Hire an AI agent that pulls comps from your MLS, drafts listing copy in your brand voice, enters new listings into MLS / Zillow / Realtor.com, and drafts post-tour follow-up emails. Cost: ~$150/month. The agent works from Slack — you ping it with the property address and it ships back the CMA, the listing, the photos-organized folder, and the social-media post variants.
Before and after
| Task | Manual process (today) | With Maya |
|---|---|---|
| CMA for a listing pitch | Agent spends 2-3 hours pulling comps, adjusting for condition and recency, formatting a PDF. Often done the night before a pitch. | Agent texts @maya with the address at 8:30am. By 8:38am: 5 active comps, 8 sold-in-90-days, suggested price band, polished CMA PDF in shared drive. Ready for a 10am pitch. |
| MLS listing entry | Agent or coordinator manually fills 40+ MLS fields from the listing agreement, property disclosure, and county records. 30-45 min per listing. Errors common. | Agent forwards listing agreement. Maya pulls county records and tax assessor data, pre-fills MLS fields, flags missing items for agent to supply. Agent reviews pre-filled form and approves. 5-10 min agent time. |
| Cross-posting to portals | Coordinator logs into Zillow, Realtor.com, brokerage site separately. 30+ min per portal. Inconsistencies between portal versions common. | Maya syncs from MLS to all connected portals simultaneously on agent approval. Single source of truth. Confirms in Slack with live URLs. |
| Post-tour follow-up | Agent drafts 4 emails after a day of tours. 15-20 min each. Often delayed to next morning. Buyers wait 24-36 hours for response. | Agent pings @maya with buyer names and tour notes. Maya drafts 4 personalized emails in 3 minutes, queues for agent review. Buyers get follow-ups the same afternoon. |
| Open house follow-up | Coordinator manually emails each sign-in attendee. High-volume open houses produce 20-40 names. Follow-up often incomplete or templated. | Agent forwards digital sign-in or photos the sheet. Maya drafts personalized emails for every attendee referencing the property, current competition, and a call-to-action. Batched send with agent approval. |
What it does
| Task | Time before | Time after |
|---|---|---|
| Comp analysis (CMA) for a listing pitch | 2-3 hours | 8-12 minutes |
| Listing copy + 5 social variants | 1-2 hours | 5 minutes |
| MLS data entry (40+ fields) | 30-45 minutes | 3-5 minutes |
| Cross-posting to Zillow / Realtor.com / brokerage site | 30 minutes per portal | Automatic |
| Post-tour follow-up email (personalized to each buyer) | 15-20 min each | 2-3 minutes each |
| Open house sign-in follow-up (24h after) | 10 min × N attendees | Batched, automatic |
| Pre-tour buyer brief (neighborhood, schools, recent activity) | 30-45 minutes | 4-6 minutes |
A day in the life
Tuesday at a 12-agent residential brokerage. Your AI listing coordinator is named "Maya":
- 8:30 AM: Agent texts
@maya pull comps for 1247 Maple Ave for a 10am pitchin Slack. - 8:38 AM: Maya posts: 5 active comps, 8 sold-in-90-days comps, suggested list-price band, suggested marketing positioning, link to a polished CMA PDF in the shared drive.
- 10:42 AM: The pitch went well — listing won. Agent forwards the listing agreement to
maya@yourbrokerage.com. - 10:50 AM: Maya pulls property data from county records, tax assessor, last MLS history. Drafts MLS listing description in the brokerage's voice. Posts in
#listings-pipeline: "1247 Maple Ave draft ready — agent please review the description and approve photos before MLS submission." - 11:15 AM: Agent reviews, makes 2 edits, hits "Approve" in Slack.
- 11:18 AM: Maya enters the listing in MLS, cross-posts to Zillow / Realtor.com / brokerage site / Instagram (image carousel). Confirms in Slack with the live URLs.
- 4:30 PM (after a tour): Agent pings:
@maya 4 buyers from today's tour, send follow-up. Maya drafts 4 personalized emails based on the agent's tour notes, queues for review. - 4:45 PM: Agent skims, hits "Send all," done. 4 buyers got contextual follow-ups in 15 minutes total instead of 60-90.
Scenario 2: Open house day
Saturday. The agent is running two open houses back to back. 23 people sign in across both properties. Normally: 2-3 hours of Monday follow-up emails. With Maya:
- Saturday 4pm: Agent photos both sign-in sheets, sends to
maya@yourbrokerage.comwith a note: "Oak Street had 14 sign-ins, Maple Ave had 9. Oak Street had a couple who seemed very interested — John and Sarah, the last two on the list." - Saturday 4:12pm: Maya reads the photos (OCR), extracts names and contact info, cross-references with CRM for existing leads. Drafts 23 personalized follow-up emails.
- Saturday 4:20pm: Maya posts in Slack: "23 follow-up emails drafted. John and Sarah flagged as hot leads based on your note — their emails have a more direct call-to-action. Preview all 23 in the drafts folder. Ready to send on your approval."
- Saturday 4:25pm: Agent skims 5 emails (spot-check), hits "Send all." 23 buyers receive personal follow-ups while the open house is still fresh. John and Sarah get a direct "would you like to schedule a private showing?" email while they're still in the car discussing the property.
The agent spent 25 minutes on admin for two open houses instead of 3 hours on Monday. Response rate on same-day follow-ups is measurably higher than next-business-day follow-ups.
Cost math
| Component | Monthly cost |
|---|---|
| Sandbox compute (ab0t.medium, auto-stop) | $10 |
| Model API (Claude Sonnet for drafting + Opus for occasional polish) | $70 |
| Sandbox Platform fee | $49 |
| MLS / portal access (already part of brokerage subscription) | $0 incremental |
| Storage | $1 |
| Total | $130-180/mo |
Compare to a virtual assistant ($1,500-3,000/month) or a transaction coordinator ($4,000-5,000/month full-time). 10-25× the leverage.
For a brokerage doing 15-30 listings/month, Maya saves ~50-80 admin hours/month per agent. The agent's freed-up time goes to client meetings — the work that actually sells houses.
ROI table: 10-agent residential brokerage, 20 listings/month
| Task | Human time (before) | Agent time (after) | Hours saved/month |
|---|---|---|---|
| CMA preparation | 2.5 hr × 20 = 50 hr | 0.2 hr review × 20 = 4 hr | 46 hr |
| MLS data entry | 0.6 hr × 20 = 12 hr | 0.1 hr review × 20 = 2 hr | 10 hr |
| Portal cross-posting | 1.5 hr × 20 = 30 hr | 0.1 hr approval × 20 = 2 hr | 28 hr |
| Post-tour follow-ups (4/week) | 1.3 hr/week × 4 = 22 hr | 0.2 hr review × 4 = 3 hr | 19 hr |
| Open house follow-ups (2/month) | 3 hr × 2 = 6 hr | 0.3 hr approval × 2 = 0.6 hr | 5.4 hr |
| Total | 120 hr/month | 11.6 hr/month | 108 hr/month |
108 hours/month at average agent value of $75/hr = $8,100/month in recaptured agent time at a cost of $150/month. That's 54× leverage before counting the quality improvements (faster follow-ups, consistent listing copy, fewer MLS entry errors).
Property-type customization
Maya's CLAUDE.md should tell her which property types your brokerage handles and what changes for each:
| Property type | CMA methodology differences | Listing copy differences | Key CLAUDE.md additions |
|---|---|---|---|
| Residential (single-family) | Adjust for beds/baths, lot size, condition, garage. 0.5-mile radius, 90-day window standard. | Lead with lifestyle and neighborhood. Schools, walkability, community feel. | School district lookup, walkability score integration, neighborhood amenity list |
| Condos / townhomes | HOA fees as a cost component. Comp by floor, view, building age. Same-building comps preferred. | Amenities-forward: pool, gym, concierge, parking. HOA what's included. | HOA fee lookup source, building amenities list, parking details |
| Luxury ($1M+) | Wider radius (2-3 miles). Smaller comp set (15-20 properties). Price/SF as primary metric. | Elevated tone, specific material callouts (Sub-Zero, Venetian plaster, heated floors). No hype words — understate and let the property speak. | "Luxury voice" rules, materials database for the area, list of banned words (stunning, gorgeous, rare opportunity) |
| Commercial | Cap rate and NOI-based pricing for investment properties. Price/SF for retail/office. Lease terms matter. | Investor-focused: cap rate, tenant profile, lease structure, upside potential. Different audience than residential buyers. | Commercial comp database credentials, cap rate calculator, zoning lookup integration |
| Rentals | Comp by unit type, amenities, neighborhood. Seasonal adjustment for rental market. | Tenant-focused: price per utility, pet policy, move-in costs, commute convenience. | Rental portal integrations (Apartments.com, Zillow Rentals), move-in fee schedule |
What goes in Maya's CLAUDE.md
You are Maya, the AI listing coordinator at [Brokerage Name]. You work with a 10-agent residential brokerage specializing in suburban properties in [Metro Area].
What you can do: Pull comps from MLS and produce CMAs. Draft listing descriptions in brokerage voice. Enter listings into MLS. Cross-post to Zillow, Realtor.com, and the brokerage website. Draft post-tour follow-up emails. Draft open-house sign-in follow-up emails. Manage the listing timeline checklist (photos scheduled, sign ordered, disclosures received).
Voice rules: Warm, factual, neighborhood-focused. Lead with lifestyle (the morning coffee on the patio, the 10-min walk to the elementary school). No hype words: banned list includes "must-see," "rare opportunity," "stunning," "gorgeous," "you'll love it." Always mention school district and walkability score. Listings are for "homes," not "houses" or "properties." Closing line: "Represented by [Agent Name], [Brokerage Name]."
Fair-housing check (mandatory on every listing): Remove any language implying preference for a protected class. Flag school-quality descriptors that could proxy for demographics. Remove "family-friendly" unless describing a specific community feature. Post-check: confirm description passes the HUD self-test.
Comp methodology: Residential single-family: 0.5-mile radius, sold within 90 days, adjusted for beds/baths (±$15K per bed, ±$10K per bath), adjusted for lot size (±$5/SF deviation from median), adjusted for condition (±3-5%). Include 5 active and 8 sold comps minimum. Price band = low-end adjusted to high-end adjusted.
Escalate to [Agent Name] when: A property is over $1.5M (luxury pricing requires partner judgment). An MLS field has no obvious answer from the listing agreement. A fair-housing language check finds a borderline item I'm unsure about.
Onboarding
- Sign up at the dashboard. Use the "Real Estate Listing Coordinator" template.
- Connect your MLS credentials (most major MLSs supported via the platform's adapters; check the integrations tab).
- Connect Slack and the email address you want the agent on (e.g.
maya@yourbrokerage.com). - Connect Zillow / Realtor.com / your brokerage site if you want auto-cross-posting (optional; some agents prefer to manually approve each cross-post).
- Customize the CLAUDE.md: brand voice ("warm, neighborhood-focused, no hype"), preferred phrasing ("call them 'homes' not 'houses'"), photo organization rules.
- Run a dry-run on an old listing — verify the CMA matches what you'd produce manually, the listing copy is on-brand.
Total: 30-45 minutes including the dry-run. The MLS connection takes the longest because every region's MLS has different auth.
30-day supervised trial
Don't throw Maya into live listings on day one. Run her in shadow mode the first two weeks — she produces the output, you compare it to what you'd have done. The goal: catch the 3-5 CLAUDE.md gaps before they reach a live listing.
| Week | What Maya does | What you do | Goal |
|---|---|---|---|
| Week 1 | Produces CMAs for every listing pitch — shadow only (you don't show these to sellers yet) | Compare Maya's comp sets to your manual CMAs. Note where she under-adjusts or over-adjusts. | Calibrate comp methodology in CLAUDE.md. Should agree with your manual output 85%+. |
| Week 2 | Drafts listing descriptions for every new listing — shadow only | Edit Maya's drafts, note recurring tone issues. Update voice rules in CLAUDE.md each time. | Draft quality should reach a point where you're editing 1-2 sentences, not rewriting paragraphs. |
| Week 3 | First live CMA (you review and present). First live listing draft (you approve before submission). | Approve or reject each piece before it leaves the brokerage. Still require 100% review. | First real-world test. Expect 1-2 CLAUDE.md tweaks based on live feedback. |
| Week 4 | Handles MLS data entry and portal cross-posting with your approval step. Sends post-tour follow-ups on agent approval. | Move from reviewing everything to spot-checking 30% of output. Flag issues to correct immediately. | By end of Week 4: Maya is in production with a one-step agent-approval gate on all outbound. |
| Month 2+ | Autonomous on CMA and listing drafts. Agent reviews before any outbound contact. | Review outputs, not inputs. Monthly CLAUDE.md tune-up based on recurring corrections. | Target: agent spends 10-15 min/day supervising vs 3-4 hours on admin. |
What Maya doesn't do
- Negotiate offers. She drafts response scripts; the agent negotiates.
- Sign contracts. Drafts only; agent reviews and signs.
- Make pricing decisions. Provides comp-based suggestions; agent picks the actual list price.
- Show properties. Pre-tour briefing, post-tour follow-up — but the human shows the home.
- Pre-screen buyers. Many states' fair-housing rules require human judgment on buyer qualification — keep that human-only.
Fair-housing considerations
Real estate has specific compliance requirements. Build these into the CLAUDE.md from day one:
- Listing language: No language that implies preference for any protected class. The agent's CLAUDE.md should include a "fair-housing language check" rule on every listing draft. Specifically banned: implying buyer demographics, school-quality euphemisms that proxy for race, "family-friendly" without context, etc.
- Buyer outreach: Maya's follow-up emails should be content-driven (about the property, the process), not demographically targeted.
- Audit trail: Every listing description, every email, every comp set is logged. If a fair-housing complaint comes in, you have the artifact.
- Photo descriptions for accessibility: Alt-text for images, at minimum. Some MLSs require it.
This isn't legal advice; check with your broker's E&O carrier and your local board.
Maya for a single broker vs Maya for a 12-agent brokerage
Two deployment patterns:
- One Maya per agent. Each agent has their own Maya in their own Slack DM. Pro: personalized to the agent's voice and workflow. Con: 12× the cost.
- One Maya for the brokerage (recommended). All agents share Maya. She lives in
#listings-pipeline, agents @-mention her with their listing requests. Pro: 1× cost, easier to keep brand voice consistent. Con: agents have to be more explicit about which property they mean.
For 5-15 agent brokerages: pattern 2 is overwhelmingly the right choice. Above 20 agents, consider 2-3 Mayas split by team or region.
Common mistakes brokerages make
Six patterns that cause brokerages to abandon Maya in the first 90 days — all avoidable:
- Skipping the brand voice setup. Maya goes live with a generic CLAUDE.md. First listing comes back sounding like every other brokerage in the market. Agents complain. The fix isn't a new AI — it's 45 minutes writing voice rules and dropping 10 example listings. Do this on day one, not after the first complaint.
- Forgetting the fair-housing language check. Maya drafts efficient listing copy, but without an explicit fair-housing check in CLAUDE.md, she'll occasionally include school-quality language or neighborhood descriptors that proxy for protected classes. Build the check into the CLAUDE.md template, not as an afterthought after a compliance flag.
- Auto-publishing without approval step. Some brokerages set Maya up to post directly to MLS and portals without agent approval. One MLS entry error or one tone-deaf listing copy and the trust is gone. Keep the approval gate for at least 90 days, then revisit.
- Using Maya for luxury listings before tuning the luxury CLAUDE.md. Luxury copy has completely different conventions — understated, materials-specific, no hype words. Maya's residential defaults produce copy that reads like a mid-market listing for a $2M property. Separate the CLAUDE.md sections by property type before going live on luxury.
- Not updating comps methodology for local market conditions. The default comp methodology (0.5-mile, 90 days) doesn't fit every market. In fast-moving markets, 30 days is more accurate. In rural markets, 2-mile radius is necessary. If Maya's CMAs are consistently high or low, the fix is usually the comp window and radius settings in CLAUDE.md — not the AI itself.
- One Maya for 20+ agents with no segmentation. Above 15-20 agents, a single shared Maya becomes a bottleneck and a context-collision risk (Agent A's listing requirements bleed into Agent B's thread). At 20+ agents, split into 2-3 Mayas by team or property type before agents start complaining about inconsistency.
Frequently asked questions
Will the MLS approve us using AI for data entry?
Most do — the listing is still attributed to your agent (the human MLS member). The AI is a tool, like a spreadsheet. Some MLSs require disclosure; check your specific MLS's policies.
Can an AI listing coordinator handle all property types?
Yes, with customization. Residential, commercial, luxury, and rentals each need different CLAUDE.md context: different comp methodology, different listing description conventions, different portal targets. One Maya can handle multiple property types if you configure the CLAUDE.md to distinguish them — separate sections for each type works well.
How do I make sure listings sound like our brokerage?
Two steps: (1) drop 10 of your best past listings into the workspace as examples; (2) write explicit voice rules in CLAUDE.md — tone, banned words, required elements. After 1-2 weeks of editing Maya's drafts, she'll be 90% on-brand without corrections.
What about fair-housing compliance?
Build a fair-housing language check into CLAUDE.md from day one. Banned: language implying preference for protected classes, school-quality euphemisms that proxy for demographics, "family-friendly" without context. Every listing draft gets the check before it reaches the agent for review.
Can Maya handle showings?
No — and you don't want her to. Showings are where humans earn their commission. Maya handles the pre-tour brief, the scheduling, and the post-tour follow-up. The showing itself is always human.
Should I have one Maya for the whole brokerage or one per agent?
For 5-15 agents: one brokerage Maya, all agents @-mention her in a shared Slack channel. Above 20 agents: consider 2-3 Mayas split by team or region. One-per-agent is 10× the cost with limited additional value at small scale.
How accurate is Maya's CMA?
Maya's CMA is as accurate as the MLS data she pulls from and the comp methodology you define. She applies your defined parameters (radius, recency, square footage adjustment, condition). Accuracy equals a well-trained human assistant running the same methodology. The agent still reviews and makes the final list-price recommendation.
What MLS systems does Maya integrate with?
Most major US regional MLSs via RETS and RESO Web API adapters. Bright MLS, CRMLS, NTREIS, Stellar MLS, Realcomp, and ARMLS are pre-built. Smaller regional MLSs may need a custom connector (half-day IT project). Check the integrations tab for your specific MLS.
Can Maya manage the transaction timeline after listing?
Yes — with transaction coordinator tasks added to her CLAUDE.md. Maya can track contingency deadlines, send reminder emails to parties, update transaction management software (Dotloop, SkySlope, Brokermint), and flag milestone slippage to the agent. This is a separate configuration from the listing coordination role.
What happens during a price reduction?
Agent says "Maya, we're reducing 1247 Maple Ave to $485K, update everything." Maya updates MLS, re-publishes to portals, drafts a price-reduction announcement email to the buyer list, and drafts a social media update. Agent reviews and approves each outbound piece before it goes.
Can Maya handle expired listings and re-listing?
Yes. On expiry: Maya flags the listing, pulls updated comps, and drafts a re-listing pitch memo with revised positioning and price suggestion. The agent uses the memo for the seller conversation. After re-listing agreement: Maya re-enters MLS with updated copy and photos.
How does Maya handle open house logistics?
Maya creates the open house event in MLS and portal listings, drafts open-house social posts (Facebook Event, Instagram Story, LinkedIn), and sets up the sign-in follow-up sequence. After the open house: agent forwards the sign-in sheet, Maya drafts personalized follow-up emails for every attendee.
What about buyer lead nurturing?
Maya can manage a drip nurture sequence for buyer leads — weekly or bi-weekly emails with relevant new listings, market updates, and neighborhood intel. The sequence is templated in CLAUDE.md; Maya personalizes each send based on the buyer's stated criteria. Agent reviews and approves before sending.
Is the $150/month cost per agent or per brokerage?
Per Maya instance. A brokerage-wide Maya shared by 10 agents costs ~$150-180/month total — not per agent. That's $15-18/agent/month versus $1,500-3,000/month for a human virtual assistant.
How do I set listing description quality standards?
Three techniques: (1) provide example listings in the workspace with a note "this is the gold standard"; (2) add explicit rules in CLAUDE.md — required sections, banned phrases, character limits per section; (3) run a grading rubric in CLAUDE.md so Maya self-checks before submitting. The combination produces consistent, on-brand copy within 1-2 weeks.