Most of the AI content written for commercial real estate answers the wrong question. It asks “which AI CRM should you buy?” when the real question for a 4–20 person firm is “which of my communication workflows should AI touch, and what is the cheapest reliable way to touch each one?” This playbook works through the full communications pipeline (inbox triage, CRM data entry from email, follow-up sequences, and listing copy with syndication) and makes an explicit call for each: use the general-purpose AI you already have, or pay for a proptech point tool. It also covers the two things vendor roundups skip entirely: how to keep a broker’s personal voice when AI drafts the words, and what deal information should never go into a prompt.
Communications Is One Pipeline, Not Four Tools
A small CRE firm’s communications problem looks like four separate problems. The inbox is overflowing. The CRM is three weeks stale. Follow-up happens when someone remembers. Listing marketing eats a full day per property.
They are one problem with four symptoms. Every workflow runs on the same raw material: the email thread. The tenant inquiry, the LOI redline, the comp request from a principal, the signed lease: each arrives as email, and everything downstream (the CRM record, the follow-up task, the marketing update) is a manual transcription of what the email already said.
That is why the biggest AI wins in CRE communications come from automating the handoffs between workflows, not from bolting a chatbot onto any single one. Our view, after building communications automation for professional-services firms handling similar confidentiality constraints: the firms that get this right treat the inbox as a data source, not a to-do list. The ones that get it wrong buy four AI subscriptions that never talk to each other.
The market has noticed the opportunity. Deloitte’s Commercial Real Estate Outlook has tracked AI and data modernization among CRE firms’ top technology priorities for several years running, and nearly every proptech vendor now ships some form of AI feature. The rest of this playbook sorts out which of those features matter for a small firm, workflow by workflow.
Inbox Triage and Email Drafting
Start here, because it requires zero new software and the payoff arrives the first week.
A broker’s inbox mixes five kinds of email: new inquiries, active-deal correspondence, tenant or vendor operations, listing-service noise, and true junk. The triage skill (deciding what each message is and what it needs) is exactly the kind of classification work current AI models do well. Microsoft Copilot inside Outlook will summarize long threads and draft replies in place; ChatGPT, Claude, or Gemini will do the same for anything you paste in, with more control over instructions.
Three triage moves that work in practice:
- Thread summarization before calls. Paste a 40-message negotiation thread and ask for open items, agreed points, and the other side’s last position. Two minutes replaces twenty of scroll-and-squint.
- Draft-with-decision, not draft-from-scratch. Tell the model your decision (“we’ll accept the TI allowance but hold at $24.50 NNN”) and let it write the courteous version. Never let it decide terms; that is your job, and the model does not know your market.
- Inquiry classification. A standing prompt that sorts forwarded inquiries into tour request / financials request / broker fishing / not-qualified, with a suggested next action for each.
The honest limitation: general models do not sit inside your inbox watching mail arrive. Copilot does, which is its one structural advantage. If your firm already pays for Microsoft 365, test Copilot on triage before buying anything CRE-specific. What none of these tools replace is judgment about which relationships deserve a personal call instead of any written reply at all. AI makes writing cheap; it makes knowing when not to write more valuable.
Prompting skill is the constraint here, not software. This is the gap a hands-on LLM fluency workshop closes fastest; the training approach is covered in the CRE AI training playbook.
CRM Data Entry from Email
Every small brokerage pays what we call the data-entry tax. A broker finishes a call or closes an email exchange, and the CRM update (new contact, changed deal stage, the note about the 1031 timeline) either takes ten minutes or never happens. Mostly it never happens. A CRM that is three weeks stale is not a system of record; it is a monument to good intentions.
This is the workflow where purpose-built tools have a real edge, because the job is continuous background capture, not on-demand drafting. Intapp DealCloud markets automatic relationship capture from email and meeting activity. Buildout, which acquired the CRE-specific CRM Apto in 2021, positions its platform around eliminating manual entry between prospecting, CRM, and marketing. HubSpot logs email correspondence against contact records natively, and its AI CRM tooling extends that with drafting and scoring features.
For a 4–20 person firm there are three viable configurations:
| Configuration | What it looks like | Best for |
|---|---|---|
| Manual + LLM assist | Paste threads into ChatGPT/Claude, get structured contact and deal fields back, paste into your CRM | Firms under ~8 people with light deal volume |
| CRE point tool | Buildout, Apto-lineage, or ClientLook-style CRM with built-in email capture | Brokerage-first firms living in listings and comps |
| General CRM + automation layer | HubSpot or similar, with email-parsing automation built to your pipeline stages | Firms with mixed business lines or investment operations |
The middle column costs real money: published pricing roundups such as The AI Consulting Network’s CRE CRM comparison put CRE-specific platforms at roughly $69–$99 per user per month. Whether that beats the manual-plus-LLM approach depends entirely on volume. A two-broker shop updating 15 records a week does not need continuous capture. A ten-person team touching 300 contacts a month absolutely does, and the subscription is cheap against the alternative of a stale database.
One warning we give every firm: AI-assisted capture propagates whatever is already in the CRM. If your database has six duplicate entries for the same property owner under three spellings, automation will happily attach new intelligence to all six. Clean before you connect. The same discipline applies to the lease and deal documents feeding your records; that pipeline is the subject of the CRE document intelligence playbook.
Follow-Up Sequences That Do Not Burn Relationships
CRE runs on long-cycle relationships. The owner who is not selling this year sells in year three, and the broker who stayed usefully in touch gets the listing. Everyone knows this; almost nobody executes it, because systematic follow-up across 500 relationships is exactly the kind of unrewarding discipline that loses to whatever is urgent today.
Automation solves the discipline problem and creates a new one: nothing damages a 15-year relationship faster than an obviously robotic touch. The fix is to be honest about which layer you are automating.
Automate the scheduling, personalize the content. The right cadence engine decides when each relationship gets attention: quarterly for warm owners, monthly for active requirements, annually for dormant contacts. That layer should be fully automated, whether in a CRM’s sequence builder or a spreadsheet with reminder dates. The message itself should be drafted by AI but anchored to something real: a comp that traded near their asset, a vacancy shift in their submarket, a zoning item from the planning commission agenda.
A drafting pattern that works: keep a short market-notes file per submarket, updated when something trades or a tenant moves. When a follow-up comes due, give the model the contact’s history plus the relevant note and ask for three sentences in your voice. The output reads like a broker who pays attention, because a broker did pay attention — AI compressed the writing from fifteen minutes to two.
What not to do. Do not enroll CRE relationships in residential-style drip campaigns; sophisticated owners recognize a mail-merge instantly. Do not automate the send itself for high-value contacts; a human should press the button so a human catches the “actually, he just listed with Marcus & Millichap” problem the system cannot see. And watch the pipeline math rather than open rates: replies and meetings are the metric, not vanity engagement. How screening and market-analysis work feeds these touchpoints is covered in the CRE deal analysis playbook.
Listing Copy and Syndication
Marketing a listing means writing the same property up five different ways: the LoopNet/Crexi listing, the OM narrative, the email blast, the social post, and the one-page flyer. Each has different length, tone, and emphasis. This multiplication is pure AI territory — it is reformatting, and reformatting is what language models are best at.
The vendors know it. Buildout’s AI listing assistant generates property and location descriptions from uploaded documents inside its marketing platform, and its broader suite auto-produces OMs, brochures, and property sites from a single data entry. And because prospects filter and search on LoopNet and Crexi, complete, well-written listings get found and inquired on more often than sparse ones, which makes copy quality a distribution question, not merely an aesthetic one.
The workflow that works for a small firm, with or without a point tool:
- Build a property fact sheet first. Square footage, zoning, NOI, ceiling heights, power, parking, tenancy: verified numbers in one structured document.
- Generate every format from the fact sheet. One prompt produces the 150-word Crexi description, the 400-word OM narrative, and the 60-word email teaser, each constrained to facts in the sheet.
- Verify numbers against the sheet before anything publishes. Every figure in the copy gets checked against the source, every time.
Step three is not optional, and this is the failure mode vendor marketing never mentions: language models fill gaps confidently. Ask for compelling copy about a 42,000-square-foot industrial building and the model may award it 24-foot clear heights it does not have. A wrong dimension in an OM is not a typo; it is a misrepresentation with legal exposure attached. The fact-sheet discipline reduces hallucination risk to a checking problem, which is manageable. Freeform generation makes it a discovery problem, which is not.
On syndication itself (pushing listings to LoopNet, Crexi, and your site), the point tools earn their fee at even modest volume. Re-keying listing data into three portals is exactly the swivel-chair work software should own. A firm with two active listings can do this by hand; a firm with twenty cannot afford to.
Keeping the Broker’s Personal Voice
The most common objection we hear from principals is also the best one: “My clients hired me, and they can tell when an email wasn’t written by me.” Correct. Voice is a commercial asset in a relationship business, and generic AI output erodes it one plausible-sounding paragraph at a time.
The objection has a practical answer, because voice is more mechanical than it feels. It lives in observable habits: greeting choice, sentence length, directness, sign-off, whether you hedge (“might be worth a look”) or commit (“worth a look”). Models are good at imitating patterns they can see. The method:
- Build a voice guide from real sent mail. Pull 10–15 emails you sent and are happy with. Ask the model to describe the patterns: length, formality, phrases you repeat, how you deliver bad news. Edit its description until it is right. That one-page document is reusable in every prompt.
- Few-shot every draft. Prompts that include two or three of your real emails as examples produce drafts that sound like you on the first pass. Prompts that only say “professional but friendly” produce LinkedIn filler.
- Keep an edit pass, always. The 30-second read-and-adjust before sending is where your judgment lives. Over time you will edit less, but the pass never goes away for client-facing mail.
There is a sharper way to say this: AI should raise your floor, not replace your ceiling. The 40 routine emails a week become consistently good instead of rushed. The five emails that matter — the pitch, the apology, the hard negotiation note — you still write yourself, maybe with the model as editor rather than author. Firms that flip this ratio, automating the important mail to save time on the trivial, get the uncanny-valley outcome their clients quietly notice.
Voice discipline is also a team standard. If three brokers at your firm all use the default ChatGPT register, your firm suddenly has one bland voice instead of three distinct ones. Per-broker voice guides prevent that, and building them is a natural exercise in an LLM fluency workshop.
Choosing Between a General LLM and a Point Tool
Vendor content never poses this question because every vendor sells one side of it. Here is the honest scorecard for a 4–20 person firm:
| Workflow | General LLM (ChatGPT/Claude/Gemini/Copilot) | CRE point tool | Our call |
|---|---|---|---|
| Inbox triage & drafting | Excellent with prompting skill | Copilot’s inbox position helps | Start with what you have |
| CRM capture from email | Manual paste-and-parse only | Built for continuous capture | Point tool once volume justifies it |
| Follow-up sequences | Drafts well; no scheduling | Cadence engines built in | CRM sequences + LLM-drafted content |
| Listing copy | Excellent from a fact sheet | Buildout-style tools integrate with syndication | LLM for copy; point tool if syndicating at volume |
| Syndication mechanics | Cannot do it | Core function | Point tool, no contest |
Two patterns fall out of this table. First, drafting work (anywhere the deliverable is words) belongs to general models plus skill, and paying per-seat proptech prices for a wrapped version of the same model is usually poor economics. Second, plumbing work (capture, scheduling, syndication) belongs to purpose-built software, and no amount of prompting talent substitutes for it.
Most firms should therefore run a two-layer stack: one general AI subscription per broker (roughly $20–$30 per user per month) for all drafting, plus targeted point tools only where the plumbing argument holds. The full framework for that decision, including when off-the-shelf proptech is simply enough, is the CRE AI buy-vs-build playbook. And when no off-the-shelf tool fits your workflow, custom automation projects in this space generally run $25–150K depending on scope, a number that only makes sense once the free and cheap layers are exhausted.
Confidential Deal Data Rules
CRE communications are full of information under NDA or informal confidence: rent rolls, tenant financials, off-market pricing guidance, a seller’s divorce. Before anyone at the firm pastes a thread into an AI tool, the firm needs three rules in writing.
Rule 1: know your tool’s data terms. Consumer AI tiers may use conversations for model training; business tiers (ChatGPT Team/Enterprise, Claude for Work, Copilot with commercial data protection) contractually exclude it. The rule is simple: client and deal data only goes into accounts on business terms. The $25-per-seat difference is not where a firm handling confidential deal flow should economize.
Rule 2: strip what the task doesn’t need. Summarizing a negotiation thread does not require the tenant’s name; “national credit tenant, 22,000 SF requirement” carries the analytical content. Redaction-by-default costs seconds and removes most of the exposure.
Rule 3: vendor AI features inherit vendor scrutiny. When your CRM or marketing platform adds AI features, your data may flow to a model provider you did not choose. Ask any proptech vendor the same two questions: which third-party models process our data, and is our data used for training? A vendor that cannot answer crisply has answered.
None of this is exotic; it is the same diligence firms already apply to E&O coverage and listing agreements, extended to a new category. Write it on one page, cover it in training, revisit it annually.
The First 90 Days of Automation
Sequencing matters more than tool choice. The pattern we recommend, and the one that builds adoption instead of resistance:
Days 1–30: drafting only. Every broker uses a general AI tool for thread summaries, reply drafts, and one listing rewrite. No new systems, no integration, no risk. The goal is fluency and a felt win; the training playbook covers how to make this stick in a small firm.
Days 31–60: one plumbing fix. Pick the single worst handoff (usually CRM capture) and fix it with the lightest tool that works. Clean the database first. Measure staleness before and after; that number is your internal proof.
Days 61–90: sequences and listing pipeline. Turn on follow-up cadences for one relationship segment, with LLM-drafted, human-sent messages. Standardize the fact-sheet listing workflow. By day 90 the firm has touched all four workflows with maybe one new subscription.
Resist the temptation to run this backwards: buying the platform first and scheduling the training “once things calm down.” Software without fluency produces the classic small-firm outcome: a per-seat invoice for features nobody uses. Fluency without software still produces faster email this week.
The communications pipeline also feeds the firm’s operational reporting (rent rolls, CAM reconciliation, investor updates), which has its own automation logic, covered in the CRE back-office automation playbook. And the strategic argument for why a 10-person firm can out-execute institutional shops on exactly this kind of adoption is the small CRE firm AI manifesto.
Frequently Asked Questions
Can AI actually draft emails that sound like me?
Yes, if you show it your writing instead of describing it. Prompts that include two or three real sent emails as examples, plus a one-page voice guide, produce drafts most brokers accept with light edits. Prompts that only say “write professionally” produce generic output. The voice-guide method in this playbook takes about an hour to set up and is reusable across every email after that.
Should a small CRE firm buy an AI CRM or use ChatGPT with the CRM we have?
Split the question by workflow. Drafting (emails, listing copy, summaries) is best served by a general model plus prompting skill, at $20–30 per user per month. Continuous plumbing (email capture into the CRM, sequence scheduling, listing syndication) needs purpose-built software once volume is real. Most 4–20 person firms land on one general AI subscription per broker plus a single point tool, chosen after a month of manual-with-LLM operation reveals where the actual bottleneck is.
Is it safe to put deal information into ChatGPT or Claude?
Only on business-tier accounts, and only with discipline. Consumer tiers may use conversations for training; ChatGPT Team/Enterprise, Claude for Work, and Microsoft Copilot’s commercial data protection contractually exclude it. Beyond the account type, strip names and identifying details the task does not need, and never paste material under NDA into any tool without checking the NDA’s terms. One page of written rules covers nearly every situation a small firm hits in practice.
What does AI CRM software cost for a small CRE team?
Published pricing puts CRE-specific CRMs at roughly $69–99 per user per month (Buildout CRM and Apto-lineage products sit in this band), general CRMs like HubSpot from free tiers up to $50–100 per seat for professional plans, and general AI assistants at $20–30 per user. A 10-person firm running the two-layer stack from this playbook typically ends up between $500 and $1,500 per month all-in, which is why sequencing (proving value with the cheap layer first) matters.
Can AI keep our CRM updated from email automatically?
Purpose-built tools can capture contacts, correspondence, and activity from email continuously; that is the core pitch of platforms like Intapp DealCloud and Buildout’s CRM suite. General models cannot watch your inbox, but they turn a pasted thread into structured CRM fields in seconds, which suits low-volume firms. Either way, the automation propagates whatever data hygiene you already have, so clean the duplicate contacts before you connect anything.
Will automated follow-up sequences annoy my contacts?
Automated scheduling won’t; automated content will. Owners and tenants respond to specific, relevant touches (“the flex building on Route 9 traded at $142/SF”) and recognize mail-merge filler instantly. Automate the cadence, draft with AI from real market notes, and keep a human on the send button for high-value relationships. Measure replies and meetings, not opens.
Can AI write listing copy for LoopNet and Crexi?
Yes, and this is one of its strongest CRE use cases, provided you generate from a verified fact sheet rather than freeform. Models reformat one property write-up into portal descriptions, OM narrative, email teaser, and social copy in one pass. The non-negotiable step is checking every number in the output against the fact sheet, because models fill missing details confidently, and a wrong clear height in an OM is a liability, not a typo.
Do we need Microsoft Copilot if we already use Outlook?
Test it before buying anything else, since it is the one mainstream option that sits inside Outlook itself rather than beside it. Copilot summarizes threads and drafts replies in place, which suits triage. For heavier drafting work (listing copy, voice-matched client mail, complex summaries), many teams still prefer ChatGPT or Claude for the control a full prompt window gives. Run both for a month; the usage pattern will tell you what to keep.
How long until a 10-person firm sees value from this?
Days for the drafting layer, a quarter for the full pipeline. Thread summarization and reply drafting pay back in the first week with no integration work. CRM capture and follow-up sequences take 30–60 days including data cleanup. The 90-day sequence in this playbook exists because firms that try to deploy everything at once usually deploy nothing — sequencing is the difference between adoption and shelfware.
The Next Step
If your firm wants a specific answer (which workflow first, which layer of the stack, what your inbox and CRM justify spending), we run a free AI-readiness assessment for small CRE firms. You bring your current workflow and tool stack; we map it against this playbook and tell you where the first real win is, whether or not that involves us.
Book a free AI-readiness assessment.
For the broader operating argument, start with the small CRE firm AI manifesto.
Arthur Wandzel
