Real Estate · AI Automation
Automation yourbroker can defend.
Instant lead response, listing workflows, and CRM routing built for the rules real estate actually operates under: TCPA consent before an automated voice dials, Fair Housing review before generated copy publishes, and MLS data moving through the RESO Web API instead of a scraper.
The problem
Why brokerage automation stalls at the pilot
The demos work. What ends them is the part no vendor page covers: whether your database carries TCPA consent per channel, whether the model writing your listing copy knows California names 16 protected characteristics, and whether your CRM will accept the write volume the build assumes.
The AI voice dials before anyone checks consent
On February 8, 2024 the FCC adopted a Declaratory Ruling placing AI-generated voices inside the TCPA definition of "artificial", effective on release. That puts automated voice outreach under the same prior express written consent standard as any prerecorded telemarketing call. Under 47 U.S.C. § 227(b)(3) the exposure is $500 per violation and up to $1,500 for willful or knowing ones, counted per call. A campaign that dials an unconsented lead list is a liability line, not a pipeline.
Generated listing copy is written against the wrong rulebook
A general-purpose model calibrates to the seven federal protected classes in 42 U.S.C. § 3604(c). California Government Code § 12955(c) names 16, adding gender identity and expression, sexual orientation, marital status, source of income, veteran or military status, and genetic information. The statute reaches whoever causes an advertisement to be published, which is the brokerage, not the vendor whose model wrote the sentence.
Routing logic nobody can justify after the fact
Under 24 CFR § 100.500, a practice with a discriminatory effect is tested in three steps, and intent is not a defense. A model that assigns leads by ZIP code, surname, or inferred income can produce that effect with nothing discriminatory anywhere in the code. The framework asks for a documented business justification and a tested less discriminatory alternative, and a routing rule that was configured in a vendor UI rather than written down has neither.
The CRM throttles the build and nobody notices
Follow Up Boss enforces a sliding ten-second window capped at 250 requests globally, 25 for PUT /v1/people, and 10 for notes endpoints, roughly halved without a valid X-System-Key header. Past the ceiling it returns HTTP 429 with a Retry-After header. A bulk enrichment pass over a 40,000-contact database hits that within seconds, and an integration that ignores the header drops records quietly rather than failing loudly.
The system
Three principles we build brokerage automation on
Brokerage automation is not a chatbot problem. It is a TCPA consent problem, a Fair Housing liability problem, and a RESO data-transport problem, in that order, and the order is why most rollouts die at the pilot.
The five-minute rule is an association, and we say so before we sell it
Real estate content repeats a 21x figure on five-minute response as though it were physics. It comes from the Lead Response Management study: three years of call logs, six companies, more than 15,000 leads and 100,000 call attempts, analyzed with MIT. It is cross-industry B2B and observational, and it was never a real estate study, so it shows an association between faster contact and higher qualification odds rather than proof that speed causes the closing. We still treat sub-five-minute response as the right operational target, then measure it against showing appointments booked in your market instead of a fifteen-year-old benchmark from someone else's.
Every automated decision gets a named owner and a paper trail
A routing rule and a lead score are decisions about consumers, and California is regulating them as such. The CPPA regulations effective January 1, 2026 bring automated decisionmaking obligations, including consumer access and opt-out rights, from January 1, 2027, with risk-assessment attestations due to the agency by April 1, 2028. We write the routing logic in plain language before it runs, record why each rule exists, and keep a less discriminatory alternative on file so the 24 CFR § 100.500 questions have answers ready.
Listing data moves on the standard the MLS already publishes
RESO stopped certifying the legacy RETS transport in 2018, so the OData v4 based RESO Web API is now the only certifiable path into MLS data, and at least 90% of MLSs across 489 functioning US systems have a certified service. The Data Dictionary gives us more than 1,700 fields and 3,100 lookups across Property, Member, Office, and Media resources, which is why an automation written for one market can be pointed at another. Refresh cadence sets the clock: MLSListings, covering Monterey and Santa Cruz among five counties, updates every five minutes, and no listing trigger we build fires faster than the feed does.
What's included
What we build inside a brokerage
Six builds, sequenced so the highest-cost miss gets closed first. Each one ships with its consent path, its logging, and its failure behavior defined.
Consent-gated lead response
The channel gate is the product, not the chat window. Every inbound lead is resolved against its own consent record before anything fires: SMS and email run on the permissions actually captured at the form, and AI voice contact never dials without prior express written consent on file, per the FCC ruling. A lead with no consent on a channel does not sit in a queue — it routes to a human immediately, which is the only handling that is both compliant and fast.
Defensible lead routing
Round-robin, geography, price band, source, and performance rules written out before they run, with the reasoning documented and the outcomes effect-tested. Routing is the one rule that decides which agent a buyer meets, which is why it is both the highest-leverage thing to automate in a brokerage and the pattern most exposed to a disparate impact claim. It gets the most scrutiny of anything we build here.
Listing content generation with a Fair Housing gate
Draft descriptions, market summaries, and campaign copy generated from RESO Property fields, then held at a review step calibrated to California's 16 protected characteristics rather than the federal seven. NAR Code of Ethics Article 10 names nine and Article 12 requires a true picture in advertising, so the gate checks both the protected-class surface and the accuracy of every claim about the property.
MLS and CRM integration built for the rate limits
RESO Web API ingestion mapped to the Data Dictionary, and CRM writes queued with exponential backoff that honors Retry-After even when the remaining-request counter suggests headroom. We check IDX terms separately from display rights, because IDX governs what you may show and is not a general license to feed MLS data into arbitrary systems.
Transaction milestone and database reactivation
Escrow and milestone sequences that fire off contract dates, plus reactivation of the leads sitting cold in the CRM. This is where the repetitive brokerage work actually lives, and it does not require a single new lead to produce a closing.
Measurement in showings, not messages
Response-time distribution rather than an average that hides the slow tail, showing and listing appointments set per hundred leads, contact-to-appointment rate by source, and an audit log of every automated decision. A brokerage does not bank replies, it banks tours, so if the automation is sending more and booking the same we would rather find that in month one than defend it in month six.
Results
What we have actually shipped
We have not published a brokerage engagement yet, so we will not imply one. Here is the automation work behind the claim, and two adjacent builds that carry directly.
How we work
Four steps, and the first one is not software
Most failed rollouts skipped step one and discovered their consent gaps after the first automated send.
Inventory the decisions and the consent
We list every decision your systems already make about a consumer, then audit what consent actually exists per contact and per channel: how it was captured, when, and for what. Portal handoffs, open-house sign-in sheets, and imported lead lists rarely carry the prior express written consent an AI voice campaign needs, and they often do not carry SMS permission either. That audit sets what can be automated in week one and what needs a re-permission campaign first.
Build the response layer, channel by channel
Instant acknowledgement and qualification go live on the channels with clean consent. AI voice waits for prior express written consent. Every path has a defined human handoff, because the fastest way to lose a listing appointment is an assistant that will not admit it is stuck.
Wire the data without breaking the CRM
RESO Web API ingestion mapped to the Data Dictionary, listing triggers set to the feed cadence rather than an optimistic one, and CRM writes queued behind a rate limiter with 429 backoff. We load-test against your actual database size before it runs on production records.
Prove it, then keep the audit trail
Response-time distribution, appointments set per hundred leads, and contact-to-appointment rate by source, reported monthly. The routing logic, its business justification, and the tested alternative stay documented and current, so the ADMT obligations arriving January 1, 2027 are a filing exercise rather than a rebuild.
How we compare
A Machina build vs. a typical real estate AI rollout
Most of these gaps stay invisible until an unconsented lead files a TCPA claim, a Fair Housing complaint names a generated listing description, or a bulk job silently stops writing to the CRM.
| Typical AI vendor rollout | Machina |
|---|---|
| AI voice outreach pitched as use case #1, consent never mentioned | Voice gated behind prior express written consent per the FCC ruling; SMS and email scoped separately |
| Listing copy generated against the seven federal protected classes | Review gate calibrated to California's 16 characteristics and NAR Article 10's nine |
| Lead scoring and routing treated as a black box | Routing rules written in plain language with justification and a tested alternative on file |
| MLS data scraped or pulled from a client-side IDX widget | RESO Web API ingestion mapped to the Data Dictionary, with IDX use terms checked separately |
| Integration built as if the CRM had no request ceiling | Queued writes with exponential backoff that honors Retry-After on HTTP 429 |
| Listing triggers fired on an assumed real-time feed | Triggers set to the actual MLS cadence; MLSListings refreshes every five minutes |
| Success reported as messages sent and response time averaged | Appointments set per hundred leads and full response-time distribution, tail included |
| Self-reported vendor benchmarks presented as if causal | Cross-industry observational research cited as association, with the study named |
| Your workflows and data locked inside the vendor platform | Built on your CRM and your MLS credentials, documented so you can run it without us |
Pair with
The rest of the real estate stack
Automation answers the lead. These decide how many arrive, and what they see when they do.
Keep reading
Go deeper
Real Estate AI Automation
Find out what your automation is allowed to do before you build it
Start with an automation audit. We will map every automated decision your systems already make, check the consent state of your database channel by channel, and show you which build returns a closing first. Twenty minutes, no obligation.