes every conversation on this podcast. The question Recchia addresses here is one Adam hears from sponsors constantly: I know AI matters. I do not know where to start, or whether to build or buy.

 

The 9% Problem: Why Awareness Does Not Become Action

 

Keyway conducted a survey of commercial real estate companies and found a gap that defines the current market moment. The firms surveyed overwhelmingly agreed that AI would materially change their business. Only 9% had done anything about it.

Eglae Recchia, CEO of Keyway, has a precise diagnosis. "I think AI, as an industry, as a whole, we've moved towards solution first, instead of what problem are we trying to solve, and therefore, what's the right solution." The consequence is a profession handed a sophisticated tool and told to find a use for it - without anyone first asking what frustrates them most in a given underwriting session.

The diagnostic Keyway uses is deliberately short: two minutes, one question. What slowed you down the most today - finding comps, building a narrative, running an OM, reviewing a rent roll? The answer defines the deployment. 

This approach reflects a broader tension Recchia identifies across the industry. CRE has historically been technology-resistant - she cites phones and Excel as the only two prior transformations that generated genuine adoption momentum. What is different with AI, she argues, is that people already want to use it. They have used ChatGPT to write emails. The curiosity exists. The gap is connecting that curiosity to a professional workflow with a defined outcome.

 

Agentic Infrastructure: What Keyway Actually Does

 

Keyway describes itself as an ‘agentic infrastructure platform’ purposefully built for commercial real estate. The phrase requires unpacking.

The platform is not a wrapper on a general-purpose LLM. Keyway runs its own proprietary agent orchestration layer that selects the optimal model for each specific task - comp generation, narrative drafting, OM analysis, rent roll anomaly detection - and applies five years of CRE-specific training to constrain and validate outputs. "The horizontal LLMs get better makes Keyway better," Recchia explains, "but where we differentiate is what did we do between then and the output that our users see."

The data layer is equally significant. The platform normalizes 50,000 publicly available data points per asset in real time. This is not a licensed data feed from a single incumbent provider. It is a normalized aggregation across public property records, transaction data, and market information - purpose-built to support the comp and underwriting workflows that CRE professionals run every day. 

Critically, the platform's outputs are consistent. A sponsor who uploads an OM today and again tomorrow receives the same analysis. That predictability is non-negotiable in a fiduciary environment. "There's no space for hallucinations in commercial real estate," Recchia states. "You don't want that to turn into different answers."

The platform's five main modules - comps, docs, property stage, key rate, and deal room - can be adopted independently or as an integrated stack. Recchia describes the architecture as Lego pieces: modular, interoperable, and configurable to a client's existing workflow rather than requiring a wholesale process change.

 

The Build-vs-Buy Calculation in CRE AI

 

The most common objection Keyway encounters has shifted in the past six months. Previously, the dominant pushback was: I can just do this with a general-purpose LLM. That objection has largely resolved itself.

Recchia recounts a recent conversation with an unnamed CTO. The firm gave its entire team open access to general-purpose LLM APIs, encouraged experimentation, and watched token costs hit $100,000 in a single month. The experiment produced neither structured outputs nor consistent analysis. 

"You have 10 people trying to do 100 things, I have my whole team only doing this one thing," Recchia says of the build-vs-buy frame. "If I had to bet on who will get it right - not only because it's Keyway, but I believe we will, right? It's as old as always, that theory."

The current objection, she says, is ROI measurement. CRE is a numbers-driven business and firms want to quantify the return before committing. Keyway's answer is to work directly with clients to model time savings against deal volume. The platform's published case studies include workflows reduced from five hours to under one minute. The challenge is that reaching that number requires firms to do an honest internal audit of how they currently work.

There is also a multi-model dimension that in-house builds consistently underestimate. AI model performance varies by task. A firm that commits its stack to a single LLM is betting on that model's future trajectory as well as its current capabilities. Keyway's orchestration layer eliminates that dependency. "You want somebody whose whole day is to find that out, know how to leverage, and give you the best possible answer every single time."

 

Security Architecture and Enterprise Deployment

 

For lenders and institutional operators, security architecture is not a secondary concern - it is the primary gating factor for any AI deployment. Keyway is SOC 2 certified and operates with full client data isolation. No client's data informs any other client's outputs. No learnings leave the client's environment.

"This is not data that anybody wants to share," Recchia says. "Everybody thinks they have their special sauce. This is information that is necessarily specifically the reason why their business works."

The platform is enterprise-grade by design. Keyway built its security architecture at inception in response to industry experience with the consequences of treating proprietary CRE data carelessly. The practical implication is that the platform passes the procurement processes of major lenders and institutional operators without modification.

Pricing is structured to match the platform's modular architecture. Clients pay a base platform fee plus usage-based costs scaled to deal volume - for example, the number of OMs processed or underwriting memos generated annually. Comps-only access carries a simpler flat-fee structure; full-platform access including OM ingestion carries incremental usage pricing. Enterprise clients at scale run into the hundreds of thousands annually; smaller operators have entry points scaled to their activity. There is no per-token billing model on the client side - usage risk sits with Keyway, not the client.

 

Frequently Asked Questions

What does Keyway do differently from general-purpose AI tools like ChatGPT or Claude?

General-purpose LLMs are trained on broad datasets and produce variable outputs for the same input - a meaningful problem when running the same OM analysis multiple times or benchmarking against a moving market. Keyway runs its own proprietary agent orchestration layer on top of multiple LLMs, selecting the best model for each task, and applies five years of CRE-specific training to constrain outputs to consistent, commercially defensible answers. The platform also aggregates 50,000 normalized public data points per asset in real time. The result is comp sets, deal narratives, and OM analyses that are both faster and more reliable than what a professional can assemble manually or through an off-the-shelf AI interface.

 

Who uses Keyway and at what stage of the deal life cycle?

Keyway's client base divides into two primary segments: owners and operators (including fund sponsors and asset managers) and lenders with exposure to commercial real estate, including those transacting with GSEs. Use cases span acquisition screening - identifying properties meeting buy-box criteria, including off-market assets - through active asset management, where the platform benchmarks properties against comps and surfaces tenant sentiment data. Investment sales officers and mortgage bankers use the platform to build underwriting narratives and stress-test market assumptions. The modular architecture means a sponsor can start with comps, add OM analysis as deal volume grows, and layer in deal room functionality as the team scales.

 

How does Keyway handle data security for enterprise clients?

Keyway is SOC 2 certified and operates a fully isolated data environment for each client. No client data is shared with or used to inform outputs for any other client. The platform draws on 50,000 normalized publicly available data points per asset rather than relying on proprietary feeds that carry their own data-sharing and compliance implications. For regulated enterprises - lenders subject to banking oversight, institutional operators with fiduciary obligations - Keyway goes through the full vendor security review process. Recchia notes that some of her most productive sales conversations are with CISOs, reflecting how central security architecture has become to CRE AI adoption decisions.

 

What does Keyway cost, and how is pricing structured?

Keyway prices on a combination of a base platform fee and usage-based costs tied to deal or document volume - for instance, the number of OMs processed annually or the frequency of underwriting memo generation. Clients who want comps-only access can access a simpler flat-fee tier. Full-platform access, including OM ingestion and deal room functionality, carries incremental usage pricing. Enterprise clients at scale run into the hundreds of thousands annually; the entry point for smaller operators scales with their activity. Recchia's approach is to start the pricing conversation from the use case - how many deals does the firm underwrite per year, how many OMs does it process - and size the cost accordingly rather than presenting a standard rate card.

 

What to Do Next

If you are a CRE sponsor and you still have not identified a specific workflow to hand to AI, the Keyway diagnostic is the right starting point: what slowed you down most in your last underwriting session? Answer that question before evaluating any platform. 

For a broader framework on how commercial real estate sponsors are building AI-powered operating systems - from deal sourcing through capital deployment - the GowerCrowd AI Accelerator Program works directly with sponsors managing $50M to $500M+ AUM. Join the waitlist for our next AI in Real Estate Accelerator executive program.

 

Adam Gower, Ph.D. is the founder of GowerCrowd and one of the most experienced practitioners in commercial real estate capital formation. With more than 30 years and $1.5 billion in transactional experience - including serving as President of a Universal Studios development division overseeing $400 million in projects across Asia Pacific - Adam now helps CRE sponsors build AI-powered systems for investor acquisition, deal management, and capital deployment. His clients collectively manage over $45 billion in assets under management. gowercrowd.com