Answer every lead and re-win your back book: 54% of brokers are still deciding

Purchase leads don't wait for business hours

A mortgage lead that sits overnight is usually gone by morning, borrowers rate-shop in a burst, and the first responder gets the application. The industry is mid-migration: lender AI use jumped from 15% to 38% in a single year, with 55% of lenders piloting or expanding (Scotsman Guide), while 54% of brokers say they still haven't decided which technology to adopt (A&D Mortgage, 2026). Half the market is modernizing while the other half deliberates, brokers who pick a lane take files from brokers who don't.

The second leak is quieter: your funded-loan book. Every file you closed at a higher rate is a future refinance, but credit bureaus sell mortgage-inquiry trigger leads, the day a past client applies anywhere, competitors get pinged. The defense is reaching that client first, with math specific to their loan.

Entropy & Co builds both systems: speed-to-lead voice and SMS automation on Vapi, Telnyx, and n8n, plus a rate-watch engine that monitors your back book and drafts outreach the moment a refinance pencils. A loan officer approves anything with a number in it. Weighing "AI" against plain automation? Start with agents vs automation, for most broker shops the answer is a mix, and the scope call says which is which.

Workflows we build for mortgage broker shops

After-hours lead answer. A voice agent on Vapi + Telnyx picks up when you can't, qualifies the basics, purchase or refi, timeline, pre-approval status, agent attached or not, books a call on your calendar, and logs the file in your CRM (Jungo, BNTouch, Shape, or GoHighLevel). Response decay is steep; we documented the curve in lead response time statistics, it hits borrowers exactly like buyers.

Rate-watch reactivation. An n8n workflow holds your funded-loan data, note rate, balance, close date, checks it against current pricing from your engine (Loansifter or Lender Price), and flags every file where the improvement clears your threshold: say 75 basis points with a break-even under 24 months. It drafts the message with the borrower's actual numbers, you approve and send. No generic "rates dropped!" blasts.

Pre-approval keep-warm. Pre-approved buyers shop for 60–90 days and go quiet. Weekly automated check-ins keep your pre-approval the live one; aging-document alerts fire before a credit pull or paystub expires mid-offer.

Document chase. The hours sink of every file: paystubs, bank statements, letters of explanation. Checklist nudges over SMS tied to your point-of-sale (Floify or your LOS portal), escalating to your processor only when a borrower stalls past 48 hours.

Milestone broadcasting. Webhooks from your LOS, Arive, LendingPad, or Calyx Point, fire SMS updates to the borrower and the referring agent at submitted, conditionally approved, clear-to-close, and funded. The agent updates are the quiet referral engine: agents send deals to brokers who make them look informed.

Referral-partner nurture. Informational market updates and milestone notes to your agent partners, structured RESPA Section 8-clean, no paid referral mechanics, just being the broker who communicates.

What you get

1. A scoped map of your lead flow, back book, and LOS/CRM stack, with an automation-vs-agent verdict on each piece

2. Voice agent on Vapi + Telnyx running your scripts, your disclosures, and your NMLS ID where required

3. SMS registered through 10DLC (The Campaign Registry) with consent capture, quiet hours, and opt-out handling built in

4. An n8n instance, self-hosted or cloud, in your account, running the rate-watch, document-chase, and milestone workflows

5. Integrations to your CRM and LOS (Jungo, BNTouch, GoHighLevel, Arive, LendingPad, Calyx Point) via API or webhook

6. Pricing-engine connection (Loansifter, Lender Price) for the rate-watch math

7. An LO approval queue: any draft with a rate, payment, or APR waits for a licensed human, templates built around Regulation Z trigger-term rules

8. An exportable audit log of every automated message and call

9. AI-disclosure language on voice and chat agents, mandatory in the EU from August 2, 2026 under AI Act Article 50, and cheap insurance everywhere else

10. A monitoring dashboard and written rollback plan before launch, not after

11. A 30-day post-launch defect window with weekly tuning

Out of scope: buying leads, quoting rates without LO sign-off, and any compensation arrangement with referral partners, that one is your compliance officer's lane, and we build inside it.

The math on one funded file

Market retainers for this vertical run $1,500–$5,000 per month (NetPartners, 2026). Against broker economics: a $350,000 funded file at 150 basis points of lender-paid compensation grosses $5,250, one incremental closing covers one to three months of a market-rate system, and the rate-watch engine points at past clients you already paid to acquire. If the alternative is hiring a human ISA, the ISA cost math runs that comparison line by line.

Want the same math run on your own book? Get a scope and quote.

How an engagement runs

1. Scope call, 45 minutes. Map lead sources, LOS/CRM, book size, and states licensed. Output: a build list with an automation-vs-agent verdict per workflow.

2. Fixed proposal, within 5 business days. Named tools, integration list, compliance checkpoints, and a quote against your scope.

3. Build, weeks 1–3. Voice and SMS first, rate-watch second. You hear every test call before anything answers a real lead.

4. Supervised launch, week 4. Agents run with human shadowing, LO approval queue live from day one. Checkpoint: you review the first 50 conversations before we widen the funnel.

5. 30-day watch. Weekly tuning against booked-call and reactivation metrics, then a handover review: we maintain it, or your team does.

Why Entropy & Co

1. A licensed human approves every number. No automated message quotes a rate, payment, or APR without LO sign-off, the approval queue and audit log sit in your own n8n instance.

2. Built to survive rollback season. 75% of enterprises rolled back customer-facing AI agents by May 2026, citing data exposure (31%) and hallucination (22%) (Sinch, n=2,500+). Our builds ship with the diagnostics and rollback plan those failures lacked, we wrote up why those rollbacks happen.

3. You own the stack. Telnyx numbers, the n8n instance, CRM access, and call recordings live in your accounts from day one. Firing us costs you nothing but the goodbye call.

FAQ

Will the AI quote rates to borrowers?

No. Rates, payments, and APR figures are Regulation Z trigger-term territory and stay with a licensed loan officer. The agent collects the scenario, loan amount, property type, self-reported credit band, timeline, and books the LO call. Drafted rate-watch messages wait in an approval queue.

Is the SMS outreach TCPA-compliant?

The build includes written-consent capture at intake, 10DLC campaign registration, quiet-hours enforcement, and automatic opt-out handling. Rate-watch messages go only to past clients with documented consent. Your compliance officer reviews every template before anything sends, and we supply the audit log.

What does it integrate with?

Broker-side LOS platforms (Arive, LendingPad, Calyx Point), mortgage CRMs (Jungo, BNTouch, Shape, GoHighLevel), point-of-sale tools (Floify), and pricing engines (Loansifter, Lender Price). Anything with an API connects directly; older tools connect through webhooks, scheduled n8n jobs, or email parsing as a stopgap.

We tried a chatbot in 2025 and turned it off. Why would this go differently?

Most deployments failed on missing guardrails, not weak models, data exposure and hallucination led the causes in Sinch's May 2026 survey of 2,500+ enterprises. This build limits the agent to qualification and scheduling, gates every number behind an LO, and ships with rollback criteria you set.

How long until it's answering leads?

Voice and SMS typically reach supervised launch in three to four weeks. The rate-watch engine lands a week or two later, depending on how clean your funded-loan data is, a book living in a CRM moves faster than a book living in spreadsheets. The scope call tells you which you have.

Get a scope and quote

One 45-minute call maps your lead flow and back book, and you get a fixed quote against your actual scope, plus a straight answer on what an AI agent costs versus what the market charges. Get a scope and quote.

Related: AI automation for the full workflow practice, software development when the build needs custom code, and AI automation for real estate teams if you run alongside an agent team.