Kizen’s AI Agents: Build, Train & Scale In Under 30 Days. Learn more.
Kizen’s AI Agents: Build, Train & Scale In Under 30 Days. Learn more.
Financial Services Solutions
min read
Kizen's Encompass Document Indexing integration is an AI-powered orchestration engine that classifies, splits, and indexes loan documents — automatically. The entire user experience lives inside Encompass. Underwriters never have to leave their familiar environment or log into another system.
If you've spent any time in mortgage operations, you know the drill. A loan file comes in with dozens, sometimes hundreds, of documents attached: pay stubs, credit reports, flood certificates, disclosure packages, and verification letters. Many of them are unlabeled. Some are misfiled. And more often than you'd like, someone has submitted a single 80-page PDF that's actually 40 different documents stitched together.
Now multiply that by every loan in your pipeline.
I'm Jamie Signorile, VP of Solutions at Kizen. We built this integration because we kept hearing the same thing from underwriters and processors: the document review and classification step is one of the most painful, time-consuming parts of the entire loan lifecycle. With roughly 60% of all U.S. mortgage originations flowing through Intercontinental Exchange’s Encompass loan origination system, the platform plays a central role in modern mortgage operations. As lenders scale, document-heavy workflows are becoming more complex — creating a clear need for deeper automation within the Encompass ecosystem. Documents come into Encompass — whether from an end-user upload or a point-of-sale integration — and they arrive ready to be organized within the system. From there, teams review each document, determine what it is, apply the appropriate label, and file it into the correct e-folder. For an average loan, that review and organization process typically takes between 45 and 90 minutes.
We set out to help improve that process.
Kizen's Encompass Document Indexing integration is an AI-powered orchestration engine that classifies, splits, and indexes loan documents — automatically. The entire user experience lives inside Encompass. Underwriters and processors never have to leave their familiar environment or log into another system. From their perspective, it's a service they open within the Encompass interface, select or drag-and-drop documents, and click a button. Kizen handles the rest in the background.
The key architectural decision here was intentional: Kizen operates purely as a backend orchestration engine. In partnership with Encompass specialists, a dedicated middleware service has been developed to manage the front-end experience. Meanwhile, the Kizen platform performs the heavy lifting, sequencing AI agents to analyze documents through various stages and deliver results back to the middleware in real time.
When an underwriter submits a document, Kizen's orchestration engine first determines what it's looking at. Using LLM-powered analysis, the system classifies the document against the lender's specific classification taxonomy — which can include hundreds of document types. The result is returned with a classification, a confidence score, and the reasoning behind the decision, so underwriters can immediately see not just what the system thinks the document is, but why.
This is where things get interesting. In mortgage workflows, it's extremely common for borrowers or third parties to submit consolidated documents — a single PDF that actually contains multiple distinct documents bundled together. Think of an initial disclosure package that's 80+ pages and contains dozens of individual forms and letters.
Kizen automatically detects when a document contains multiple sub-documents. When it does, the system indexes the consolidated file, determines the page ranges for each distinct sub-document, splits them out into individual records, and then classifies each one independently. For example, an 82 page Initial Disclosure package was broken into 42 distinct sub documents, each individually classified, all processed in a few minutes.
As documents move through the orchestration pipeline, the underwriter sees real-time status updates in their Encompass interface. Each document displays its processing status, classification result, confidence score, and the AI's reasoning. For multi-document packages, child documents are displayed sequentially with their page ranges, and an integrated PDF viewer lets underwriters click through each section without leaving the dashboard.
We were deliberate about designing for the edge cases, because mortgage documents are messy. The integration supports several exception workflows:
Reclassification — If the AI's classification doesn't look right, the underwriter can override it with a manual classification. The system supports searching by category or specific document type across the lender's full taxonomy.
Re-splitting — If a multi-document package wasn't split correctly, the underwriter can provide new page ranges and trigger a re-split without starting over.
Single-to-multi conversion — Sometimes a document classified as a single document actually contains multiple sub-documents. The underwriter can flag it, provide page ranges, and the system will split and reclassify.
Multi-to-single conversion — The reverse scenario. If a document was split but should actually be treated as one package, the underwriter can consolidate it back and reclassify.
Every one of these actions is tracked with a full audit trail, including the user who performed the action, so compliance teams have complete visibility.
Once a document is classified — whether by the AI or through a manual override — it is automatically sorted into the appropriate e-folder within Encompass. This eliminates one of the most tedious manual steps in the process and ensures documents are consistently organized from the moment they're classified.
This is the capability I'm most excited about, and the one I believe is a genuine differentiator.
Every time an underwriter overrides a classification or manually classifies a document that the system couldn't handle, Kizen records that action in a detailed audit log. The system analyzes the history of all reclassifications, looking for patterns and trends. These reclassification patterns improve AI accuracy over time, specifically tuned to each lender’s document mix.
In practice, this means the system gets smarter over time, specifically tuned to each lender's document patterns. If underwriters are consistently reclassifying a certain type of document, the system learns from that and adjusts. It's a closed-loop improvement cycle that compounds in value the more it's used.
This stands in contrast to many competitors in this space who rely on static OCR technology or, in some cases, offshore teams manually reviewing every document. Those approaches don't learn. They don't get faster. And if you submit a batch on a Friday afternoon, you might not get results until Monday!
The numbers tell a clear story. Manual document indexing takes 45 to 90 minutes per loan application. With Kizen's integration, that drops to roughly 10 minutes — and most of that time is the underwriter reviewing results and handling the occasional exception, not doing the classification work itself.
For lenders processing hundreds of loans per month, this translates directly into increased underwriter throughput, faster loan cycle times, and significant cost savings on every loan originated. It also means fewer errors, more consistent filing, and a complete audit trail for every document decision.
And the timing matters. As mortgage rates trend downward, origination volumes are expected to rise. Lenders who can process more loans without proportionally scaling headcount will have a meaningful operational advantage.
This integration is the first step in a broader vision. Today, we're solving the document classification and indexing problem. But the underlying orchestration architecture we've built is designed to go much deeper into the underwriting workflow.
Future iterations will ingest additional loan data — borrower information, collateral details, loan terms — and enable capabilities like automated data validation, rule-based decisioning aligned to a lender's specific credit policies, and AI-assisted underwriting recommendations. The goal is to accelerate the entire underwriting process, not just the document step.
Whether that takes the form of a deeper Encompass integration or a standalone Underwriter Copilot product that works across loan origination systems, the foundation is in place.
If you're a lender running on Encompass and spending too much time on document workflows, we'd love to show you what this looks like. We're also going to be at the upcoming ICE conference in Las Vegas on 03/16-03/18— come find us at our booth #721and we'll walk you through a live demo.
Jamie Signorile is the VP of Solutions at Kizen, where he designs and builds AI-powered orchestration solutions for enterprise workflows. Connect with him on LinkedIn, https://www.linkedin.com/in/jamiesignorile/
Patrick Rader is the Director of Enterprise Sales at Kizen. To learn more about how Kizen’s solutions can help benefit your business, reach out to him at [email protected] and connect with him on Linkedin, https://www.linkedin.com/in/patrick-rader-a60020134/