Intake to Decision Flow for Private Lenders: Resetting an AI Underwriting Build Around the CRM

A lender's AI underwriting can work in a demo and still fail in the business, because nothing connects it to the place where deals actually arrive. That was the position of a private lender whose AI assets were scattered across personal accounts and whose intake-to-decision flow never touched the CRM.

The client is a leadership team at a private real estate lender. This article covers the reset plan we agreed on a working call, before anything was rebuilt. If you run lending for investors, our page for mortgage loan officers and private lenders shows the wider work.

The situation

The lender had been building AI underwriting with us for some time. Skills and prompts had ended up spread across the personal accounts of different people. The underwriting automation itself needed a reset, and the flow from a new deal coming in to a decision going out ran separately from the CRM where the team tracks loans.

The call brought together the lender's leadership team and ended with agreed design decisions rather than a finished system.

What was broken

Four gaps came up, and they were more connected than they looked:

• AI assets lived in personal accounts, so no one place held the lender's prompts, skills and work.

• The underwriting automation needed to be reset.

• The intake-to-decision flow was not wired to the CRM, so a new deal in the CRM did not start any AI analysis.

• There was no single record of calls, tasks and vendor discussions.

What we decided

Each gap got a specific decision.

• One admin account. All AI assets get centralized under an admin Claude account owned by the company, not by an individual.

• A CRM trigger. We rebuild the intake-to-decision flow so that a CRM event starts the analysis. The proof is a test, started by a CRM event, that processes one live deal from intake to decision, which we committed to build and demonstrate by the next Monday follow-up.

• A map of AI and human steps. Using the lender's draft loan origination flow, we map each Claude automation task onto it and mark where AI handles a step and where a person does. The lender's side was to send the origination draft and its revised underwriting guidelines as they became available.

• A credit box and an exceptions tracker. A new senior hire is to define the credit box and underwriting standards, and to stand up a tracker for exceptions to them.

• Recorded calls as the source of truth. The team will turn on call recording in its meeting software and run a summary routine, so decisions, tasks and vendor discussions have one searchable record.

What changed

This is a plan-stage result, so no outcome was reported. What changed was ownership and sequence. The AI work now has an owner, which is the company. The test that decides whether the flow works is defined in advance as one real deal, end to end, started by the CRM.

The credit box matters most for the long run. An AI underwriter can only apply standards that someone has written down. Assigning that job to a named role, with an exceptions tracker beside it, gives the automation something firm to apply and gives the team a record of the exceptions.

Sequence your own reset

If your AI work is spread across personal logins, fix ownership first. Move prompts, skills and notes into one company-controlled account before you build anything new.

Then decide what starts the process. Pick the CRM event that means a deal has arrived, and prove the flow on one live deal before you scale it. Write down which origination steps AI handles and which stay with a person. Finally, record your calls. A summary routine on recorded calls gives you a dependable history of what was decided and who owns the next step.

Frequently asked questions

Why centralize AI assets under an admin account?

Assets in personal accounts leave when the person does and cannot be reviewed by the team. An admin account owned by the company keeps the prompts, skills and work in one place, so the lender controls its own underwriting automation.

What does a CRM-triggered AI underwriting test look like?

A deal event in the CRM starts the analysis, and the result flows through to a decision step. The test planned here runs one live deal end to end, which shows whether intake, analysis and decision actually connect before more deals are added.

What is a credit box and why does AI need one?

A credit box is the written set of standards for the loans a lender will make. An AI underwriter can only apply standards that exist on paper, so defining the box, and tracking exceptions to it, comes before trusting the automation with decisions.

Why record calls for an AI project?

Recorded calls with a summary routine create a single source of truth for knowledge, tasks and vendor discussions. Without one, decisions made on calls get lost, and the AI build ends up following instructions nobody can find.

If your AI tools are not connected to the place your deals arrive, book a 30-minute AI follow-up and reporting audit. We map your intake-to-decision path and show where the first trigger belongs.