Week 2: Building Out an AI Software for LOs

My name is Christian and I am building an AI database agent for loan officers. I run a digital marketing company that serves over 200 mortgage professionals, and this is what we are working on next.

Last week we built the foundation. This week we built the brain.

Introducing the Snapshot Engine

The biggest thing we built this week is something we are calling the Snapshot Engine. It looks at every single person in a loan officer's database and figures out the best reason to reach out to them.

Is this person ready to buy a home? Do they have equity they could be using? Are they still paying for mortgage insurance they no longer need? The software figures all of that out automatically, scores each contact across six different opportunity types, and picks the best one.

Those six opportunity types are purchase and move-up, cash-out refinance, rate refinance, PMI removal, annual mortgage review, and a neutral check-in as a fallback. Every contact gets scored and ranked before a single message is ever written.

Then We Connected It to AI

Once the Snapshot Engine scores a contact, the system hands that information off to Claude, and Claude writes the outreach message.

Not a template blast. A real, personal text or email that sounds like the loan officer actually knows the person. Because it uses their real information to craft it. No self-introductions. No "I came across your information in my database." Just a warm, natural message that references something specific about that contact.

We also built a compliance checker that runs on every message before it goes anywhere. It flags rate mentions, guarantee language, and missing opt-out text. Zero violations on our test run.

The Bumps We Hit Along the Way

It was not all smooth sailing. Files ended up in the wrong folders. The AI kept trying to introduce itself to people it already knew. Early messages sounded like generic marketing blasts instead of real human outreach.

We fixed all of it. We updated the prompt to remove introductions, strip out marketing language, and reference the relationship naturally based on the contact's data. Once we added real contact data to a test contact and reran the system, the opportunity detection changed, the tone changed, and the message felt genuinely personal.

We launched it on live test contacts connected to a real GHL account. It worked.

What This Means for Loan Officers

Every single person in a loan officer's database gets a personal, thoughtful follow up without the loan officer having to do anything. The leads that have been sitting dead in a CRM for years finally get worked.

And the message does not sound like it came from a robot. It sounds like it came from someone who pays attention.

Where We Are Heading Next

Here is something I want to be clear about. I do not think this should be a spray and pray AI agent that blasts everyone in a database at any time without any control.

What I actually want to build is a dashboard where a loan officer logs in and immediately sees their opportunities laid out in front of them. 20 refinance opportunities. 10 purchase opportunities. 5 PMI removals. Real numbers based on real data from their actual contacts.

From there, the loan officer chooses which opportunity they want to activate. They press a button, and the AI personalizes and sends messages to every contact in that group at scale. The loan officer stays in control. The AI does the work.

That is the product we are building toward.

Stay Tuned for Next Week

Next week we are building the two-way conversation manager. When someone replies to an outreach message, the AI reads their response, figures out their intent, and keeps the conversation going. If they are ready to talk, it books the call automatically and hands off to the loan officer at exactly the right moment.

See you next week.