This is a demo description, Please dont mind
3 Months (Ongoing)
Timeline
Manager + Me
Team
Manager + Me
Landscape
Filing an insurance claim is the most critical and high anxiety touchpoint in our application. When users submit documents missing vital information or upload them in the wrong format, it triggers an IR (Information Request) query from the insurer. This extends the claim timeline and causes immense frustration. The goal was to put control back into the hands of the user. By intercepting these errors proactively with AI, we could eliminate the manual review waiting period and create a much more seamless experience.
SUMMARY
Problem
During the highly stressful claim filing process, users frequently uploaded incorrect, illegible, or misplaced documents. This triggered endless query loops with the insurer and delayed claim payouts significantly.
Solution
Designed an AI powered document scanning interface that validates uploads in real time. The system categorizes document errors and provides clear, actionable steps for the user to fix issues before the claim is submitted.
Impact
Reduced overall claim initiation time by 26%, saving users up to 4 days of processing delays and significantly lowering manual support intervention.
I worked directly with a single Product Manager on this initiative. While the PM defined the backend AI logic and researched the core requirements, I took full ownership of the end to end design execution. We collaborated closely on defining the error hierarchy and ran multiple post launch reviews to refine the interface based on actual user behavior.
Journey
1. Workflow Auditing & Logic Mapping I mapped the existing claim submission flow alongside the PM to identify exactly where document drop offs and errors occurred. We isolated the most common user mistakes, such as blurry photos or missing hospital stamps. 2. AI Feedback Structuring Designing an interface for AI requires carefully managing how the machine talks to the human. I structured the UI to clearly separate the AI's findings, prioritizing critical errors over minor suggestions to ensure users knew exactly what action to take next. 3. Live Testing and Iteration We launched the initial version in three months to gather real world data. By tracking live queries and user drop offs during the scanning phase, I identified new friction points. I then executed a second design iteration to polish the scanning feedback loop further. We continue to treat this as a living project, optimizing the UI based on incoming data.
Hurdle:
Cognitive Overload from AI Flagging Users often upload multiple multi page PDFs at once. If the AI flagged every single minor flaw simultaneously across all documents, the interface would completely overwhelm the user, causing them to abandon the claim entirely.
Solution:
I designed a strict triage system for the AI scanner interface. The design prioritizes "Important Issues" at the very top. These are critical flaws that guarantee a claim rejection or an IR query. "Recommended Issues" (flaws that might slow down the process but will not cause outright rejection) are placed in a secondary, less visually urgent section. This allowed users to fix the most critical errors first without feeling paralyzed by a massive list of corrections.
The AI validation system successfully transformed a manual, error prone process into a seamless, user controlled experience. By catching mistakes before they reach the insurer, we dramatically improved the reliability of our claims ecosystem. Key Metrics Achieved: -26% Reduction in average claim initiation time. -4 Days Saved on average per user by eliminating the manual query and re-upload cycle. -Drastically reduced the volume of manual IR queries raised by the Pazcare support and operations teams.