Making AI output usable in loan origination
AI-enabled origination · Work in progress
The work
I am designing interfaces and repeatable interaction patterns for an AI-enabled loan-origination platform. The work supports processors and underwriters reviewing information across documents and domains such as credit, assets, and collateral. The platform is intended to extract information, compare it with customer-provided and system data, and reduce the manual work involved in preparing information for loan decisions.
I am developing an internal design kit aligned with brand standards so these patterns work consistently across several agent domains. The design needs to help people understand the information presented, assess AI-generated conclusions, and recognize where they can review or act.
Making AI-assisted work understandable
In loan origination, people need to understand the basis for an AI-generated assessment and where their own judgment is required. I use these questions to guide the interface:
- What comes directly from documents or system records, and what has the AI inferred?
- What information is missing, uncertain, or contradictory?
- Is this a recommendation, an action awaiting approval, or something already done?
- What do I need to verify, and which decisions remain mine?
- If something looks wrong, how can I check its source, review what happened, and correct or escalate the problem?
Turning complexity into clear decisions
I work with business, product, and engineering partners to reconcile business priorities, product intent, technical capabilities, and user needs. I make clear why an interface pattern was chosen, what it leaves unresolved, and what evidence would justify changing it.
I translate those decisions into repeatable interface patterns, giving people consistent ways to read complex information, assess recommendations, and understand the actions available to them across the platform.
Status
This is ongoing, unlaunched work. The questions above guide the design; improvements in speed, confidence, and decision quality remain goals rather than measured outcomes.