AI Agents for CRE Prospecting and Lease Expirations

By Jude Lee · · Workflow

Two commercial real estate brokers reviewing a lease expiration schedule and property map in a modern office

The part of prospecting that is actually mechanical

Ask a producer what they do to build a target list and you’ll get a description of research, not selling: pulling a lease expiration schedule out of old abstracts, checking the county recorder for a recent deed transfer, scanning a public tenant’s filings for store-closure language, cross-referencing whether anyone at the firm has touched the owner in the last 18 months, then writing thirty emails that each need one specific, true detail to not get deleted.

That assembly work is mechanical. The judgment on top of it is not. That distinction is the whole design brief for a prospecting agent.

In most brokerages, the model is not the bottleneck. The bottleneck is that lease abstracts live in scanned PDFs, contact history lives in one broker’s inbox, and nobody has written down what a good prospecting email actually contains. Fix those three things and a mediocre model performs well; skip them and the best model available produces confident noise.

The signals a prospecting agent can actually reach

An agent is only as good as the data you legally and technically let it touch. Four buckets, in rough order of reliability:

Your own abstracted leases and rent rolls. The richest and most defensible source, because nobody else has it. If your lease data still lives as scanned PDFs, that’s the prerequisite project — see rent roll and lease abstraction automation. Expiration dates, option windows, and escalations are the raw fuel for expiration-driven outreach.

Public records. County recorder and assessor systems publish deed transfers, mortgages, and assessed values; availability and format vary widely by county, and some charge for bulk access. Ownership-change and loan-maturity signals both live here. For publicly traded owners and tenants, SEC filings are free, structured, and searchable via EDGAR — useful for portfolio-level occupancy and store-closure language.

Licensed data platforms. CoStar, Crexi, and Reonomy each publish their own terms; automated extraction or agent access to licensed data is a contractual question, not a technical one. Read your agreement before pointing any agent at it, and ask the vendor directly about API access.

Your CRM and email history. Who at the firm has touched this owner, when, and what happened. This is what stops the agent from sending a cold intro to someone your colleague toured with last quarter.

What the agent run actually looks like

  1. Assemble the window

    The agent queries your abstracted lease data for expirations and option-notice deadlines inside a defined window — say 12 to 30 months out, since that’s when tenant-rep and renewal conversations typically start. It returns a raw candidate list with property, tenant, square footage, expiration, and source document.

  2. Enrich each candidate

    For each row, the agent checks public records for a recent ownership transfer, checks EDGAR if the tenant or owner is public, and pulls whatever licensed data you’re permitted to feed it. Every enrichment field carries a source link back to the record it came from.

  3. De-duplicate against firm history

    The agent queries the CRM: has anyone here contacted this entity in the last N months? Is there an open deal, a signed agreement, or a do-not-contact flag? Suppressed rows go to a separate list with the reason, so nothing silently disappears.

  4. Draft the outreach

    Using a defined skill (below), the agent drafts an email per target that names one specific, verifiable fact — the expiration quarter, the recent transfer, a comparable lease your firm handled nearby — and proposes a single next step. Drafts land in a review queue, not the outbox.

  5. Log and schedule follow-through

    Once a broker approves and sends, the agent writes the activity back to the CRM, sets the follow-up task, and adds the target to the next cycle’s suppression list. This write-back step is what keeps the pipeline honest.

Connecting it to your stack without a rebuild

The connective tissue here is MCP — the Model Context Protocol, an open standard for giving an AI assistant governed access to specific data and tools. Rather than exporting CSVs and pasting into a chat window, you expose named tools like search_lease_expirations, get_contact_history, create_activity, with scoped permissions and an audit trail of every call.

The practical version for most brokerages: a small custom MCP server over your own lease and CRM data, plus whatever official connectors your existing vendors publish. We walk through the build in connecting Claude to your CRE systems via MCP. Two design rules worth repeating: read tools and write tools should be separate and separately permissioned, and every write should be attributable to a human approver.

An agent that can read your whole pipeline and write to none of it is a research assistant. An agent that can write without an approver is a liability. The useful configuration sits deliberately in between.

The outreach skill: making “in your voice” reproducible

A skill is a packaged, reusable instruction set that teaches an assistant to do one job the same way every time — the format, the constraints, the examples, the things it must never do. For prospecting outreach, a workable skill specifies: maximum length; exactly one verifiable specific fact and where it must come from; no market claims without a cited source; no invented comps; a single clear ask; and firm signature and disclosure requirements.

The payoff is consistency across a team of producers who each write differently, and a single place to fix a problem when the drafts start drifting. Write it once, version it, review it quarterly.

Where this breaks, and what stays human

Four honest failure modes:

Stale or wrong ownership. County records lag, entities are layered, and the LLC on the deed is not the decision-maker. The agent should surface the record and its date; a human confirms who to actually call.

Confident fabrication. Ask an agent for a comp it doesn’t have and it may produce a plausible one. Require source links on every factual claim in a draft, and treat any unsourced specific as a defect.

Timing and relationship judgment. Whether to approach an owner now, through whom, and with what positioning is the job. Automating the assembly buys time for that judgment; it does not substitute for it.

Compliance. Commercial email in the U.S. is governed by the CAN-SPAM Act, enforced by the FTC, which sets requirements around accurate header and subject information, a valid physical address, and honoring opt-outs. Calls and texts fall under the TCPA and FCC rules, plus the National Do Not Call Registry, and state laws add more. If you prospect outside the U.S., your own regime governs instead — CASL in Canada, the ePrivacy Directive and GDPR across the EU, and comparable rules elsewhere — and several are stricter about consent than CAN-SPAM. An agent that sends at volume amplifies whatever your compliance posture already is. Confirm your outreach program with qualified counsel before you scale it, and verify current requirements with the governing regulator directly rather than trusting a summary.

Build, buy, or neither

Off-the-shelf CRE tools
CRM sequencing in Buildout, Apto, or ClientLook already handles cadence, templates, and activity logging. Data platforms increasingly ship their own AI search. If your prospecting bottleneck is “we don’t send follow-ups consistently,” buy the sequencing and stop — no agent required. Faster, supported, and cheaper.
Custom agent over your own data
A custom agent earns its keep when the signal you prospect on lives in data only you have — your abstracted leases, your tour history, your submarket knowledge — and when assembling that signal across systems is what eats the hours. It’s a real build with real maintenance. Read the honest version in custom vs off-the-shelf CRE software.

A third answer people skip: no AI at all. If you have 40 target accounts and three producers, a shared spreadsheet and a calendar reminder outperform any agent, at zero cost and zero governance overhead. Agents pay off on volume and on cross-system assembly, not on small, well-known lists.

Modeling the payoff without inventing numbers

Don’t take a vendor’s savings claim. Model your own with a formula you can defend:

Hours recovered = (research minutes per target × targets per month) ÷ 60 × share the agent genuinely removes. Then value them at your loaded hourly cost, and separately ask the question that actually matters: what do those hours get reallocated to, and what is a marginal tour or LOI worth in your shop?

On that last term — expect the removable share to land somewhere between most and nearly all of the assembly step, and nowhere near all of the surrounding judgment work. Confirm the real figure in your own pilot rather than assuming it.

Your number
Research minutes per target, timed on 10 real targets
Illustrative input — measure it yourself
Your number
Targets researched per month, per producer
Illustrative input — pull from your CRM
Measure it
Share of the assembly step the agent actually removes
Assumption to test in a pilot, not a benchmark

The honest caveat: recovered hours only become revenue if someone actually redirects them to calls and tours. Time saved that leaks into inbox time is worth nothing.

Start narrow. One asset class, one submarket, one 90-day expiration window, drafts to a review queue for a month. Track draft approval rate, but track reply rate and meetings booked alongside it — a pilot where brokers approve nearly every draft and nobody answers them is the most common way this work looks successful and isn’t. If approval rates stay low, the fix is almost always the data or the skill definition, not a bigger model. And on the plumbing: connector standards in this space are still churning as of early 2026, so assume you’ll rewrite integrations and keep your business logic separate from them.

Related: rent roll and lease abstraction automation, CRE CRM comparison, and the broker-ops payback model.

Not sure where to start?

Get a free automation audit: we map your deal pipeline, marketing, and back-office workflows and show you what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides