What does a governance-first AI rollout look like in a brokerage?
Policy, training, and measurement come before access, and the guardrails live inside the tools rather than only in a memo. That is how I took enterprise AI from policy to production for every agent and employee across 30 offices.
- Period
- 2025 onward
- Scale
- 1,100 advisors and staff across 30 offices
- Focus
- AI Governance, Enablement Platform, Fair Housing Compliance
Problem
Agents were already using AI. The only open question was whether they were using it inside a policy we wrote or outside one we never published. Three constraints stack up in residential brokerage: Fair Housing law and MLS rules apply to every piece of public marketing, advisors handle confidential client financials and negotiation strategy every day, and the sales force consists of independent contractors working across company systems and personal devices. In that environment, an unmanaged AI rollout creates immediate legal and brand exposure, and a policy vacuum reads as permission.
Approach
We wrote a clear AI Governance Policy first and kept the rules concrete enough to train across every office:
- Use approved enterprise tools for anything touching company or client data.
- Never paste confidential client financials, contracts, or personal data into unapproved public models.
- Apply Fair Housing and MLS accuracy standards to every AI-assisted listing description and marketing asset.
- Keep a human publisher of record responsible for verifying every output before it reaches a client or the public.
Then we built the guardrails into the tools. I negotiated the multi-year Google Workspace and Gemini agreement directly with Google, and we deployed Gemini inside our existing admin and data-protection controls, under enterprise terms that keep company data out of model training. Gemini went where 1,100 agents already worked, in Gmail, Docs, and Drive. Claude and Claude Code went to the corporate team, where deep reasoning, contract and vendor analysis, and code mattered.
To move beyond generic chat, I designed and built the enablement platform myself with Claude Code, on Vercel and Supabase behind company single sign-on. It combined role-based learning paths, a curated real estate prompt library, a cited policy assistant grounded in company documentation, and guided listing and property image studios.
For AI-enhanced property imagery, we built compliance into the rendering pipeline itself: every generated image automatically received a 'Digitally Enhanced' watermark in the brand typeface before download, making MLS and consumer disclosure automatic rather than optional. We trained with real failure modes rather than abstract theory, and I shared the governance and enablement framework with peers at the Sotheby's International Realty Brain Trust in April 2026.
I built the measurement myself: monthly active use by office, cohort, and usage tier, with automated and passive events stripped out, so training went to the offices and the try-once-and-stop users who needed it.
Result
Gemini for Workspace and the internal AI enablement platform went into production for every agent and employee with policy, training, and telemetry in place before launch, while Claude powered internal workflows across finance, operations, IT, and support.
By pairing clear rules with purpose-built tools that made the compliant path the easiest path, AI shifted from shadow experimentation to a governed daily operating capability across the brokerage. The goal was never adoption for its own sake. It was changed behavior.
Every regulated, relationship-driven business, from wealth management to insurance to healthcare, has the same gap: model capability is ahead of what the enterprise actually uses. Governance before access, role-specific workflows inside existing tools, and cohort-level measurement turn AI from a compliance liability into operating leverage.