Direct Mortgage and Multimodal: loan documents from 10-week closings to 5
Direct Mortgage Corp., a 29-year US residential mortgage lender, deployed Multimodal's AgentFlow (Document AI plus Decision AI) to classify and extract data from loan documents and automate parts of underwriting. Per Multimodal's case study, closing time fell from 10 weeks to 5, document backlogs fell from 1,199 (May 2024) to 4 (Jan 2025), cost per processed document fell 80%, and approvals ran up to 20x faster. Every figure is vendor-published; the client CEO is named and quoted.
| Metric | Before | After |
|---|---|---|
| Loan closing time | 10 weeks | 5 weeks |
| Document backlog | 1,199 (May 2024) | 4 (Jan 2025) |
| Cost per processed document | 80% reduction | |
| Application approval speed | manual review | up to 20x faster; approvals within 24 hours |
| Straight-through processing accuracy | 97 to 100% across major document types | |
| Documents processed per day | about 172, a 16% rise |
The problem
Mortgage lending runs on documents: pay stubs, tax returns, bank statements, insurance and title paperwork, hundreds of formats, all reviewed by hand. “Direct Mortgage Corp. (DMC) is a 29-year veteran in residential mortgage lending” (source), and by May 2024 its document backlog stood at 1,199 with loan closings taking 10 weeks.
What was built
“DMC partnered with us to implement AgentFlow, a suite of AI tools powered by generative AI… Combining Document AI, Decision AI, and generative AI, Multimodal deployed AI-powered workflows that automated the entire document lifecycle” (source). The models were customised on DMC’s internal datasets, built from GPT-3.5, Llama 2, and LightGBM. The pilot was deliberately hard, per CEO Jim Beech: “We gave [Multimodal] the most difficult form to process, a paystub. That was our test pilot. I couldn’t throw anything more difficult at them than that. And in 30 days, they had it resolved” (source).
The outcome
Closings halved, backlog to near zero. Per the case study: “Faster loan closing time: Loan processing dropped from 10 weeks to just 5… Queue elimination: Document backlogs dropped from 1,199 in May 2024 to only 4 by January 2025, bringing borrower queues to zero” (source).
The headline ratios. “200+ types of processed documents; 20x faster application approval process; 80% cost reduction per processed document… Borrowers now receive approvals within 24 hours” (source). These are vendor-stated multipliers; the baselines and definitions behind them are not published, which is why they stay amber here.
Accuracy and throughput. “Automation achieved 97–100% straight-through processing accuracy across major document types, including 100% accuracy in categories like verification of employment and title commitments… Daily processing capacity rose by 16%, reaching an average of 172 documents per day… 50% of bank statements are now auto-processed” (source), and automated verification “reduces what once took an hour of manual effort to just 10 minutes” (source).
How this was verified
This case is pending, and the caveat is the same as for every vendor-published story: all figures come from Multimodal’s own customer story, and no independent source reports them. The 20x and 80% headline numbers in particular need their definitions confirmed before they can be treated as measured outcomes. What keeps the file on the record: the client is named, the CEO is quoted at length and on the record (“Nobody is doing what we’re doing with [Multimodal], not even close”), and the backlog figures carry dates, which makes them checkable. Green requires Direct Mortgage to confirm the closing-time drop, the backlog numbers and dates, and the cost and speed baselines on the record.
Sources
Cited in this case file. Tier 3 = vendor or first-party. Each figure was checked against the live source on 2026-07-07, and against the cited archive capture.
- Multimodal customer story, “Mortgage application AI at Direct Mortgage Corp.” (Tier 3, vendor-published with the client CEO named and quoted). https://www.multimodal.dev/customer-stories/mortgage-application-ai · archived
Multimodal AgentFlow: Document AI + Decision AIModels customised on DMC's own documents (GPT-3.5, Llama 2, LightGBM)
- Status
- pending
- Method
- Sourced by The Internet Ninja against the public record: every figure below is quoted from the cited source and, on 2026-07-07, checked against the live page over the open network, with the cited web.archive.org capture also confirmed. This case is pending, it has not yet cleared TIN's green validation bar. The single source is Multimodal's own customer story, vendor-published marketing material; no independent source reports these figures, and headline multipliers like 20x warrant extra caution until the definitions behind them are confirmed. The mitigating factor: the client is named and its CEO, Jim Beech, is quoted on the record. A green badge requires Direct Mortgage to confirm the closing-time drop, the backlog figures and dates, and the cost-per-document baseline on the record.
- Provider
- Multimodal, AgentFlow document and decision AI
- Client
- Direct Mortgage Corp. · Mortgage lending (US)
- Disclosure
- named
What changed at Direct Mortgage after deploying Multimodal?
Per the vendor case study, loan closing time fell from 10 weeks to 5 and document backlogs fell from 1,199 in May 2024 to 4 by January 2025, with automated classification and extraction across 200+ document types.
Why is this case file pending, not verified?
All figures are vendor-published, and the 20x and 80% headline multipliers need their definitions confirmed. A green badge requires Direct Mortgage to confirm the numbers on the record.