Shadow AI in Commercial Real Estate: Why Bans Fail and What Works

Shadow AI, meaning employees using AI tools without IT or leadership approval, is a growing governance concern. Microsoft and LinkedIn’s 2024 Work Trend Index found that 78% of AI users bring their own AI tools to work, not that 78% of all employees use unapproved tools. In commercial real estate, where every document touches underwriting assumptions, tenant-landlord terms, transaction history or borrowing data, unapproved use is not just a productivity shortcut. It risks uncontrolled disclosure. We broke this down in a recent conversation with Bisnow: “How Companies Can Eliminate Shadow AI And Gain Value From Artificial Intelligence Tools.”

What is shadow AI?

Shadow AI is any use of an AI tool that a company hasn’t sanctioned. An analyst pastes a rent roll into a public chatbot to reformat it. An associate uploads an LOI to a free summarizer. A broker connects an assistant to a third-party API no one vetted.

None of this is malicious. It’s people trying to get work done faster with tools that are genuinely good at the job. The problem is that the data leaves the building and no one can say where it went.

Why is the risk higher in CRE than in other industries?

Most industries have a mix of sensitive and non-sensitive data. CRE does not. A lease is a financial instrument. A rent roll is a valuation input. A tenant’s sales reporting is confidential by contract. A borrowing covenant is material non-public information.

There is no low-stakes document to practice on. That changes the math on a single careless upload — the cost implication isn’t a policy violation, it’s potentially millions of dollars and a broken counterparty relationship.

The second risk is quieter and arguably worse: AI tools are fluent and confident, so a plausible-looking output gets trusted. When someone who isn’t technical receives a clean, structured abstract in nine seconds, the instinct is to use it, not to verify the co-tenancy clause it silently dropped.

Why banning AI tools doesn’t work

The instinct is to issue a policy and block the domains. It doesn’t change behavior. It just pushes the same activity onto personal devices, where you have even less visibility than before.

What actually works is structured enablement: showing people what reliable, safe AI looks like for their specific role.

That’s also where most CRE firms are stuck. The tools arrived before the governance did, so the response was ad hoc piloting — roll something out, hope for ROI. What surfaces instead is that the tool doesn’t solve the actual problem, the team was never trained on it, or the underlying data was never clean enough for it to work. This is a strategic sequencing problem, and it’s the reason so many AI pilots quietly stall.

Consider a listing proposal. Handing a broker a chat assistant and an empty text box doesn’t produce a reliable polished proposal; instead it produces a different kind every time, depending on who wrote the prompt. The work only becomes consistent when the firm’s own standards are captured as a repeatable skill the firm owns, and the proposals run through a purpose-built automation instead of improvised prompting.

What to measure

Skip adoption metrics. Logins tell you nothing. Track outcomes:

• How many employees are successfully using AI to automate rote work today?

• How much more productive and engaged are they?

• Of the outputs being generated, how much is actually reliable?

The sequence: enablement, then agents

The strategic sequencing problem has a strategic sequencing answer. Two layers, in order.

Layer one is enablement. Build AI fluency where it pays off first — training on repetitive, templatizable tasks, with the firm’s own language captured as skills the firm owns. Every firm should start here, whether they work with Bryckel or on their own. This solves the behavior problem: people stop reaching for unapproved tools because the approved one now does the job better.

Layer two is infrastructure. Complex, high-stakes workflows can’t be one-shotted from a chat box. A full lease abstraction, a multi-property market survey, an LOI review against the landlord’s playbook — these need multiturn agents, enterprise-grade tooling and token optimization to be dependable. Bryckel deploys those agents inside your existing environment, working through the AI assistants your team already uses.

Coaching solves the behavior problem. Agents solve the infrastructure problem. Run them in that order and the data never leaves, the output stops varying by operator, and you have an AI capability instead of an AI pilot.

See more from Bryckel on Bisnow →