Mortgage Lead Scoring: What Belongs in the Score

Mortgage lead scoring ranks opportunities by how likely each one is to become a funded loan. Done right, a loan officer opens the CRM and already knows where to start.

Done wrong, it produces a compliance problem that looks like arithmetic.

What Actually Belongs in a Mortgage Lead Scoring Model

Two categories of input, and telling them apart is most of the work. Lead scoring for loan officers only works when the inputs describe intent rather than the person.

Behavior. What the borrower did. Opened an email, clicked, replied, called back. Time on a rate page. A second visit to the site. These are choices the borrower made.

Loan facts. Purpose, loan amount, timeline, whether a rate lock is in play. Facts about the transaction rather than the person.

Now notice what is missing from both lists.

Age. Employment status. ZIP code. Income band. Those show up on nearly every lead scoring guide in this industry. They are attributes rather than signals of intent. They also carry the problem in the next section but one.

Scoring Conversion Is Not the Same as Scoring Eligibility

Most platforms give you one number and let you assume it covers both. It does not.

A borrower who will apply and then be declined scores high on intent and is worth almost nothing. One who qualifies easily but is a year from buying scores low and deserves a nurture sequence.

Two different predictions. If your score reflects one, know which. Otherwise you are ranking who answers the phone and calling it a pipeline.

The Fair Lending Problem Nobody Mentions

Here is the part every article on this subject leaves out.

Score leads on age, income, employment, or geography. Then give the low scorers fewer attempts, slower response, and less of a loan officer’s day. What you have built is a system that delivers a different level of service based on protected characteristics.

Intent does not enter into it. Disparate impact is measured on outcomes.

The score makes this harder to catch than an obvious policy would, because it reads as math. Nobody wrote down “call these borrowers less.” The model did it and the model came from a template.

What to do instead. Build the score on behavior and loan facts only. Keep demographic attributes out of the model entirely, including proxies like ZIP code. And be able to show, on request, exactly which inputs the score uses and what each one weighs.

This is Shape’s operating read rather than legal advice. Take your scoring inputs to your own compliance officer before you turn a model on. Revisit whenever the inputs change.

How to Build a Mortgage Lead Score Model That Works

Four steps. The second one is where almost everybody stops short.

One. Pick five to seven signals. Behavior and loan facts only. Past seven, you cannot tell which is doing the work.

Two. Score your last 30 funded loans retroactively. Run your proposed model backward against loans that actually closed. This is the step that separates a model from a guess.

Three. Drop what did not cluster. A signal with no pattern across your funded loans is noise wearing a point value.

Four. Re-check quarterly. What predicted a funded loan in a purchase market predicts something different when refinance volume returns.

Steps two through four are why AI CRM platforms for mortgage advisers handle scoring differently than a general sales tool. A model retraining on your own outcomes does step four on its own.

Anyone can assign point values. That is the easy part, and it is where every mortgage lead scoring template stops.

Scoring is one of three signals that should drive your queue, alongside time since inquiry and loan purpose. How the three work together is in mortgage lead management.

Scoring is one piece of a larger system; see where automation fits the loan lifecycle.

Frequently Asked Questions