Case Study: How a $75M Direct Lender Automated Underwriting Intake and Reclaimed 18 Hours Per Week
Every direct lender at the $50M–$100M deployment mark hits the same wall. Volume is growing. The ops team isn't. And the bottleneck is almost always the same place: underwriting intake.
Files arrive as email attachments. Stips are scattered across threads. Someone on the ops team spends their morning assembling a package that should have been structured before it ever hit the queue. By the time the underwriter touches it, the file is already two days old.
Here's what that problem looked like for one $75M direct lender — and what happened after they automated it.
A Two-Person Ops Team Running on Manual Assembly
This lender operates in mid-market specialty finance, deploying roughly $75M annually with a team of 12. Two ops staff handled all back-office work: underwriting intake, document collection, stip tracking, and file prep.
Their intake process looked like this:
- Borrower documents arrived via email in no consistent format
- An ops staff member manually sorted PDFs, identified document types, and logged them into a tracker
- Missing stips were identified by reviewing the file manually, then chased via email
- Files moved to underwriting only after a full manual review pass
- Average time from application receipt to underwriter-ready file: 3.1 business days
The ops team was spending roughly 18 hours per week on this single workflow. Nearly half of one FTE — doing work that added no credit judgment, no relationship value, and no origination output.
The math is brutal. At $75M deployed, that's 18 hours per week spent on document sorting.
What the Workflow Actually Needed
Before building anything, the Starter Stack team ran a workflow diagnostic. This is standard practice. You don't deploy agents against a broken process — you map it, find the actual friction points, and build against those.
The diagnostic surfaced three specific failure modes:
- No structured intake channel — Documents arrived via email with no naming convention, no required format, and no automated acknowledgment. The ops team was the intake system.
- Stip tracking lived in a spreadsheet — Missing documents were identified manually, logged manually, and chased manually. No automated flag when a required item was absent.
- File handoff to underwriting was informal — No structured checklist, no confirmation of completeness, no audit trail of what arrived when.
These aren't technology problems. They're workflow design problems — ones that technology can solve once the process is mapped correctly.
What the Agent Does
Starter Stack built a custom underwriting intake agent configured to this lender's specific document requirements, credit policy stips, and handoff criteria. The agent doesn't replace underwriting judgment. It handles the repeatable, structured work that was consuming 18 hours per week.
On every new application, the agent:
- Receives and classifies incoming documents from a designated intake channel — identifying document type, borrower entity, and application reference automatically
- Cross-references the file against the lender's stip checklist, flagging missing or expired items within minutes of receipt
- Generates a structured intake summary showing document status, outstanding stips, and any items requiring human review
- Triggers an automated stip request to the broker or borrower for missing items, with a timestamped follow-up queue
- Routes the completed file to underwriting only when the intake checklist is satisfied, with a full audit trail attached
The agent runs on Starter Stack's managed infrastructure. The lender doesn't manage software, monitor dashboards, or maintain any part of the system. When something needs adjustment, Starter Stack handles it.
The Results
The engagement went live in 26 days. Results measured at 90 days post-deployment:
- Time to underwriter-ready file: 3.1 business days → 0.8 business days
- Ops time on intake: 18 hours per week → 3.5 hours per week
- Stip chase cycle: 2.4 follow-up emails per file → one automated request with a structured response queue
- File completeness at handoff: 61% on first pass → 94%
- Ops staff reallocation: 14.5 hours per week freed for exception handling, borrower servicing, and portfolio monitoring support
The two ops staff members didn't lose their jobs. They stopped sorting documents and started doing work that actually requires human judgment.
What Did Not Change
Worth naming directly: the lender didn't rip out their existing systems. Their LOS stayed in place. Email infrastructure stayed in place. Underwriting process and credit policy stayed exactly as written.
The agent plugged into the existing intake channel. No new system, no new login, no new vendor relationship for the ops team to manage. Starter Stack runs the system. The lender's team uses the output.
Client data doesn't enter a shared platform and doesn't train any shared model. The agent is built to this lender's specific stip requirements and credit logic. That specificity is what makes it work.
Why Generic Automation Fails Here
Most SaaS tools in this category are built for volume lenders with standardized document sets. A $75M direct lender in specialty finance doesn't have standardized document sets. Stip requirements vary by deal type, collateral, and borrower profile. A generic classification tool trained on residential mortgage files doesn't know what to do with a UCC filing, a rent roll, or a personal financial statement formatted differently by every borrower.
This is the same pattern documented in the document intelligence case study for a CRE debt lender: the failure mode isn't the technology — it's the mismatch between a generic tool and a specialized workflow.
For mid-market lenders, the challenge isn't just automation. It's automation that encodes your credit logic, your stip requirements, and your handoff criteria — not someone else's. The mid-market back-office breakdown covers this pattern in more detail.
This Problem Is Not Unique to This Lender
Underwriting intake is the most common bottleneck we see at the $50M–$100M deployment range. Volume has grown past what manual assembly can handle. The ops team is capable. They're just doing the wrong work.
The same dynamic shows up in bank statement analysis for revenue-based financing lenders, where the intake bottleneck is financial data extraction rather than document sorting. Different workflow, same root cause: repeatable structured work consuming hours that should go toward judgment-based tasks.
If your underwriting intake looks anything like what's described above, 18 hours per week is a floor, not a ceiling. Most lenders in this range are spending more.
What a First Engagement Looks Like
Starter Stack starts every engagement with a workflow diagnostic. You don't commit to a build before you know exactly where the friction is. The diagnostic identifies the highest-impact workflow to automate first, maps the current process, and defines what the agent needs to do.
From there, build and deployment runs under 30 days to a live first workflow. No rip-and-replace. No software to manage. One point of accountability.
To see what this looks like for your underwriting intake specifically, Starter Stack runs a 30-minute workflow assessment — identifying where the hours are going and what an agent would actually do about it.
FAQs
What is underwriting intake automation for direct lenders? Underwriting intake automation uses AI agents to receive, classify, and organize borrower documents, check files against a stip checklist, flag missing items, and route complete files to underwriting — without manual sorting by an ops team member. The agent handles the structured, repeatable work so your ops staff can focus on exceptions and judgment-based tasks.
How long does it take to deploy an underwriting intake agent? For a non-bank direct lender with a defined stip checklist and an existing intake channel, Starter Stack typically deploys a live first workflow in under 30 days. The process starts with a workflow diagnostic before any build begins.
Does automation require replacing the existing loan origination system? No. The agent integrates with the lender's existing intake channel and document workflow. No requirement to replace the LOS, change email infrastructure, or adopt new software. The agent plugs into the existing process and handles the structured work within it.
What happens to the ops staff when intake is automated? In this case, 14.5 hours per week were freed from document sorting and reallocated to exception handling, borrower servicing support, and portfolio monitoring. The ops team didn't shrink. The work they do became more valuable.
Does the agent use a shared model trained on other lenders' data? No. Starter Stack builds each agent to the specific lender's credit logic, stip requirements, and handoff criteria. Client data doesn't enter a shared platform and doesn't train any shared model. Deployment runs on Starter Stack managed infrastructure, or optionally within the client's own environment.
What's the difference between a managed service and a SaaS tool for this use case? A SaaS tool gives you a dashboard and a configuration interface — you manage it, maintain it, and troubleshoot it. Starter Stack is a managed service: agents are built, deployed, and run by Starter Stack. When something needs adjustment, Starter Stack handles it. No per-seat licenses. No software for your team to manage.
What types of lenders benefit most from underwriting intake automation? Non-bank direct lenders deploying between $20M and $300M annually with small ops teams typically see the strongest results. The trigger is usually origination volume growing faster than the ops team can handle manual file assembly. Specialty finance, business credit, CRE debt, and working capital lenders all fit this profile.
18 hours per week on document sorting is not a staffing problem. It's a workflow problem. And workflow problems have a specific solution.
Book a 30-minute workflow assessment at starterstack.ai to see exactly where your intake hours are going.