Company of the Week:
A property was built in 2017. Straightforward enough.
… Or was it 2018?
It might sound innocuous, but that discrepancy actually provoked one of the most interesting moments in our recent demo with Keyway.
The platform had pulled information from multiple diligence documents for the same deal, and some of them disagreed on the year built. Rather than sneakily picking one and carrying on anyway, Keyway shone a spotlight on the conflict, showed exactly which documents each answer came from, and asked the analyst to make the final call.
Anyone who’s spent time inside a deal room knows how quickly these discrepancies multiply when you’re working across appraisals, environmental reports, OMs, sponsor materials, rent rolls, financials, and everything else that gets dumped into diligence. This also neatly encapsulates what Keyway is trying to do more broadly: take the repetitive parts of investment management, automate as much as possible, but still keep people involved where judgement is still required.
Watch The Keyway Demo Here:
AI Maturity in Real Estate Survey
Keyway, is running a new survey - AI Maturity in Real Estate - to dig into what separates firms actually scaling AI from those still stuck running isolated pilots.
The survey is open to professionals across acquisitions, lending, asset management, finance, legal, property management, and executive leadership.
It's confidential, takes 5-10 minutes, and every respondent gets the final report, including which business functions are seeing the most AI adoption, which organizational models (centralized AI teams vs. function-by-function ownership) are working, and what's separating firms that scale from firms that stall.
This builds on Keyway's State of AI Adoption in Real Estate report from earlier this year, which surveyed 150+ U.S. real estate professionals and found the industry is long on experimentation, short on follow-through:
45% of firms are running AI pilots, though only 9% have reached enterprise-wide deployment.
Just 8% consider their data infrastructure fully ready to support AI at scale.
Trust, data readiness, and system integration are the top three barriers to scaling.
More than half of firms expect to significantly increase AI investment over the next 2 years.
We’re Going Live in 3 Hours!
Today (9/16) at 10am PT, we’re unpacking the biggest stories where technology changes the physical economy. From AI infrastructure and data-center power to the tools changing how commercial real estate firms operate, we’ll debate what matters.
An unusual origin story
Keyway didn’t start life as a pure software company.
The business began around 2019 as an owner and operator of multifamily real estate, while simultaneously building its own tech to help with acquisitions, market analysis, and workflow automation. They did a bit of everything, basically.
“We’d done the discovery on ourselves,” Joe Moskowitz - Keyway’s Head of Growth and part of its founding team - told us. “We knew what our pain points were. We knew what the gaps in the data we had were and we filled all those gaps.” By late 2023 the company had decided the larger opportunity lay in the technology itself. It divested the real estate portfolio, and has operated as a pure tech business since around mid-2024.
You can actually still see that operating history in the product.
One side of Keyway remains heavily focused on multifamily market intelligence and comps. The other is now much broader; covering document analysis, diligence, underwriting workflows, deal management, and enterprise AI.
Where things got interesting
Keyway can handle some pretty familiar CRE AI jobs.
Upload a rent roll, and it can standardize the data. Upload a T12, and it can map the financials into a client’s own chart of accounts. Ask questions about a loan agreement or environmental report, and it can return an answer with references back to the source document. Where the demo became more intriguing was when those individual tasks started feeding into the same deal workflow.
Joe showed us Keyway’s Deal Room: the workspace it uses to organize the documents and extracted data for an individual transaction. It lined up overlapping data points side by side, making it easy to see where the documents agreed, and - more importantly - where they didn’t.
That’s where our 2017 vs 2018 problem appeared.
Keyway did not automatically decide which figure belonged in the final deal record. The user reviewed the source files instead, and selected the value they wanted. As Joe put it, he was making “the final judgment call.” That decision then flows through the rest of the work. If the analyst changes a data point that affects another calculation, or a written section of an IC memo, Keyway can flag the dependency and update the relevant narrative.
Basically - and mercifully - you don’t have to fix the same piece of information manually in five different places, and hope you remembered all of them.
For an investment team producing materials that may eventually be challenged by an IC, lender, or senior management, having the source attached is obviously useful too. If somebody asks where a number came from, the analyst can go straight back to both the document and specific section that produced it.
Breaking down the workflow
That approach carries through the rest of the platform. “We’ve taken investment management workflow and broken it down into its pieces,” Joe explained.
Rent roll abstraction is one piece, while T12 normalization is another. Document comparison, underwriting, lease audits, IC memos, loan narratives, and other tasks can then be combined, depending on how a particular firm actually works.
Neat.
This is especially useful because CRE teams rarely follow exactly the same process. One client may need its financial data mapped into a particular ERP schema. Another may want a human approval checkpoint halfway through the workflow. Yet another might need a very specific investment committee format before anything can move forward.
Joe said the conversations Keyway is having with CRE firms have changed noticeably over the past year. Previously, many prospects were still asking broad questions about what AI could do, and what tools were available. Now, they’re showing up with much more specific requests.
They know what output they need, which system the data has to enter, where a person needs to approve the work, and which parts can happen automatically afterward.
That is a pretty good indication of where enterprise AI adoption is heading in CRE. It also explains why Keyway has broken the investment process into individual pieces that can be combined around each client.
As firms get more specific about exactly how they want AI to work, Keyway can take those requirements and turn them into a repeatable process which fits the way the team already operates.
The bigger problem
Most firms now have at least a few people who are genuinely good at using ChatGPT, Claude, or another model. The problem is that the optimal workflow often lives with that one person.
One analyst might have built an excellent first-pass acquisition prompt. Another has figured out a reliable way to review leases. Someone else has their own CAM reconciliation workflow.
That’s all great… but the people sitting next to them are doing something completely different. It’s not exactly the most efficient way to work.
Joe described the risk as an “individual level imbalance and uneven fiefdom,” where some employees become highly proficient with one model, and other people use entirely different approaches.
“There’s no uniformity,” he said - but Keyway’s new Skills Library is designed around changing that.
The platform can package a workflow into a reusable Skill that can then be kept personal, shared with selected colleagues, or deployed across an entire organization. Current examples include CAM reconciliation, acquisition first pass review, lease audits, and underwriting model creation.
That means if one smart cookie analyst develops a particularly effective workflow, it doesn’t have to remain that analyst’s secret sauce (to mix our food metaphors). The process can instead be formalized and reused by everybody else.
You don’t have to write the Skill yourself (thank the lord)
An obvious potential problem is that a serious workflow can require an intimidating set of instructions.
Joe showed us the underlying content behind one of Keyway’s CAM reconciliation Skills. It wasn’t exactly War and Peace, but it was still long enough that expecting every property manager or analyst to write it themselves would be unrealistic.
That’s why Keyway built a Skill Studio on top.
Instead of writing all of the instructions manually, a user explains what they’re trying to achieve, the output they need, and the format they want. Keyway asks follow up questions, and builds the reusable Skill behind the scenes.
Those Skills can then work against the firm’s own information. Through Key Pilot, Keyway can connect into systems such as SharePoint and OneDrive, as well as broader data infrastructure, allowing users to work across years of internal documents rather than starting from scratch with every new deal.
Joe said Keyway can now put together workflows that feel highly tailored to an individual client in “a week, a couple of days, not even.” That speed is one of the bigger changes AI is starting to bring to enterprise software.
Our take
Plenty of what Keyway does will already sound familiar. Rent rolls, T12s, document chat, lease abstraction, deal screening, underwriting, IC memos - you know the score.
The genuinely interesting idea here is what happens when those individual AI tasks become part of one repeatable process.
CRE firms have spent the last few years letting employees experiment with general-purpose AI tools, and some of those experiments have certainly produced useful workflows. The next challenge is sharing the wealth, and making sure the good ones don’t stay trapped on one analyst’s laptop.
Keyway is trying to give firms a way to take the workflow that’s actually effective, add the data and controls the organization requires, then let the whole team use it consistently.
For CRE firms that have already spent a couple of years experimenting with AI, that feels like a pretty logical next step - stop treating good AI use as an individual skill, and start turning it into something the whole organization can actually rely on.