Embedded Files
Case Study
Note: To comply with my non-disclosure agreement, I have omitted and obfuscated confidential information in this case study. All information in this case study is my own and does not necessarily reflect the views of my current employer.
My Role
My Role
🤔 Concepting / ✏️ Sketching / 📝 Wireframing / 🎨 UI/UX Design /
🤖 Prototyping
Tools
Tools
Figma
Platform
Platform
Desktop
Timeline
Timeline
3 Day Innovation Hackathon
The Business Problem:
The current underwriting governance ecosystem is characterized by extreme Information Density and Systemic Silos. Underwriters are forced to navigate a "manual-discovery" landscape that actively hinders speed-to-market.
Essential data is scattered across 40+ corporate bulletins, disparate Business Unit guidelines, and individualized authority letters.
Locating specific requirements for complex exposures is a non-linear, time-intensive process that lacks a "Single Source of Truth."
Bottlenecks when manual search fails, underwriters must escalate to Corporate Underwriting. This dependency on human availability creates decision latency, stalling policy issuance and increasing the risk of inconsistent risk selection.
The Strategic Solution:
Retrieval-Augmented Generation or (RAG) interface, purposely designed as a persistent "Side-Car" to the underwriter's primary workspace. This architecture was chosen to solve for Trust and Context. Side-Car "reads" the active policy on screen, using RAG to fetch specific governance clauses from the 40+ bulletins in real-time.
To eliminate the "Black Box" AI problem, every AI-generated answer includes deep-link citations. Clicking a citation automatically scrolls the main document viewer to the exact paragraph in the Governance Manual.
The UI personalized search results by cross-referencing the user's individual authority letter, ensuring they only see guidance relevant to their specific premium capacity and limits.
If AI detects a query that exceeds the user’s authority, the UI contextually triggers a "Draft Referral" button, prepopulating an escalation form with the necessary governance citations for Corporate UW.
The Result:
The challenge wasn't just finding data—it was creating a verifiable path to a decision. By using a Side-Car UI, we kept the Underwriter as the 'Human in the Loop,' providing them with the intelligence they need without ever forcing them to leave their primary workflow.
We estimated a 65% Reduction in search latency. Underwriters now resolve governance inquiries in seconds, eliminating the "hunt-and-peck" manual search process.
Reduced 'clarification friction' by 30% through the implementation of a transparency-first AI UI. By designing a system that surfaces the 'Source of Truth' alongside AI insights, we enabled autonomous decision-making for junior underwriters and reclaimed significant bandwidth for the Corporate Underwriting team.
Alignment with Governance, real-time data retrieval ensured that every decision was backed by the most current version of corporate bulletins, virtually eliminating compliance oversights.
Results section:
Designed a logic-gate system within the Side-Car. If the RAG engine identifies that a query result exceeds the user's "Authority ID" (stored in their profile), the UI dynamically swaps the "Approve" button for a "Draft Referral" & Governance Review CTA.
Contextual Pre-Population: We engineered a "Data Bridge" where the AI extracts the specific governance clause it just found and auto-populates it into a referral template.
Result: The underwriter doesn't have to "prove" the rule to the VP; the AI has already cited it.
The "Referral Preview" Modal: To maintain the "Human-in-the-Loop" principle, we designed a transition state where the underwriter can review, edit, and add personal notes to the AI-generated referral before it is sent to Corporate Underwriting.
Retrospective:
Challenge: The core challenge involved mitigating the "trust-gap" inherent in high-stakes underwriting by transitioning from a black-box AI model to a
transparent, verifiable decision-support system. I solved for high cognitive load and "information density" by architecting a non-obstructive Side-Car UI that paired AI-synthesized insights with deep-link citations, ensuring underwriters could validate 40+ governance sources without losing document context. This strategic intervention transformed "hard-stop" manual bottlenecks into a streamlined, actionable workflow that automated the transition from insight to referral.
The Takeaway: The success of the Side-Car (RAG) architecture proved that enterprise AI must function as a contextual assistant rather than a siloed destination, prioritizing the user’s "flow state" through a persistent sidebar that minimizes cognitive switching. By implementing deep-link citations that anchor AI insights to exact source coordinates, I transformed the interface from a speculative "search tool" into a verifiable "navigation engine." The strategic value was ultimately realized through automated referral logic, where the system identifies authority breaches and proactively handles the heavy lifting of drafting escalations, bridging the gap between raw data and decisive action.
The Takeaway: The success of the Side-Car (RAG) architecture proved that enterprise AI must function as a contextual assistant rather than a siloed destination, prioritizing the user’s "flow state" through a persistent sidebar that minimizes cognitive switching. By implementing deep-link citations that anchor AI insights to exact source coordinates, I transformed the interface from a speculative "search tool" into a verifiable "navigation engine." The strategic value was ultimately realized through automated referral logic, where the system identifies authority breaches and proactively handles the heavy lifting of drafting escalations, bridging the gap between raw data and decisive action.
The Result: The Side-Car RAG interface, we optimized the underwriting workflow for "flow state," replacing high-friction tab-switching with a data-agnostic system that scales from 40 to 4,000 bulletins across global business units. We mitigated AI mistrust by implementing deep-link citations and "One-Click" referral logic.
Our team secured 1st Place in the internal Innovation Hackathon by developing a data-agnostic RAG interface designed for global scale. By decoupling the UI from specific datasets, we demonstrated a solution that maintains high technical viability whether processing 40 or 4,000 policy bulletins, offering immediate, cross-departmental business impact.
Google Sites
Report abuse