A mortgage AI assistant becomes useful only after it knows three things: how your business operates, where a borrower should go next, and when it should step back. Those decisions are not one prompt. They are a connected setup spanning identity, scheduling, follow-up, products, disclosures, escalation, and testing.
The complete LoanOfficer.ai setup walkthrough brings those controls into one sequence. You start with a name and approved links, define when meetings can be booked, decide how quiet conversations are restarted, teach the assistant your rules and specialized products, test borrower questions in a sandbox, and finish by placing the assistant on your website.
This guide turns that walkthrough into a practical implementation plan. Watch the full lesson in the <a href="/academy/ai/complete-ai-assistant-setup-walkthrough">LoanOfficer.ai Academy</a>, or use the <a href="/es/academia/ia/configuracion-completa-del-asistente-de-ia">Spanish training</a> with a bilingual team.
1. Give the assistant a clear identity
Open <strong>User Settings</strong> and select <strong>AI Assistant</strong>. Give the assistant a name that distinguishes it from you while it schedules appointments and follows up with leads. The goal is clarity: the borrower should understand that an assistant is helping while the relationship still belongs to the licensed mortgage professional.
You can also assign the assistant a phone number. That makes it possible to exchange text messages with the assistant and receive updates on your phone, while the platform keeps assistant chats available under its dedicated chat icon.
If you use delayed replies, choose the interval deliberately. A delay of up to three minutes changes the pace without changing the answer. Test the result in a real conversational rhythm; the longest setting is not automatically the most natural for a new inbound lead.
2. Build an approved link library
The assistant needs an authoritative destination for every action it recommends. Start with the loan-application URL. Then add only the resources you want it to share, such as your calendar, YouTube channel, social profiles, reviews, or a focused educational page.
Important Links can also become smart tracking links. That gives you a click count for the exact destination the assistant shared. The distinction matters: a link sent is an activity; a link opened is an intent signal.
Name links by purpose rather than by internal shorthand. ‘Start a loan application’ is easier for the assistant to match correctly than ‘LOS app v3.’ Review the list whenever a website or application URL changes so an otherwise good conversation never ends at an outdated page.
3. Configure scheduling around the calendar you really keep
Set the appointment types and meeting lengths you actually offer, then define availability by day. Add a minimum-notice window that gives you enough time to review context before a call. The walkthrough uses 12 hours as an example for avoiding unexpected last-minute appointments; your value should reflect your process.
Time Off is as important as weekly availability. Add vacations, conferences, and other unavailable dates before a campaign or assistant interaction sends someone to the calendar. The assistant can discuss an appointment at any hour, but it should offer only times inside the window you approved.
The same section provides an inline calendar, pop-up, or badge for a website. After embedding it, complete a booking from a phone and desktop. Confirm meeting types, time zone, minimum notice, blocked dates, and the contact-information step.
4. Design follow-up as a sequence, not a pile of reminders
When a lead stops replying, the AI Assistant can wait for a chosen period and write a follow-up from the conversation context. A second step can wait longer, and later steps can continue at the intervals that fit your process. The assistant writes the message; you control the cadence.
Review every step on one timeline. One day, then two days, then three days, then seven days can be reasonable when each message has a purpose. The same number of touches compressed into a short window can feel excessive even if each individual setting looked harmless.
Use the re-enrollment buffer to stop a recently completed sequence from immediately restarting. A good follow-up system preserves persistence without treating every pause as permission for another burst of messages.
5. Decide whether the Production Coach belongs in your workflow
The AI Production Coach is optional and separate from borrower follow-up. If you enable it, describe your business, experience, production goals, current challenges, resources, and team. That context lets the coach respond to the way your operation is actually staffed rather than offering generic productivity advice.
Be specific about constraints. ‘I need more leads’ is less useful than explaining your lead sources, weekly call time, current pipeline, and where production stalls. Also identify the people who can help so recommendations can account for delegation instead of assuming the loan officer does everything.
Leave the coach off if you are not ready to maintain the context. Disabling it does not change how the assistant handles live borrower conversations.
6. Train behavior before adding more information
Training should begin with rules. Tell the assistant how to end an interaction, what to do when someone asks to get started, when a disclosure is required, and which situations should transfer to a person. Direct if/then instructions are easier to test than broad requests to ‘sound professional.’
Then add identity and company context: your experience, business details, team, and the facts the assistant needs to represent you accurately. Keep this material current. A biography from two companies ago is not harmless background when the assistant uses it in a live answer.
Use Offerings for specialized products that need additional explanation. The assistant already has broad mortgage-program knowledge; the highest-value additions are the programs, geographic conditions, and company-specific details it could not know without you.
7. Use Sandbox Mode as an acceptance test
Sandbox Mode is where configuration becomes evidence. Ask a product question, a rate question, a scheduling question, and a question that should trigger escalation. The walkthrough demonstrates a no-income-documentation question that leads to useful alternatives and a qualifying question rather than a one-line answer.
For rate questions, confirm that the answer uses your Rate Watch configuration and includes the disclosure you require. For product questions, confirm the assistant distinguishes standard knowledge from the specialized information you entered. For scheduling, confirm it respects notice and time off.
Write down a small test set and rerun it after meaningful changes. A sandbox is most valuable when the same difficult scenarios are checked consistently, not when it is used once for an easy greeting.
8. Treat guardrails as operating policy
The guardrails area includes built-in handling for RESPA and TILA compliance, content moderation, and consumer opt-out requests such as STOP. It also supports configurable rate disclosures and automatic escalation based on selected words or topics.
Technology does not replace your company's compliance review. Use the available controls to encode approved behavior, test the resulting conversations, and revise the rules when your policies or products change. The important operational question is not whether a toggle exists; it is whether the complete response behaves as expected.
Escalation deserves its own test. Ask about a topic that should transfer, confirm the assistant stops trying to resolve it, and verify that the appropriate person can see and continue the conversation.
9. Put the configured assistant on your website
Once the assistant behaves correctly, the Widget section lets you publish it on your website. Customize its icon, logo, colors, and title, then copy the provided code into the intended page. The visitor can open the widget and start a conversation without first navigating to a separate form.
The website version should inherit the same approved links, business context, scheduling rules, and guardrails you tested. Still, the surrounding page changes the questions people ask. Test the widget on the exact page where it will appear, not only inside settings.
Check mobile placement carefully. The launcher should remain visible without covering navigation, consent text, or another primary action. Open it, ask a representative question, follow a shared link, and complete a booking before sending traffic to the page.
A practical 45-minute launch checklist
Spend five minutes on name, phone access, and delayed-reply pacing. Spend five minutes validating the application URL and approved links. Use ten minutes for appointment types, availability, notice, and time off. Use another ten for follow-up timing and the re-enrollment buffer.
Reserve ten minutes for behavior rules, identity, company context, and specialized offerings. Finish with five minutes in Sandbox Mode testing rates, product eligibility, scheduling, opt-out handling, and escalation. Configure the website widget only after those tests pass.
The sequence matters. Publishing a beautifully branded widget before the assistant knows your calendar, application link, or escalation rules creates a polished entry point into an unfinished process. Configure first, test second, embed third.
Where are the AI Assistant settings?
Open User Settings and select the AI Assistant tab. The configuration areas include Assistant, Scheduling, Follow-Up, Coach, Training, Sandbox Mode, Guardrails, and Widget.
Can the assistant have its own phone number?
Yes. The walkthrough shows an optional phone number that lets you text the assistant and receive updates on your cell phone.
How long can a reply be delayed?
Delayed Reply can be set for up to three minutes. It changes response timing to create more natural pacing.
Can the AI book while I am on vacation?
Not when the dates are entered under Time Off. The assistant uses your configured availability and blocked dates when offering appointments.
Does AI follow-up use fixed scripts?
The walkthrough shows the assistant deciding and writing the follow-up from the conversation context while you set the timing and number of steps.
Add specialized or company-specific products that need context beyond standard mortgage-program knowledge.
How should I test the assistant?
Use Sandbox Mode for product, rate, scheduling, opt-out, and escalation scenarios. Check the answer, disclosure, qualifying question, approved link, and handoff.
Can I embed the AI Assistant on my website?
Yes. Customize the widget's icon, logo, colors, and title, then copy the supplied embed code into your site.
A reliable mortgage AI assistant is configured as a system, not decorated as a chatbot. Give it the right identity and links, let it see the calendar you actually keep, define follow-up boundaries, teach it your rules and specialized products, and make it prove those decisions in the sandbox. Then place it on your website. Watch the <a href="/academy/ai/complete-ai-assistant-setup-walkthrough">complete English walkthrough</a>, share the <a href="/es/academia/ia/configuracion-completa-del-asistente-de-ia">Spanish lesson</a> with your team, or <a href="/trial">Start 14-Day Trial — $1</a>.