Most AI announcements in mortgage are about capability: it can text, it can call, it can write. The upgrade LoanOfficer.ai shipped this week is about something loan officers have been asking for much more loudly — control. The assistant's settings were rebuilt from a single long form into five purpose-built sections, a sandbox now sits next to them so you can test a change before a borrower ever sees it, and compliance guardrails were pulled into one screen where six of them are permanently on and cannot be turned off.
The second half of the release goes the other direction. The assistant is now embedded in the platform's top menu, on every page, reading live data from your own CRM. You can ask it to summarize your pipeline, tell you who to follow up with today, draft a rate-drop email, or show this week's appointments — without navigating away from whatever you were doing.
This article walks through the whole release the way you'd actually set it up: section by section, then the guardrails, then the platform-wide assistant. If you'd rather watch it, the full walkthrough is in the <a href="/academy/ai/ai-assistant-upgrade">LoanOfficer.ai Academy</a>, and there's a Spanish version at <a href="/es/academia/ia/mejora-del-asistente-de-ia">Academia</a>.
Where the new settings live
Everything in this release starts in the same place: open <strong>User Settings</strong>, then select <strong>AI Assistant</strong>. What used to be one scrolling form is now organized into five labeled sections — Assistant, Scheduling, Follow-Up, Coach, and Training — with sandbox mode on the right.
The reorganization isn't cosmetic. Each section answers one question about your assistant: who it is, when it can book, how it follows up, how it coaches you, and what it knows about your business. When those questions were mixed together in one form, most loan officers configured the first third of it and never came back. Split apart, you can finish one section in two minutes, test it, and move on.
A reasonable order for a first pass: Assistant, then Scheduling, then Follow-Up, then Training, then Coach. Assistant and Scheduling change borrower-facing behavior immediately, so they earn the first ten minutes.
Assistant: identity, phone number, and approved links
The Assistant section holds the details that appear in every conversation: your assistant's name, the phone number it communicates from, and the important links it's allowed to share with a borrower.
The link list deserves more attention than it usually gets. An assistant that can only send addresses you approved cannot invent a rate sheet, an outdated application page, or a competitor's calculator in the middle of a conversation. Populate it with the three or four URLs you actually want in a borrower's hands — your application, your calendar, your reviews page, and, if you use it, your Property Pulse report link.
On naming: pick something that doesn't pretend to be a licensed originator. Borrowers respond fine to an assistant that is clearly an assistant, and the alternative creates disclosure problems you don't need.
Scheduling: availability, minimum notice, and days off
Scheduling is new in this release, and it's the section most likely to change your calendar this week. Instead of one blanket availability window, you set availability day by day — so a Tuesday stacked with agent meetings looks different from an open Thursday.
Two controls do the heavy lifting. A <strong>minimum notice window</strong> prevents the AI from booking a borrower into the next fifteen minutes, which is the single most common complaint about AI appointment setting. And <strong>time-off blocks</strong> stop it from booking while you're out of the office, including the week you actually take off.
Set the notice window to how you really work. If you need thirty minutes to pull a file and review credit before a call, thirty minutes is the number. A too-short window produces appointments you show up to unprepared, which costs more trust than a slightly later slot ever would.
Follow-Up: what happens when a conversation goes quiet
The Follow-Up section controls sequences: how long the assistant waits after a conversation goes cold, and what it sends next. This is where most of the revenue in the release sits, because most lost leads in mortgage aren't lost to a competitor — they simply stopped replying and nobody followed up again.
A pattern that works: a short first wait, then progressively longer ones. Fast at the start and patient afterward reads as human. A fixed interval — every three days forever — reads as software, and it's the fastest way to get a borrower to opt out.
Pair the timing with content that changes. A second touch that repeats the first touch teaches the borrower to ignore you. A second touch that adds something new — a rate observation, a document they'll need anyway, an answer to the question they didn't finish asking — gets replies.
Coach and Training: your goals, and your voice
Coach configures the AI production coach. You give it your business information, your current focus, and your goals, and it guides you on your production against them. It's the one section that isn't about talking to borrowers — it's about what you do next.
Training holds the context that lets the assistant communicate like you: how you explain programs, the phrases you use, and the things you never say. Loan officers consistently underinvest here and then complain the AI sounds generic. Ten minutes of specifics — how you explain a buydown, how you handle the rate question on a first call — changes the output more than any other setting on the screen.
Both sections compound with the top-menu assistant described below, because the same context feeds it. A well-trained assistant gives better pipeline answers, not just better borrower texts.
Sandbox mode: test it before a borrower does
Sandbox mode sits to the right of the settings and lets you run a conversation exactly as a borrower would experience it. That's the point of the whole redesign: change a follow-up rule or add a training note, then run it in the sandbox before it reaches anyone real.
Test the uncomfortable messages, not the easy ones. A rate question, a credit question, an annoyed reply, a request to stop texting. Those are where a training gap shows, and they're the exact conversations that would otherwise surface first in front of a live borrower.
If you have a compliance stakeholder, the sandbox is also how you get sign-off. Running the scripted scenarios your compliance team cares about, in a place where nothing sends, is far easier than asking them to trust a settings screen.
The guardrails screen: six protections that can't be turned off
Compliance controls used to be spread out. In this release they're consolidated onto one guardrails screen with a deliberate split between what you control and what you can't.
Six protections are always on and cannot be disabled — including RESPA and TILA compliance handling, honoring opt-out requests, and redacting Social Security numbers and email addresses. That's an intentional product decision: the most common source of AI compliance exposure isn't a bad model, it's a well-meaning user switching a safeguard off to make the assistant sound more helpful.
Below the permanent six, you control the policy layer: whether the assistant may reference average rates, the required rate disclaimer language, the auto-handoff keywords that escalate a conversation to a human immediately, and the restricted topics it won't discuss on your behalf. Set the handoff keywords with your own liability in mind — words like attorney, bankruptcy, dispute, and complaint belong on almost every list.
Ask Anything: an assistant on every page, reading live data
The assistant is now embedded in the platform's top menu with substantially more capability behind it. Ask it to summarize your pipeline. Ask who you should follow up with today. Ask it to draft a rate-drop email, or to show this week's appointments. Or type your own question.
The important detail is the data source: it pulls live information from the platform, so answers reflect your actual pipeline rather than a generic model response. And you get them without leaving the page you're working on, which is what turns it from a demo into a habit.
Practically, this replaces a category of small tasks — building a call list, checking what's on the calendar, drafting a one-off borrower email — that used to require three clicks and a report. The morning version of this is a single question: who should I call today?
A 30-minute setup plan
If you want a concrete plan: spend five minutes in Assistant confirming your name, number, and approved links. Spend ten in Scheduling setting real day-by-day availability, a minimum notice window that matches your prep time, and your known days off. Spend five in Follow-Up setting a short first wait and a longer second.
Then spend five minutes in Training writing down how you explain your two most common programs and one thing you never say. Spend the last five in the sandbox running three hard conversations. Save Coach for the end of the week, when you can enter real goal numbers rather than guesses.
Everything here is live now for every LoanOfficer.ai account at no extra cost. If you're not on the platform yet, you can <a href="/trial">try it for 14 days for $1</a>, or <a href="/demo">book a demo</a> and we'll walk the guardrails screen with you.
Where do I find the new AI assistant settings?
Open User Settings and select AI Assistant. The screen is now split into Assistant, Scheduling, Follow-Up, Coach, and Training, with sandbox mode on the right.
Scheduling is a new section. You set availability day by day, add a minimum notice window so appointments can't be booked too close to now, and block out days you're out of the office.
Which guardrails can't be switched off?
Six protections are always on, including RESPA and TILA compliance handling, honoring opt-out requests, and redacting Social Security numbers and email addresses.
What compliance settings do I still control?
References to average rates, the required rate disclaimer, the auto-handoff keywords that escalate a conversation to a human, and the restricted topics the assistant won't discuss.
How do I test a change before borrowers see it?
Use sandbox mode next to the settings. It replays the conversation as a borrower would experience it, so you can validate a training note or a follow-up rule first.
What can the top-menu assistant do?
It answers using live data from your platform — summarize your pipeline, tell you who to follow up with today, draft a rate-drop email, or show this week's appointments, without leaving the page.
No. It's included for every LoanOfficer.ai account. New customers can start a 14-day trial for $1, and plans are $197/mo after that.
The through-line of this release is that AI in mortgage becomes useful at exactly the point it becomes controllable — sectioned settings you can reason about, a sandbox to test in, protections that don't have an off switch, and then an assistant everywhere because you finally trust it. Watch the full walkthrough in the <a href="/academy/ai/ai-assistant-upgrade">Academy</a>, read the <a href="/press/loanofficer-ai-releases-major-ai-assistant-upgrade">release announcement</a>, or <a href="/trial">start for $1</a> and configure it yourself this afternoon.