The Deployment Gap
A third of commercial real estate firms already have AI in the building. Almost four in five still process their documents by hand. This is Kolena’s second annual study of AI adoption in CRE — where last year’s asked what buyers wanted, this one asks what actually changed. Based on 667 conversations with 277 commercial real estate companies.
The participants
This report draws on 667 anonymized buyer conversations with 277 unique commercial real estate companies, held between September 2025 and July 2026. Where a company spoke with us more than once, only its most recent conversation is counted — so each company appears exactly once.
Investment and asset management firms lead, followed by property managers, developers, and CRE lenders — the corners of commercial real estate that carry the heaviest document loads.
The sample spans boutique operators to national enterprises: about a third are under 50 people, while roughly one in seven has 1,000 or more employees.
There is no single AI buyer in CRE. Leadership leads, but the rest spreads across asset and property management, investment, finance, IT, and operations — AI initiatives here are cross-functional, and the people feeling the pain are rarely in one department.
Companies operated across major US markets, concentrated in New York, California, Texas, and Florida. Role, size, and geography were disclosed in roughly half of conversations; those breakdowns describe the companies where the detail was stated.
CRE didn’t start more AI projects. It finished more of them.
The headline shift is not that more companies are trying AI — it’s that more are finishing. Companies actively scaling AI in production jumped from 1.5% to 9.7% — a six-fold increase (p<0.001). Meanwhile the share merely piloting barely moved (13.0% → 13.7%).
Read those two facts together and the story is clear: the growth isn’t coming from a wave of new pilots. It’s coming from companies that were already piloting graduating into production. Overall, companies in active execution rose from 14.9% to 24.5% (p=0.003).
Roughly one in ten CRE companies has AI genuinely running in production. Nearly seven in ten are still evaluating or exploring. The market has not adopted AI — a determined minority has, all the way, while the majority remains at the threshold.
AI is already in the building. It just isn’t doing the work.
Ask a CRE firm whether it uses AI and a third will say yes. 34% of companies are using a general-purpose AI tool — ChatGPT, Claude, or Copilot. But ask who still processes their documents by hand, and the number is 78%. Nearly a third of the market — 29% — has both: a general AI subscription, and a document workflow that is still entirely manual.
The obvious assumption is that general AI is a stepping stone — firms start with a chatbot, get comfortable, and graduate to systematic automation. The data does not support that. Companies using a general LLM are barely further along in production deployment than those using none (29% vs 22%), and they are no more likely to have tested AI on their own documents (13% vs 15%).
Why it stalls, in buyers’ own words
Across the conversations, CRE teams described the same four walls when they tried to push a general chatbot into real document work:
- It never becomes automation. One VP of technology built a lease-audit script with a general AI model in half a day — but it still required manually dropping in every file. The chat window became one more manual step, not a replacement for one.
- It doesn’t know the domain. One asset manager codes financial line items into a proprietary format with more than 600 categories. A general model has no way to know that taxonomy; a purpose-built agent is designed around it.
- It doesn’t scale or govern. A firm running due-diligence questionnaires through individual, unshared chatbot subscriptions described the result as error-prone, unscalable, and requiring constant policing.
- It becomes shadow work. One team runs its files through a purpose-built agent and separately through a general chatbot to catch what it might miss — doing the same job twice.
The requirement that emerges is not “smarter AI.” It is workflow-native AI: something that already understands what a lease abstraction is, what a rent roll reconciliation involves, and what the 600 line items in a chart of accounts mean — that runs on a batch of files rather than one at a time, connects to the systems the work already lives in, and leaves an audit trail. A chat window, by design, does none of those things.
It’s not the technology. It’s the org chart.
If two-thirds of CRE hasn’t reached execution, what’s holding them there? We classified every company’s single primary open question. The result is decisive: the remaining work is almost entirely internal.
39% are working through an internal decision — agreeing on a use case, aligning stakeholders. 26% are still evaluating fit against their own workflows. Competing priorities account for 15%, an unproven business case 11%. Companies that concluded AI simply cannot do what they need? 5.4%.
Classified a second way: of every company with an open question, 87% have only internal work left — a decision, a missing demonstration, a process step. Just 13% face a genuine hard blocker in the technology.
What “fit” and “internal step” actually mean
Inside the two largest buckets is the real texture. When a team questions fit, it is mostly whether AI maps to their multi-step workflow (37%) or simply that they haven’t yet seen it run on their own files (33%). The worry that their documents are too messy — often assumed to be the blocker — is just 6%. And when an internal step stands in the way, the most common by far, at 51%, is agreeing internally on what to use AI for at all — well ahead of budget (29%) or security review (14%).
Note what this means. No one has said no. The organization simply hasn’t yet decided yes.
Where does your organization stand?
Three questions. See how your situation compares with the 277 CRE companies in this study — and what the companies that moved forward did next.
—
—
The proof gap: only 1 in 7 has seen AI run on their own documents
Just 14% of CRE companies have tested AI on their own files. Six in ten saw a relevant demo but never tested it. Another 20% didn’t see enough to judge at all.
This is the single most actionable number in the study. Companies aren’t rejecting AI — they are stalling in a state of not-quite-knowing. A demo on someone else’s documents settles nothing about whether it works on yours, and buyers know it: recall that a third of fit questions are precisely “we haven’t seen it on our own files.” The evaluation isn’t incomplete because teams are lazy. It’s incomplete because nobody ever put their documents in front of the thing.
The use cases are broadening beyond leases
Lease abstraction is still the most-requested workflow, but it cooled sharply — from 44.6% to 32.1% (p=0.002) — as the conversation widened. Rising fastest: rent roll analysis (+7.5 points) and underwriting (+7.2 points), both highly significant.
The pattern is a market maturing past its first obvious use case. Lease abstraction was the entry point for CRE AI; rent rolls, underwriting, and due diligence are where it goes next.
What buyers want: efficiency, not magic
CRE teams are strikingly consistent — and modest — about what they want from AI. Efficiency (92%) and time savings (87%) dominate. Revenue growth barely registers at 14%. This is an industry that sees AI as a way to give staff their hours back, not as a growth engine.
One shift stands out: cost avoidance fell from 45.8% to 32.5% (p<0.001). As companies moved from evaluating to executing, their goals narrowed to the most concrete, countable wins — hours saved, errors caught — rather than broad claims about cost.
The fundamentals are not in doubt: 53% are in high or critical pain, 78% still work manually, and 22% are actively weighing replacing an outsourced provider with AI.
What is this workflow really costing you?
Most CRE teams in this study handle document work in-house — so the real cost is staff time. Pick your most painful workflow and see what it adds up to.
Choose how it’s handled, then enter your numbers.
Meeting commercial real estate where it is
What buyers want AI to plug into barely changed from a year ago: Excel, SharePoint, and Yardi still lead. The stability is the point — integration into the systems the work already lives in remains non-negotiable, even as maturity and use cases shift.
It is worth reading this list next to Finding 02. The reason a chat window stalls is written here: the work lives in Excel, SharePoint, and Yardi, and a general chatbot does not go there.
Enterprises scale. Smaller firms experiment.
Enterprises (1,000+ employees) are more than twice as likely to be scaling AI in production — 18.2% vs 8.8% for smaller firms — even though overall execution rates are closer (33% vs 22%). The divergence isn’t about who starts. It’s about who finishes.
This tracks the report’s central thesis. Scaling AI is an organizational act: it needs a decided use case, an internal owner, and a process to push it through. Enterprises have the machinery for exactly that kind of decision — and, it turns out, the patience.
The companies that crossed the gap started with one workflow.
Not a platform decision. Not a committee. They picked the workflow that hurt most, saw it run on their own documents, and decided from there. Only the first step asks anything of you.
How we did this
- Source. Anonymized buyer conversations, analyzed by purpose-built Kolena research agents. Current period: 667 CRE conversations (Sep 2025–Jul 2026). Baseline: 478 CRE conversations (Jan–Sep 2025), from the dataset behind our 2025 report.
- Unit of analysis: the company. Conversations are deduplicated to unique companies — where a firm spoke with us multiple times, only its most recent conversation counts. Final samples: 277 companies (2026) and 323 companies (2025).
- Scope. Commercial real estate only: investment and asset management, property management, development, CRE lending, and commercial brokerage. Residential brokerage is excluded.
- Significance. All year-over-year changes tested with two-proportion z-tests. Findings that did not reach significance are not reported as changes.
- Anonymization. No companies or individuals are named. Outsourcing providers and competing tools are referenced by category only. Integration platforms are named.
About Kolena: Kolena builds purpose-built AI agents that automate document-heavy workflows — lease abstraction, underwriting, rent roll reconciliation, loan and claims processing — for commercial real estate, insurance, and finance teams. An internal research-grade Kolena agent was used to code and analyze the conversations behind this report.