It’s 9:40 on a Tuesday in Frisco. You’ve got a pre-approval to write, a realtor breathing down your phone about a Saturday showing, and an inbox with eleven follow-ups you swore you’d answer yesterday. You open ChatGPT to draft one email, then freeze. Is it even safe to paste this in? Nobody ever told you. So you close the tab and write the email by hand, slower, again.
That hesitation is the enemy here. Not AI. The not-knowing.
Here’s the verdict first: you can have three AI tools running by Friday, safely, for under $70 a month. The thing standing between you and that is one rule nobody handed you in plain English. I’ll hand it to you now.
One disclaimer up front: I’m not a compliance officer and this isn’t legal advice. Run any AI workflow that touches borrower data past your compliance team before you scale it.
The one rule that decides everything
Free tiers of ChatGPT, Claude, and Gemini may train on or keep what you type, depending on the product and that day’s privacy policy. Borrower names, socials, addresses, and income figures do not belong in a free public chatbot. The fix is not “don’t use AI.” The fix is the right-tier rule: anything client-specific goes in a paid tier built for it (Teams, Enterprise, or an API with zero retention), and free tools handle generic drafting and research. That one distinction is the whole game.
Hold onto the right-tier rule. Every tool below is just an application of it.
Tool 1: Claude, for writing
Claude (the latest Opus model) is the best of the three at sounding like a person. Email drafting, plain-English rewrites, long-document summaries, tone control. That’s its lane.
| Plan | Retention | Notes |
|---|---|---|
| Free / Pro | Inputs may be used per current terms | Drafting only. No client data. |
| Teams | Zero training on inputs | Safer for internal workflows |
| API (zero-retention) | Enterprise-grade | What custom HTS builds run on |
The use case: the borrower follow-up rewrite. Two stiff paragraphs become a warm, tight email in under a minute.
Model: Claude (the latest Opus model)
Prompt: "Rewrite this email in a warmer, more direct tone.
Keep it under 120 words. End with a single clear ask.
Don't change any factual details."
What never goes in the free tier: borrower names, SSNs, dates of birth, account numbers, pay stub or W-2 figures, bank statements, full email threads with PII. Strip the PII out, get the rewrite, paste the names back in. That is the workflow.
Draft in the AI, edit in your CRM. That’s faster than drafting in your CRM. Treat the AI like a writer, not a database.
Tool 2: Perplexity, for market intel and prospect research
Perplexity is a search engine that hands you a synthesized answer with citations instead of ten blue links. For any question that needs a real answer and not a list, it beats Google on speed.
The use case: the Monday-morning market update.
"What were the major US mortgage rate movements last week?
Cite Mortgage News Daily and the MBA Weekly Applications Survey."
You get a sourced answer in about 90 seconds. Drop the highlights into a market-update email, add your own read on it, and send to your top 30 past clients and realtor partners before your coffee’s cold.
What never goes in it: anything from a borrower file, your internal pipeline, anything under your shop’s NDA. Perplexity has the same posture as Google. Smarter results, same fence line. That’s the right-tier rule again: a public search tool is a free tier, so it gets public questions only.
Tool 3: An AI notetaker, and why this is the one that pays for itself
This is the tool most LOs skip, and it’s the one I’d install first. Here’s why.
A notetaker joins your Zoom, Teams, or Google Meet call, transcribes it, and hands you a summary plus action items. Fathom, Otter, and Fireflies do roughly the same job. Pick the one that fits your stack and your shop’s data agreement.
The use case: the realtor partner lunch. You meet an agent, she mentions she’s listing three homes next month and her buyers keep getting outbid. Without a notetaker, that’s gone by Friday. With one, you walk out with a four-bullet summary and action items. You paste them into Jungo, set a two-week follow-up, done in 90 seconds. The version of you without the notetaker spends 15 minutes on it, and usually doesn’t do it at all.
That last part is the whole point. The intel that actually grows your book lives in conversations. If the conversation isn’t captured, your CRM can’t remind you of something it never heard.
The rules here are about consent, not retention: never record without telling people (the notetaker should announce itself, or you should), don’t auto-share full borrower-call transcripts outside your shop, and keep internal pricing talk internal. Summaries you can share. Raw transcripts you don’t.
Most of what an LO knows about a referral partner lives in the conversation, not the file. Capture the conversation or the file stays empty.
Put a real number on it
Meet Carlos. He’s a solo loan officer in Arlington, closes about four loans a month, runs everything himself. Carlos is an illustrative composite, not a real client, but you’ll recognize him.
Two leaks cost Carlos money every week, and neither shows up on a report.
The first is drafting. He hand-writes borrower follow-ups and partner emails. Call it five hours a week he could halve with a writing tool. At a conservative $75 an hour for his time, that’s roughly $187 a week, near $9,700 a year, spent typing things a tool drafts in a minute.
The second is the forgotten follow-up. Say Carlos has two real realtor conversations a month where he means to circle back and forgets one. Over a year that’s 12 dropped threads. If even one in four would’ve turned into a referred deal, that’s 3 deals. At a $4,000 average commission, that’s $12,000 walking out the door.
Add it up: about $21,700 a year, most of it invisible. The tools that plug both leaks cost him under $70 a month, around $840 a year.
Your numbers will be different. Your hourly value, your close rate, your average commission, all of it moves the math. The point isn’t the exact figure. It’s that the cost of “I’ll get to AI later” is a real number, and it’s bigger than the subscription.
Skip this if
This isn’t for you if you’re not closing yet, or if your shop already runs an enterprise AI setup with these workflows wired in. If you’ve got no consistent process to speed up, a tool just helps you do the wrong thing faster. Get the process repeatable first. Then automate it.
The 30-minute setup
Claude Pro for drafting and summaries, PII stays out. Perplexity Pro for Monday updates and partner research. An AI notetaker on every meeting that matters.
Under $70 a month. About 30 minutes to set up. Time saved per week, conservatively, in HTS engagements I’ve watched: 3 to 5 hours. Every one of those three choices is just the right-tier rule applied to a different job.
What this is not
This is the starter stack. Three off-the-shelf tools any LO can install today with no IT support. It is not a custom workflow, and it is not an AI agent running your business.
If you want AI plugged into Jungo, your email, your phone system, and your loan pipeline end to end, that’s the next conversation. We build that inside the HTS Operating System tier. We did a version of it inside a recent HTS engagement, and the piece they used most wasn’t the impressive part. It was a 90-second daily summary.
Ready to go further?
Get the starter stack in place. If you then want a real workflow built around it, book a discovery call. It’s 30 minutes, and you’ll leave with a one-page plan whether or not you hire us.
Compliance note: This post discusses AI tools in the context of mortgage workflows. None of the recommendations here is a substitute for review by your shop’s compliance officer. Any workflow that processes borrower PII should be vetted before deployment. Pricing and plan features for third-party tools change frequently: verify current terms with the vendor.