Brokerage ops

Have Claude build two HubSpot agents that turn broker loan-request emails into scored Leads for a commercial lender

Consultant John Hulina of CRM Assist describes using Claude to build a loan-intake system inside HubSpot for a commercial lender (the lender is not named) whose four loan officers receive more than 1,000 inbound emails a month. Claude first analyzed the HubSpot portal: email volume per inbox, the Lead pipeline and its stages, owner IDs, and the lending criteria. Claude then wrote two HubSpot custom agents: a Submission Check agent decides whether an email is a loan request and whether the request is a duplicate, and creates a Lead named by property and city, assigned to the loan officer who received the email; a Loan Intake agent extracts the loan details and scores the request against the lending criteria as Meets, Needs review, or Does not meet. In simulated tests on 10 real broker contacts, 9 were correct on the first pass; the miss matched a new request to the wrong existing deal, so Claude rewrote the rule so a new property from the same broker counts as a new deal. Testing also caught CC'd contacts, auto-replies and calendar accepts triggering the agent, so Claude rebuilt the trigger with sender, recipient and subject filters. Meets creates a follow-up task, Needs review creates a review task, and Does not meet marks the Lead Disqualified and creates a task for a person to review and send the decline. Claude also set per-agent run limits and a monthly HubSpot credit cap and built five pilot reports. Hulina says the project reached launch-ready in days.

At a glance

Tools used
ClaudeHubSpot
Category
Brokerage ops
Audience
investor, broker
Difficulty
Builder
Source
How I Used Claude to Build an AI Loan-Intake System Inside HubSpot
John Hulina (CRM Assist) · Blog · Oct 17, 2026
First step
Write your lending criteria as a short checklist (loan size range, property types, markets, maximum loan-to-value) so the intake agent has a written rule to score each broker request against.

Compliance note: Keep a person sending every decline and every reply to a broker, store borrower and sponsor financials only inside your CRM's permission controls, and cap AI credit spending before go-live so a misfiring trigger cannot run up the bill.

Suggested prompt

Written by this directory, inspired by the source. Replace the [bracketed] parts.

Here is an email a mortgage broker sent to our loan officer: [paste email]. First, decide whether the email is a new loan request, a follow-up on an existing request, or not a loan request (auto-reply, calendar accept, CC). A new property from the same broker counts as a new request. If the email is a new loan request, extract the property address, city, property type, loan amount, purpose, and any loan-to-value or debt-service numbers the broker gave. Score the request against our criteria as Meets, Needs review, or Does not meet, and quote the criterion behind the score: [paste lending criteria]. Do not draft a decline; list what the loan officer should check before replying.

Share on X Read the source

← All use cases