AI in real estate,
shipped on your Yardi, MLS, or Zillow stack. Not on a slide deck.

An `ai real estate company` like ours ships production AI workflows on your real-estate stack. Not slide decks, not pilots that die at month 4. The day-to-day work:

• Scope the workflow that moves a P&L line: most often lead-to-tour first-touch, leasing chatbot, AI CMA + listing copy, tenant-screening triage, or CRE underwriting memo.
• Get the integrations in place against your Yardi / AppFolio / RealPage / Buildium / Zillow / MLS tenant.
• Pick the right model per workflow: Sonnet 4.6 for narrative and reasoning, Haiku 4.5 for high-volume routing, GPT-5-mini for structured rent-roll and application extraction.
• Bake in FHA prompt-review on every leasing path and FCRA-conscious flows on every tenant-screening path; ship behind a licensed-agent-in-loop flag.
• Operate the workflow long enough to prove cost-of-ownership before scaling.

We do not sell a product. We ship one workflow at a time and report cost-per-decision monthly. If you want a `real estate ai company` that delivers a live integration in four to six weeks, this is it. Most engagements start with real-estate AI advisory — a 1-2 week discovery audit that maps FHA / FCRA exposure per workflow before any build commits.

Yes, all four are buyer-vocab anchors and each has a different integration shape.

• `Yardi ai` integrations go through the Voyager API (or for legacy tenants, the older webservice layer). Yardi is the largest multifamily PMS by units, the API is stable, and rate limits are reasonable for chatbot-volume calls.
• `AppFolio ai` integrations use the AppFolio REST API: friendly auth flow, decent webhook coverage, slightly tighter rate limits than Yardi.
• `RealPage ai` covers a portfolio of products (OneSite, Spherexx, Lead2Lease, IMS); the integration depends on which RealPage product is your system of record, and we'll scope the right API surface in the audit.
• `Buildium ai` is the smaller-portfolio operator's tool. REST API is clean, webhooks exist but reconciliation jobs are a good idea for production reliability.
• `Ai mls integration` covers Bright MLS, CRMLS, Realtor.com, and brokerage-side feeds. Listing intake + comp pulls work; field-mapping per MLS region is the engineering, not the model.

The audit names which integrations are clean and which are webhook-fragile before any pilot ships. Adjacent build: property ops automation for leasing follow-up, tenant-screening triage, and renewal workflows on the same PMS spine.

We design every leasing and lead-handling AI to be Fair-Housing-Act-aware. To be clear: "FHA compliance" is a continuous design and review practice, not a one-time certification. Operator specifics:

• FHA prompt review pre-launch: every leasing-chatbot prompt path goes through an FHA review before pilot launch. No protected-class signal (race, color, religion, national origin, sex, familial status, disability) is allowed in the prompt context, the response-shaping prompt, or the disqualification logic.
• Audit log on every reply with the full prompt, response, retrieved context, and the model + version used. If a fair-housing claim ever surfaces, the audit log is the defense.
• Steering-adjacent language filtered (no "good neighborhood for families with kids," no "quieter building," no "safer area") via a deny-list run on every outbound reply.
• Adverse-action-adjacent questions escalate (rejected-application appeals, denied-tour reasons) to a human leasing manager. The AI does not adjudicate.
• Disqualification logs include a written, model-generated rationale that the leasing manager reviews.

What we don't claim: an FHA "certification". There isn't one. What we ship is a defensible operator practice that survives plaintiff-side review.

AI can triage tenant screening. It cannot deny a tenant. The Fair Credit Reporting Act is explicit: when a denial is based in whole or in part on a consumer report, a human reviewer must make the adverse-action call, and the applicant must receive a written adverse-action notice with reasons, the consumer-reporting-agency contact information, and a dispute pathway.

• The AI reads the application and surfaces inconsistencies (income vs. rent ratio out of policy, employment gaps, eviction-record flags, credit-score band), then routes to a human leasing manager with a structured triage report.
• The leasing manager reviews, decides, and issues the adverse-action notice (or approves).
• The AI never denies, never auto-rejects, never auto-issues the notice.
• Every triage report is audit-logged.

If a vendor pitches autonomous tenant denial or AI-issued adverse-action notices, walk. The FCRA exposure is high four to five figures per violation and class actions stack.

An `ai cma` workflow runs in two cost buckets:

• Model layer on Sonnet 4.6 with prompt caching warm: roughly $0.14–$0.22 per CMA for a typical residential comp pull (comp set retrieval + narrative + sensitivity range).
• Infrastructure layer (MLS feed pulls, comp-database lookup, PDF generation, audit-log writes): typically $0.06–$0.12 per CMA.
• Total: roughly $0.20–$0.34 per CMA. An agent running 40 CMAs a month sits around $8–$14/month in model + infra. On a 12-agent brokerage, model + infra usually lands in the $200–$400/month range.

The honest cost ceiling is engineering: getting the MLS field mapping right (every MLS has slightly different field names for the same attribute), getting the comp-cluster logic to defensibly justify which 8 comps were selected from 80 candidates, and getting the listing-description voice to match the brokerage tone. That work is the engagement, not the per-CMA spend.

Three places we'll say no, covered in §8 above and worth repeating:

• Fair-Housing-Act steering risk: we will not ship a leasing chatbot, lead-qualification flow, or recommendation engine that selects, ranks, or describes housing in a way that could constitute steering. That includes subtle failure modes like tone-shifting replies based on inferred demographic signals from name, language, or zip code.
• Autonomous tenant denial: FCRA requires a human reviewer on every adverse-action call when the denial is based in whole or part on a consumer report. We ship triage and risk-flagging; we do not ship the denial.
• AI giving real-estate "advice" across state-license boundaries: pricing strategy, negotiation, disclosure interpretation are licensed activities in most US states; our chatbots provide information (here's the comp set, here's what the data shows) and the licensed agent provides advice.

Beyond those three: any workflow where the metric won't move within the pilot window, where the data isn't clean enough to build a baseline, or where the regulatory posture is unclear. We'll say so in the audit before the pilot.

No. Augment, not replace. Pricing strategy, negotiation, advice-shaped client conversations, and the adverse-action call stay with licensed agents and trained leasing managers. The AI we ship for real-estate organizations is decision-support and rate-mover:

• Lead-to-tour: replies to the inbound lead in 90 seconds and qualifies on budget / timeline / pre-approval.
• Listing + CMA: drafts the listing copy and the CMA from the comp set.
• Leasing chatbot: handles tour-scheduling, applicant WISMO, rent-reminder nudges, and renewal nudges.
• Tenant-screening triage: surfaces inconsistencies on tenant applications and routes risk-flagged cases to a human.
• CRE workflows: drafts the first-pass underwriting memo and the quarterly IR update from the operating data.

In every pattern we ship, the licensed agent (or the leasing manager, or the analyst, or the IR officer) signs off on the consequential decision and the AI's output goes to them with the rationale pre-written, saving 60–90% on inbound-lead first-touch latency, 6+ minutes per resident touchpoint, and 4+ hours per CRE deal underwrite. That's the realistic claim a `real estate ai company` should make: agents keep doing the deals; the AI handles the first-touch, the drafts, the routine routing, and the structured extraction.

Three tiers, pricing-locked across the cluster:

• `Real estate ai consulting` audit: Fixed-fee, 1–2 weeks. We shadow brokerage / leasing / underwriting team, score candidate workflows by hours returned × time-to-ship × FHA/FCRA/license risk, and deliver a 90-day roadmap with per-workflow cost bands and an honest "these won't pay back yet" list.
• Pilot to production: fixed-bid fixed, 4–6 weeks. One workflow shipped end-to-end on your Yardi / AppFolio / RealPage / Buildium / Zillow / MLS stack, FHA-reviewed, FCRA-conscious, audit-logged, licensed-agent-in-loop, with a walk-away point at week 5. If the metric won't move, we stop before production hardening and you don't pay phase 2.
• Continuous real-estate AI team: monthlynth, no annual contract. Embedded PM + AI engineer + real-estate-ops analyst shipping the next workflow on your roadmap, with per-workflow monthly cost-of-ownership reporting and a compliance review cadence on the FHA + FCRA prompt paths.

Most `ai real estate company` engagements we run start with the audit, ship the first workflow on the pilot, and move to monthly for workflows two through five. Cost-per-decision reported monthly on every shipped workflow. No per-decision number, no engagement.