82% of agents use AI: only 17% see results. We build the system that closes it.

The 17% problem

Between 68% and 82% of agents now use AI tools, per NAR's surveys, yet only 17% report a significant positive impact, and 46% report none at all. That is not an adoption problem. It is a systems problem: a ChatGPT subscription writes a listing description, but it does not answer the Zillow inquiry that lands at 2:47 AM, route it to the right agent inside 5 minutes, and follow up for the months a typical buyer takes to transact.

The response-time data shows where the money leaks. In WAV Group's 2014 responsiveness study (384 brokers, 11 states), 48% of buyer inquiries never got a response, and the average first response took 917 minutes. Roof AI's November 2025 secret-shopper test of 74 top US brokerages found 41% still never responded, and only 9% replied within 5 minutes. Eleven years and several CRM generations moved the never-responded rate seven points. We keep every one of these numbers dated and sourced in our real estate lead response time statistics reference.

We build the part the subscription does not include: intake, routing, first touch, and 12-month nurture, wired into the CRM your team already runs. Teams and brokerages only, the economics below explain why a solo agent should not pay for this.

What you get

Every item names the tool it runs on:

• Lead-source audit: Zillow, Realtor.com, Homes.com, sign calls, and website forms mapped into one intake layer

• Sub-60-second first touch: SMS via Twilio or Telnyx, optional AI voice first call on Vapi with a disclosure line

• Claim-or-release lead routing in Follow Up Boss, kvCORE, Sierra Interactive, or Lofty, unclaimed leads escalate after 5 minutes

• 12-month buyer and seller nurture (email plus SMS), branched by timeline and price band, written and loaded, not templates left for later

• Open house capture: QR sign-in (Open Home Pro, Spacio, or self-hosted Tally) to an n8n webhook to your CRM, with same-hour segmented follow-up

• Sphere reactivation: purchase-anniversary, rate-watch, and home-valuation triggers across your past-client book

• A2P 10DLC registration and TCPA consent capture on every form before a single text goes out

• Listing-launch checklist automation and showing-feedback chase

• Response-time, claim-rate, and per-source ROI dashboards with an agent-written weekly report

• Audit log of every automated message, with a kill switch per workflow

What we do not do: sell you leads, run unsupervised AI on your past clients, or take the buyer-consult conversation away from your agents.

Three builds, end to end

The AI ISA lane. A human ISA costs $50,000–$85,000 a year fully loaded and the seat turns over roughly annually; AI ISA platforms run $100–$500 a month plus about $0.25–$0.50 per managed minute. The hybrid wins, and we show the break-even arithmetic in our AI ISA vs human ISA cost math: software takes the first 60 seconds and the overnight inquiries, a human takes the warm hand-off and every pricing conversation.

Open house follow-up. Fourteen sign-ins on Sunday should not decay into six contacts by Tuesday. QR sign-in to n8n to CRM, buyer/neighbor/agent segmentation, SMS inside the first hour, the full open house follow-up build is published node by node with an itemized stack cost of roughly $75–80 a month.

Listing-side ops. Pre-listing checklist runs, photo and copy QA, showing-feedback chases, and price-reduction triggers tied to days-on-market. The operator version lives in our playbook for listing agents.

How it ships

1. Scope call, week 0. Lead sources, CRM, team structure, and your real response times pulled from your own data. You get a fixed quote.

2. Build, weeks 1–3. Workflows in n8n against your CRM's API, 10DLC registration filed, every nurture message written. You approve each template before anything sends.

3. Supervised launch, weeks 3–4. New leads only, daily log review, your team sets the hand-off thresholds.

4. Handover, week 5+. Dashboards live, your ops lead trained, kill switches documented. We stay on for tuning or hand you the keys, you own the accounts either way.

Proof you can check

We publish the homework. The response-time reference above is our own primary-source audit, every stat dated, every sample size named. The open house build is public down to the n8n nodes and the monthly invoice. Client builds ship with the same documentation standard: message templates, workflow diagrams, and an audit log you can read without us on the call.

That standard exists because the failure mode is documented: 75% of enterprises have rolled back customer-facing AI agents, with data exposure (31%) and hallucination (22%) the top causes, per Sinch's May 2026 survey of 2,500+ companies. We cover the pattern in why companies are rolling back AI agents, and we build agents that survive it.

Get a scope and quote

What the market charges

Scope drives price, but the market bands are public. Agencies running this category of work for real estate teams charge $1,000–$4,000 a month in retainers. AI ISA platforms add $100–$500 a month base plus per-minute usage. What moves the number: how many lead sources need wiring, your CRM's API quality (Follow Up Boss is clean, some legacy systems are not), SMS volume against 10DLC throughput limits, voice minutes, and how many agents sit on the routing tree.

The yardstick is one recovered deal. A 2.5% commission side on a $400,000 sale is $10,000 of gross commission, one inquiry rescued from the 41% that currently get no answer covers months at the top of that retainer band. We quote your exact scope at the link below, in writing, before any contract.

Why Entropy & Co

Three claims you can check:

1. Disclosure by default. Every voice agent ships with an AI-disclosure line. The EU AI Act's Article 50 makes disclosure mandatory in the EU from August 2, 2026, we treat it as the floor everywhere.

2. Model-agnostic plumbing. Routing and logic live in n8n, not inside any vendor's product. Five frontier model releases shipped in seven weeks this spring; your stack should not be married to one of them.

3. Rollback-safe. Every automated message is logged, every workflow has a kill switch, and every hand-off threshold is yours to set and audit.

FAQ

Can AI legally text my leads?
Only with prior express written consent. We register your numbers under A2P 10DLC, capture TCPA consent on every form, honor quiet hours, and wire opt-outs into the CRM itself, a stop request halts every sequence at once, not just the one that triggered it.

Does this work with our CRM?
Follow Up Boss, kvCORE, Sierra Interactive, Lofty, and Real Geeks all expose APIs or webhooks we build against. If your system cannot support claim-or-release routing or event-triggered sends, we say so before contract, sometimes the honest first step is a CRM migration, and we will tell you.

What should the AI never handle?
Pricing objections, commission conversations, grief, divorce, and estate sellers, and repeat clients, those route to humans by explicit rule. AI takes first touch, qualification, and scheduling. The hand-off thresholds are documented in your runbook and yours to tighten at any time.

We already bought AI tools that did nothing. Why is this different?
You are in the 46% NAR found reporting no impact. A tool answers one task; this is intake-to-close plumbing with measurement attached. We audit your existing stack first, most teams keep their CRM and most of their current subscriptions.

How long until it is live?
Two to five weeks for most teams: scope call, build against your CRM's API, supervised launch on new leads only, then handover. A2P 10DLC registration is the usual long pole, carrier approval takes days, so the filing goes in during week one.

Get your build scoped

Send your lead sources, CRM, and team size, you get back a fixed quote and the workflow map we would build. Get a scope and quote

Related: AI automation is the parent service. See paid ads when lead volume itself is the constraint, email marketing for the nurture layer alone, and software development when your team needs custom tooling beyond workflows.