Confer Solutions AI - AI-Powered Business SolutionsConfer Solutions AI
RFP & Buyer's Guide
Updated April 28, 2026•11 min read

12 Questions to Ask Your Mortgage LOS Vendor in an RFP (and the Answers That Matter)

Most mortgage LOS RFPs ask the wrong questions. Twelve that separate vendors who do real work from vendors who slide a deck — with the right-answer pattern and the red-flag tell for each.

Yatin Karnik

Founder & CEO, Confer Solutions

TL;DR

Most mortgage LOS RFPs are too long, too generic, and don't separate marketing AI from production AI. The 12 questions below put each vendor on the spot to demo real work on your real data. Each comes with the right-answer pattern (what a production-ready vendor sounds like) and the red-flag tell (the response you should read as marketing). Use them in 60-minute working sessions with each finalist rather than a 200-page written response.

How should an operations leader run a tight mortgage LOS RFP?

Three working sessions per finalist. First session: the AI agent demos (questions 1, 3, 7, 10). Second session: integration, compliance, and audit trail (questions 2, 4, 8, 9). Third session: implementation, pricing, and operational fit (questions 5, 6, 11, 12). Cap each at 60 minutes with all the questions on a shared doc. Vendors that can't show real work in this format aren't ready for your shop.

The 12 questions

1

Question

Show me your AI document agent classifying 30 of our real loans on this call.

What a production-ready answer sounds like

Vendor opens a sandbox, you upload three loans' worth of documents, the agent classifies each in <5 seconds with confidence scores. Vendor commits to the same demo on a larger sample for the proof of concept.

Red flag

We have AI features but a live demo would require a separate engagement. Translation: it's marketing, not production.

2

Question

What happens to a TRID 3-day timer if your servers restart mid-clock?

What a production-ready answer sounds like

Timers run in a durable workflow engine (Temporal, Cadence, or equivalent). State is persisted at every transition. Server restarts, deploys, and worker crashes do not lose the clock.

Red flag

We have monitoring and alerts. Translation: the timer is a cron job and your CFPB exposure is real.

3

Question

Walk me through how your platform calculates Schedule C income with depreciation, business use of home, and 50% meals; show the math.

What a production-ready answer sounds like

Vendor opens a 1084-compliant calculator, demonstrates each add-back with the calculation, shows 2-year trending logic, and confirms the same inputs produce identical outputs every time. Math is deterministic, not LLM-generated.

Red flag

Our LLM handles that. Translation: income/employment defects are the #1 critical defect category per ACES; you do not want an LLM in the math path.

4

Question

How many Encompass fields do you sync, in which directions, and which E-Folder containers do you map?

What a production-ready answer sounds like

Specific number (e.g., 180+), confirmation of bidirectional sync, named E-Folder containers (U1–U15) mapped explicitly. Demo of a document classified by AI and routed to the correct container.

Red flag

We have an Encompass integration. Translation: it might be a one-way export.

5

Question

What is your average implementation timeline for a lender at our volume; and what's the longest one you've done?

What a production-ready answer sounds like

Specific range (e.g., '30–60 days for hybrid, 8–12 weeks for standalone at 5K loans/year'). Reference customer or scoped statement of work. Honest answer about edge cases that extend timelines.

Red flag

It depends. Translation: the vendor doesn't have a repeatable implementation pattern.

6

Question

How do you price for a lender with seasonal volume; say, 1,000 loans in Q1 and 3,000 loans in Q3?

What a production-ready answer sounds like

Usage-based pricing scales with funded volume. No per-seat licensing penalty for hiring temp staff in peak season. Annual commitment with overage at a known rate.

Red flag

Per-seat licensing with annual commitment. Translation: you'll pay for inactive users in Q1 and re-negotiate to add temp staff in Q3.

7

Question

Walk me through QM/ATR verification on a loan where the borrower has 7-year employment history but two recent 30-day employment gaps.

What a production-ready answer sounds like

All 8 ATR factors verified per loan with stored evidence. Specific handling of employment gaps (continuity of income vs. employment, residual income tests). Block-or-disclose workflow gating until factors are documented.

Red flag

Our compliance team would review. Translation: ATR is post-close QC, not workflow architecture.

8

Question

Where does your immutable audit log live, and can you show me an example entry?

What a production-ready answer sounds like

Append-only ledger in a database with row-level security. Every entry shows operator, timestamp, before/after state, and reasoning where applicable. Demo of an audit query against a live loan record.

Red flag

We log to standard application logs. Translation: regulator-grade audit trails will be reconstructed from logs and inferences. That's a finding waiting to happen.

9

Question

Can a third-party LLM or BI tool act on loan data through your platform; and how?

What a production-ready answer sounds like

Yes, via Model Context Protocol (MCP) or open API. Specific tools exposed (e.g., 32+ MCP tools across underwriting, document AI, compliance). Authentication and tenant isolation explicitly described.

Red flag

We have an SDK. Translation: each integration is a per-customer engineering project.

10

Question

Demo your voice AI agent. Call it from my phone right now.

What a production-ready answer sounds like

Vendor provides a number, you call it, the voice AI answers with mortgage-vocabulary fluency, can pull live loan-status data when given a sample loan number, and transfers cleanly to a human with full call context.

Red flag

Voice AI is on the roadmap. Translation: borrower self-service is not in scope this quarter.

11

Question

What does your post-close QC look like; sampled or 100%?

What a production-ready answer sounds like

100% pre-fund continuous QC: income reproducibility, AUS findings vs. final terms, TRID timing audit, HMDA completeness, ULDD validation against investor specs. Defects route to a review queue before close, not a discovery report after fund.

Red flag

We sample 10% post-fund. Translation: critical defect rate stays at the ACES 1.79% industry average and repurchase exposure compounds.

12

Question

Show me the contract change request you'd send for a new HMDA edit-check rule the FFIEC published last quarter.

What a production-ready answer sounds like

Configuration change in admin UI, deployed same-day or next-day. No code change required. Demo on a sandbox tenant.

Red flag

That would be a custom development engagement at $X/hour. Translation: every regulatory update is an IT capacity tax.

Related reading

Frequently asked questions

What is the most important question to ask a mortgage LOS vendor in an RFP?

Question #1; show me your AI document agent classifying 30 of our real loans on this call. It's the single best filter between vendors with production AI and vendors with marketing AI. The follow-up question; what happens to a TRID timer when your servers restart; separates durable workflow engineering from cron-job timers and identifies your highest-severity compliance risk in one question.

How long should a mortgage LOS RFP take?

An RFP focused on the 12 questions above can be completed in 3–4 vendor sessions of 60 minutes each, plus a written response covering pricing, security, and compliance documentation. The mistake most lenders make is running a 6-month RFP that exhausts internal capacity before any vendor is meaningfully evaluated. Tighter RFPs with concrete demo requirements get better outcomes.

Should we include Encompass in our RFP if we're already on Encompass?

Yes; the question is whether you're upgrading Encompass, replacing it, or running a hybrid. Including Encompass in the RFP forces the incumbent to compete on capability rather than relationship. The hybrid option (keep Encompass as system of record, add AI-native LOS in front via 180+ field sync) is often the right answer for mid-sized lenders that have invested heavily in Encompass.

What pricing model should we expect for a modern mortgage LOS?

Usage-based annual contracts are the modern model. Per-seat licensing penalizes seasonal staffing fluctuations and feels less aligned with funded-loan economics. Per-loan pricing is fine but watch for tiered overage rates that punish growth. Total cost of ownership comparisons need to include AI plug-in costs, implementation services, integration build-out, and the labor cost of ongoing customization; those are often the largest line items.

How do we evaluate AI capabilities without bringing in a data scientist?

Pick three real loans from your shop. Ask the vendor's AI to classify the documents, calculate a 1084 income on the self-employed file, walk through ATR verification on a borderline DTI scenario, and explain what would clear it to approve/eligible. If the AI can do real work on your data in real time, it's production-ready. If the demo requires a sandbox preload or a follow-up engagement, the AI is still in the marketing stage.