Mortgage Loan Officers · Chapter 02 of 10
The lender's systems, and what you run yourself
You do not choose your LOS. You rarely choose your pricing engine, your credit vendor, or the AUS. This chapter maps every system a loan file passes through, marks who owns each one, and draws the line where your own tools, and Claude, are allowed to sit.
Why this chapter
Almost every technology decision in your day was made above you: your company (the bank, credit union, or broker owner) licenses the loan origination system, the point of sale, the pricing engine, the credit vendor, and the AUS connection. You operate inside those systems, you rarely administer them, and you never own the vendor contract. What you can add on your own is a much shorter list, and borrower data has a hard boundary around it under federal law. This chapter draws that map before Chapter 08 sets Claude up inside it.
The two layers: company systems and your own tools
Every system a mortgage desk touches falls into one of two ownership buckets. The company licenses and administers the first bucket; you view, trigger, or draft inside it, but you do not choose the vendor. The second bucket is the small set of tools an individual LO commonly buys for themselves, on top of whatever the company already provides.
| Category | Named systems | Who owns it | What you touch | What you never see |
|---|---|---|---|---|
| Loan origination system (LOS) | ICE Encompass, nCino Mortgage Suite, MeridianLink Mortgage, Calyx Point/Path, LendingPad, Arive, Byte | Company selects and contracts it at the institution level verified [1] [3] [5] [6] | Enter and manage the loan file, order services, track conditions, communicate with the borrower where wired to a POS | The vendor contract, admin configuration, integration keys, and pricing terms the company negotiated |
| Point of sale (POS) | Blend, nCino Mortgage Suite POS (formerly SimpleNexus), Floify, BeSmartee, Maxwell | Company selects it; the borrower interacts with it directly verified [7] [8] [9] | Monitor the borrower's application progress, review uploaded documents, follow up on gaps | Back-end configuration and the borrower-data pipeline into the LOS |
| CRM | Total Expert, Surefire, Jungo, Bonzo, Velocify, BNTouch, Shape, HubSpot | Enterprise-scale CRMs are typically company-administered; lighter add-ons are more often LO-run, though no vendor page states a formal self-purchase policy report [9] [10] [11] [12] | Pipeline stages, campaigns, texting and email sequences, contact records | The company's compliance configuration of the enterprise CRM and its texting-registration setup |
| Pricing engine (PPE) | Optimal Blue, LoanSifter, Polly, Lender Price, MeridianLink PPE | Company selects it and sets margins and overlays report [13] | View live rates, request a lock | The margin table, overlay logic, and investor relationships behind the price you see |
| Credit / tri-merge | Xactus, CoreLogic/Cotality Credco | Company selects the credit reporting agency and pays per-pull fees claim [14] | Pull credit or a pre-qualification soft pull with a documented permissible purpose, review the resulting report | The agency's back-end contract and per-pull cost structure |
| Automated underwriting system (AUS) | Fannie Mae Desktop Underwriter (DU), Freddie Mac Loan Product Advisor (LPA) | The company's LOS is configured with AUS credentials; the GSEs set the rules report [15] | Submit the file and read the findings | The GSE's internal risk model and rule engine |
| Disclosure / document prep | DocMagic, Mortgage Cadence | Company selects and integrates the vendor into the LOS claim [17] | Confirm the generated numbers match what was quoted; rarely touch document generation directly | The audit-engine rule set that checks TRID tolerance, QM/ATR, and RESPA before you ever see the output |
| E-closing / RON | Snapdocs, Notarize (rebranding to Proof, transition completing 2026-09-14) | Company, or the title/settlement partner, selects the platform claim [18] | Coordinate the closing date and confirm the borrower is scheduled | The notary network and title-branch integration behind the scheduling |
| Income / employment / asset verification | The Work Number, Experian Verify, Truv, Argyle, Plaid | Company selects and pays for the vendor, often more than one; the borrower authorizes the pull claim [19] | Review the resulting verification report | The per-pull cost and the vendor's data-matching pipeline |
Read the ownership column literally. In every category above, the loan officer operates inside a system someone else selected and configured. The one partial exception is the CRM row, where lighter tools are commonly run by an individual LO, though no vendor states this as a formal company-side policy; that means your own company's answer, not the vendor's marketing page, decides what you may install. operator
The loan's path through the systems
A single file crosses most of the systems above in a fixed order. This table walks that order, names which system is doing the work at each stage, who is acting on it, and what you personally are checking rather than entering. The exact timing rules (how many days a disclosure has to go out, what resets a closing disclosure) belong to Chapter 05 and Chapter 09; this table stops at which system does the work.
| Stage | System | Who acts | What you check |
|---|---|---|---|
| Lead | A lead source (Zillow, LendingTree, Bankrate, a builder-affiliated captive lender, or Homebot for a past client) feeding the CRM report [20] | You, or a company ISA | The source is recorded, the consent method is on file, and the lead's cost or referral basis is understood before you spend time on it |
| Application | The POS (Blend, nCino, Floify, BeSmartee, or Maxwell) verified [7] [8] | The borrower, self-serve, with you monitoring | The application is complete, required documents are uploaded, and nothing is stalled on the borrower's end |
| Credit | Xactus or CoreLogic/Cotality Credco, called from the LOS claim [14] | You or a processor | A permissible purpose is documented for the pull, and the tri-merge result lines up with what the AUS will expect |
| Pricing and lock | The company's PPE (Optimal Blue, LoanSifter, Polly, Lender Price, MeridianLink PPE) report [13] | You view live pricing and request the lock | The lock confirmation matches the quoted terms and expiration date exactly |
| Disclosures | The LOS, generating documents through DocMagic or Mortgage Cadence claim [17] | The system generates automatically from LOS data | The generated numbers match what was quoted; timing rules covered in Chapter 05 and Chapter 09 are met |
| Processing | The LOS conditioning workflow; ICE's stand-alone automated conditioning engine (launched 2026-04-01) auto-generates conditions and matches them to documents claim [2] | A processor, with you following up on borrower-side items | Which conditions are open, whose responsibility each one is, and whether anything is stuck on the borrower |
| Underwriting | DU or LPA, plus a human underwriter report [15] [16] | The underwriter renders the decision | The AUS findings and any manual conditions the underwriter adds beyond the automated result |
| Closing | Snapdocs or Notarize/Proof, coordinated with title claim [18] | Title or settlement schedules the signing; you confirm the date with the borrower | The closing disclosure figures are accurate and the borrower is prepared for the signing appointment |
ICE's own public position on Encompass is worth reading closely because it defines the ceiling for every AI tool touching this path, including Claude: the company states its AI features assist document conditioning and analysis but will not make final decisions on approvals, pricing, or disclosures. claim [2] That is the same line this guide draws for Claude, independently of what any vendor claims about its own tools.
What you may take out
Borrower information (income, assets, Social Security number, account numbers, and most of what sits in the LOS and POS) is nonpublic personal information, NPI, under the Gramm-Leach-Bliley Act. The FTC's Safeguards Rule requires financial institutions, explicitly including mortgage brokers, to maintain a written information security program protecting that information. verified [22] A 2023 amendment sharpened the consequence: a breach exposing unencrypted customer information for 500 or more consumers must be reported to the FTC within 30 days of discovery. verified [23]
The practical rule that follows: NPI stays inside company systems. It does not go into a personal AI account, a personal spreadsheet, or a personal note-taking app, unless your company's written policy explicitly says otherwise.
What personal AI, including Claude, can reasonably work on without touching NPI: the desk file you built in Chapter 01 (your NMLS ID, programs offered, current limits), the program matrix (rate and product facts that are not tied to one borrower), your own call notes and to-do lists, and de-identified summaries you write yourself with names and numbers stripped out. This is a reasoned boundary based on what GLBA covers, not a statement from any regulator about personal AI use specifically. inference No regulator statement was found addressing an individual loan officer's personal use of a consumer AI assistant with borrower data; every CFPB, FHFA, and GSE document located in this research targets institutional AI/ML systems used in underwriting, pricing, servicing, or valuation, not a single LO's own tools. report [25] Treat that gap as a reason to ask, not a reason to assume you are clear; see the "Ask compliance the right questions" workflow below.
The LO's own toolkit
Once the company systems are accounted for, a short list of tools is commonly bought by the individual loan officer, budget and company policy permitting. Prices are dated 2026-09-03 unless noted, and every figure below is a third party's reported number or a vendor's own claim, not a price this guide re-verified against a live checkout.
| Tool | Category | Price (2026-09-03) | Evidence |
|---|---|---|---|
| Bonzo | CRM engagement layer / conversation automation | About $99 per user per month | report [9] |
| Jungo | CRM, built on Salesforce | About $96 per user per month, plus the underlying Salesforce license | report [9] |
| HubSpot | General CRM, not mortgage-specific | Tiered published pricing, not re-verified in this pass | inference [12] |
| MBS Highway | Rate and market education, client alerts | All-Access about $199.95 per month; a "Certified Mortgage Advisor" tier cited at $1,997 initial plus $297 per year | report [21] |
| Homebot | Past-client home-equity and rate-alert emails | Cited at roughly $125 to $300 per month for a loan officer tier | report [21] |
| Mortgage Coach | Total Cost Analysis presentation builder | Not published; the vendor claims 80 percent of the Top 25 Retail Lenders use it | claim [21] |
Review platforms, scheduling tools, and video tools built specifically for loan officers were not published or confirmed in this research pass. That is not the same as "none exist"; it means the source list behind this chapter did not surface a named vendor and price for that category. Ask your compliance officer for the company's approved list before signing up for one, and ask the vendor directly for current pricing and whether the product touches borrower NPI in a way GLBA covers.
Why your company is about to have an AI policy
Two GSE bulletins landed on a near-identical 2026 timeline, and both point at your company, not you directly, but both will produce a written AI policy that reaches your desk. Freddie Mac Bulletin 2025-16, issued 2025-12-03, requires seller/servicers to have a comprehensive AI/ML governance framework, effective 2026-03-03. Fannie Mae Lender Letter LL-2026-04, issued 2026-04-08, requires seller/servicers to maintain policies covering AI/ML risk management, applicable law, "trustworthy and ethical use," employee communication, and at-least-annual program review, effective 2026-08-06. report [24]
Both bulletins are written at the institutional level, aimed at how a lender governs AI/ML inside underwriting, pricing, and valuation systems. Neither one names personal generative-AI use by an individual loan officer. But a company that is now required to write an AI governance policy for its institutional systems is very likely to extend that policy, or write a companion one, to cover what its loan officers do with tools like Claude on their own desk. If your company has not yet told you where personal AI use of borrower data stands, that policy is coming; asking early is the safer move than waiting to be told.
Where Claude fits and where it never touches
Given the ownership map above and the GLBA boundary, Claude's working lane on a mortgage desk is narrow by design: drafting, summarizing, and preparing lists from information you choose to give it, never entering data into the LOS unattended, never producing a disclosure, and never making or implying a pricing decision. That mirrors the line ICE itself draws for its own embedded AI inside Encompass. claim [2]
- Claude never enters data into the LOS, POS, or CRM without you reviewing and clicking the entry yourself.
- Claude never drafts or sends a disclosure; disclosures are generated by the company's LOS and doc-prep vendor from verified loan data.
- Claude never states or implies a rate, an approval, or a pricing outcome; those come from the company's pricing engine, credit vendor, and AUS.
- Claude works from what you type or paste in: your own notes, a de-identified summary, or the desk file and program matrix from Chapter 01, not a live pull from a system holding borrower NPI.
Application intake prep summary
Trigger: you just got off a discovery call with a prospective borrower and have your own handwritten or typed notes.
What Claude prepares: reading only what you choose to type in (no live connector to the POS or LOS), Claude turns your notes into a document checklist matched to the loan scenario you described (pay stubs, W-2s, bank statements, and so on) and a short list of open questions to ask before the application is submitted.
What you check and do: you verify the checklist against your company's actual required-documents list for the loan type, fill in anything Claude flagged as unclear by asking the borrower directly, and you are the one who enters anything into the POS or LOS.
Never automated: Claude never receives borrower NPI beyond what you deliberately choose to type (a first name and a loan scenario is enough; a Social Security number or full account number is not needed and should not be pasted in), and it never submits anything to a company system on its own.
Conditions chase list
Trigger: an underwriter returns a conditions list on a file in processing.
What Claude prepares: you paste the conditions list with names, loan numbers, and any other identifiers removed first. Claude turns the raw underwriting language into a plain-English request addressed to each party who owes something: the borrower, the title company, the employer's HR department, and so on, grouped by who needs to act.
What you check and do: you re-attach the borrower's identity and loan number before sending anything, verify each request accurately reflects the actual condition (Claude is translating language, not interpreting underwriting intent), and send the requests yourself through the channel your company approves.
Never automated: identifiers are stripped before the conditions list ever reaches Claude, and Claude never sends a request directly; it only drafts the plain-English version for you to review and send.
Ask compliance the right questions
Trigger: you want to start using Claude on your desk and do not yet have a written answer from your company about it.
What Claude prepares: a short checklist for you to hand to your compliance officer or manager, covering: which AI tools, if any, are currently approved for use with borrower data; whether the desk file, program matrix, and de-identified summaries described in this chapter are acceptable to keep in a personal AI account; what the company's position is on the GSE AI governance bulletins referenced above and whether a written personal-AI-use policy is planned; and what the company's data-retention and training-opt-out settings should be if a paid Claude plan is approved for company use (Chapter 08 covers the plan-level detail).
What you check and do: you are the one who asks the question and gets a written answer before touching any borrower data in a personal tool; a verbal "that should be fine" is not the same as a policy.
Never automated: nothing in this workflow authorizes Claude to touch borrower data before compliance responds; the checklist itself contains no borrower information at all.
What stays human
- Every entry into the LOS, POS, or CRM: Claude may draft the content, you click the entry.
- Every disclosure, which is generated by the company's LOS and doc-prep vendor, never by Claude.
- Every pricing, rate, or approval statement, which comes from the company's PPE, credit vendor, and AUS.
- Whether and how borrower NPI, as defined by GLBA, ever reaches a personal AI account; the default answer is that it does not.
- Getting a written answer from compliance before treating any of this chapter's Claude workflows as approved for your company.
Do this today
- Map your own desk against the ownership table above: which systems are company-provided, and which of the LO toolkit items you already pay for yourself.
- Write down, in one sentence, what your company's current written policy says about personal AI use and borrower data. If you cannot write that sentence, that is your answer.
- Send the "Ask compliance the right questions" checklist to your compliance officer or manager this week, before you attach anything but your own notes to a Claude Project.
Sources
- ICE Mortgage Technology, Developer Connect and Encompass Partner Connect documentation, developer.icemortgagetechnology.com/developer-connect/docs/welcome, retrieved 2026-09-04
- National Mortgage News, ICE Aurora and automated conditioning engine coverage, nationalmortgagenews.com/news/new-ai-mortgage-tools-unveiled-at-ice-experience-2026, retrieved 2026-09-04
- nCino, Mortgage MCP release announcement, ncino.com/news/ncino-releases-mortgage-mcp-connect-ai-agents, retrieved 2026-09-04
- nCino, AI capability update, ncino.com/news/ncino-advances-ai-mortgage-lending-with-powerful-new-capabilities, retrieved 2026-09-04
- MeridianLink, mortgage origination system product page, meridianlink.com/solutions/mortgage-origination-system, retrieved 2026-09-04
- Setshape, loan origination system comparison, setshape.com/blog/top-loan-origination-systems, retrieved 2026-09-04
- Blend, platform product page, blend.com/platform, retrieved 2026-09-04
- iJungo, mortgage POS systems review, ijungo.com/top-mortgage-pos-systems-review, retrieved 2026-09-04
- Floify, Lender Edition product page, floify.com/lender-edition, retrieved 2026-09-04
- Leadpops, mortgage CRM software guide, leadpops.com/guide/best-mortgage-crm-software-2026, retrieved 2026-09-04
- BNTouch, CRM texting compliance page, bntouch.com/mortgage-blog/crm-text-messaging, retrieved 2026-09-04
- ICE Mortgage Technology, Velocify product page, mortgagetech.ice.com/products/velocify, retrieved 2026-09-04
- HubSpot, remote MCP server general availability changelog, developers.hubspot.com/changelog/remote-hubspot-mcp-server-is-now-generally-available, retrieved 2026-09-04
- Banking Bridge, top pricing engines for 2026, bankingbridge.com/post/the-top-5-pricing-engines-for-2026, retrieved 2026-09-04
- Xactus, credit product page, xactus.com/credit, retrieved 2026-09-04
- The Mortgage Hub, DU and LPA overview, themtghub.com/how-desktop-underwriter-du-and-loan-product-advisor-lpa-impact-mortgage-applications, retrieved 2026-09-04
- ICE Mortgage Technology, AUS partners and samples reference, developer.icemortgagetechnology.com/developer-connect/reference/aus-partners-and-samples, retrieved 2026-09-04
- DocMagic, automated compliance page, docmagic.com/automated-compliance, and DocMagic blog, blog.docmagic.com/docmagic-reaches-300-million-mortgage-esignings, retrieved 2026-09-04
- Snapdocs, remote online notarization solutions page, snapdocs.com/solutions/remote-online-notarization, and Proof, platform rebrand announcement, proof.com/blog/platform-rebrand, retrieved 2026-09-04
- Argyle, verification provider guide, argyle.com/blog/choosing-a-verification-provider-for-mortgage-lending, and Truv, payroll product page, truv.com/products/payroll, retrieved 2026-09-04
- Leadpops, Zillow mortgage leads review, leadpops.com/blog/zillow-mortgage-leads-review-2026, and EZ Loan Docs, Zillow mortgage leads review, ezloandocs.com/zillow-mortgage-leads-reviewed, retrieved 2026-09-04
- Capterra, MBS Highway listing, and Leadpops, Homebot pricing reference, and Trust Engine, Mortgage Coach page, trustengine.com/mortgage-coach, retrieved 2026-09-04
- Federal Trade Commission, Safeguards Rule library page, ftc.gov/legal-library/browse/rules/safeguards-rule, retrieved 2026-09-04
- Jones Day, FTC amended Safeguards Rule breach-notification alert, jonesday.com/en/insights/2023/11/ftc-requires-nonbank-financial-institutions-to-report-data-security-breaches-under-amended-safeguards-rule, retrieved 2026-09-04
- Harris Beach Murtha, Fannie Mae and Freddie Mac AI standards alert, harrisbeachmurtha.com/insights/fannie-mae-and-freddie-mac-set-new-ai-standards-for-mortgage-lenders, retrieved 2026-09-04
- MLO tech-stack research pass (MLO-WS-B), internal research note on the absence of a regulator statement addressing personal AI use of borrower data, retrieved 2026-09-04
Starting from zero? Use the field manuals first.
02 · BEFORE YOU BUY SOFTWARE
If you're still choosing the buyer, offer, stack, agents, content, customers, and delivery, don't start with a tool: start with the manuals. These create local worksheets only: no checkout, outreach, or account setup.
Find the first manual
Not sure which guide to open first? Start from your stuck point and let the finder recommend the next manual and the first worksheet question. No personal data is collected: it just points you at the right starting move.
Work the starter pack
Move through buyer, offer, stack, agents, content, customers, and delivery as a sequence of short local worksheet sessions. One artifact at a time, in order, so you build the business instead of browsing for it.
See what "done" looks like
Read a fully worked field manual before writing your own version, so the output feels concrete: a real, numbered workflow with the risky steps gated, not a fake proof claim.