Transforming Insurance Intelligence with the power of AI
I led the end-to-end redesign of a legacy insurance claims reporting ecosystem, consolidating 100+ fragmented dashboards into a unified, reporting ecosystem with the help of AI. Bridging the gap between real user needs and data points, I reduced time-to-insight by 25% and established a new corporate standard for accessible and user-driven dashboard design.
Lead Product Designer | Timeline: ~12 Months | Tools: Figma, Figma Make, FigJam, MS Power BI, MS Copilot
Skills: UX Strategy, In-depth User Research, Data-driven Decisions, WCAG 2.1 AA Design System, Prompt Engineering
Summary
Static dashboards are everywhere!
My client, a leading insurer, decided to revamp an entire platform of 100+ fragmented, legacy dashboards and disconnected data sources. Reasons: stagnant operations and compliance. I was curious about the first reason for obvious reasons and wanted to dive deeper into the problem area.
7 diverse business user groups lacking defined KPIs, resorted to manual calculations, leading to critical latency in decision-making (including executive decisions). Dashboard implementation team used band aid solutions to provide immediate outcomes that hindered scalability and appropriate decision making.
I led a cross-functional transformation to replace reporting chaos with a unified intelligence layer. By auditing data at the SQL/DAX level and conducting deep-dive KPI validation using in-depth user research, I bridged the gap between raw data marts and user goals. I architected a three-tiered dashboard ecosystem (Executive, Manager, Analyst) underpinned by a custom, inclusive design system.
To solve executive "data noise," I integrated a Copilot-driven summary layer using prompt engineering for specific user groups. I engineered custom prompts to bypass raw numbers, instead delivering high-level reasoning and actionable focus areas.
AI-Generated Executive Insight Example: "Spike in pending claims due to regional licensing delays." (9 words - quick focus area!)
It wasn't simple as it looks, diving deep into the business context and data context had real challenges which I will uncover in the following sections.
The Impact
25%
Increase in Time-to-Insight
By simplified Information Architecture Design
3.7/5
Usability benchmark achieved for legacy reporting
By iterative Usability Testing Sessions
2 Hours
Instead of 2 Days spent in analyzing KPIs per 10 dashboards
By using Copilot to analyze the discovered KPIs
I interviewed 20 users, 7 focus groups as per the claim lifecycle groups. I analyzed 10K+ data columns and my design strategy resulted in a 30% increase in claims processing visibility and eliminated manual KPI calculation for 50+ business users.
The Problem Space
Problem Area
Important decisions including the executive decisions were delayed due to the slowness in reaching the correct data points by the users. The legacy reporting system did provide insights but was lacking complex calculations KPIs as per the user personas and not to mention the distributed data sources which further even confused the users while requesting the complex KPIs during dashboard creation.
After 8 initial interviews (total 20), I found two major problems:
1. The KPI clarity Problem
It wasn’t that the teams were less productive but the reporting system in the entire claim lifecycle had decision making latency issues from the end-user’s side.
ven though the business users were best at their jobs, they were not satisfied with the dashboards they used. At first, it looked like the reports were at fault because they did not have the correct KPIs. But further in-depth contextual inquiry revealed that the business users were unable to translate their goals into the required KPIs for their reports to the implementation team. This was a major eye-opening problem. For instance, a user mentioned downloading the report data and using their own calculations to get to the data point to make business decisions. They did not clearly know how to explain the KPI or their goal.
I asked them in-depth questions around what they did with the data, why they needed that data point and how they used it, calculated it and why?
“I was not sure how to ask for this metric I calculate because it involves many permutations and combinations to say it is the right one.”
- A Senior Claims executive
2. The Calculator Problem
One particular business user mentioned using a physical calculator to feel assured about the savings value for the month. She mentioned that the data is not updated to get the latest value and on the op of that she needs to pull it rom 3 different sources.
This was surprising to me!
Me: “If you do not find the savings value on any of these dashboards then where do you find it? How do you know what it is?”
User: “I always find the savings value on my calculator!”
I asked them in-depth questions around what they did with the data, why they needed that data point and how they used it, calculated it and why?
Apart from this the data sources were fragmented (and disconnected at peak hours) not just the reports. Our team created a unified data source or data marts to serve this purpose. On the top of that DAX queries were written to pull the right KPIs after validating those and interpreting and articulating those from the users.
The interviews truly helped in uncovering the real problems and root causes of the legacy system usage pattern and what it was lacking; which were clearly unseen all these years.
The Problem Space
The Decision-First Mapping
I created a decision mapping as soon as I found out the roles and their hierarchies in the organization and their critical business goals. I added potential solution decisions to ensure we are setting the right expectations from the beginning.
Role | Business Goal | Critical Decision Area | Solution Design |
|---|---|---|---|
Executive
| Root Cause Analysis | "Is our risk exposure within acceptable thresholds for Q4?" | AI-Powered Summary: A Copilot-driven narrative highlighting "Loss Ratio" spikes without requiring manual drill-downs. |
Claims Manager | Operational Velocity | "Where are the bottlenecks in the claim lifecycle causing latency?" | Exception-Based UI: High-contrast symbols flagging claims that have exceeded "Mean Time to Settlement" (MTTS) targets. |
Data Analyst | Portfolio Health | "What specific attributes (e.g., region, adjuster) are driving the spike in reserves?" | Interactive Drill-Downs: 1,000+ attribute accessibility via cross-filtering, powered by optimized DAX queries. |
Mental Models
System Evaluation Diagram
Before: No user goals considered in the reporting system
After: User aligned report design (goal and role based, hierarchical design)
Power BI's complex calculation were backed by user goals, their actions and business decisions
Executives now walk into board meetings with the insight already surfaced — no requests raised, no data exported, no waiting. The meeting no longer depends on whether the report is ready. It always is.
Managers now navigate directly to role-specific KPIs without waiting for data to load or reports to be built. Monday mornings changed.
Analysts now move through data the way they think — by role, by goal, by decision. The notes page they kept open to remember filter combinations is closed. They haven't needed it since launch.
AI Solution
Executives weren't asking for AI. They were asking to stop drowning. When one executive said "I cannot go to 10 different dashboards to get numbers, I need a quick insight on what I need to work on in the board meeting" — that wasn't a feature request. It was a design brief. The Copilot summary layer was the answer to that brief, not a technology experiment. Power BI's complex calculations were backed by user goals, their actions and business decisions. I mapped user goals by analyzing the distributed data tables and created goal oriented KPIs.
The Discovery: Validating the Truth
The Process
I conducted 7 focus group sessions in two rounds. The first round was all about gathering the pain points around report usage behavior and the second one was about understanding KPIs they use today vs what KPIs they need. It was not straightforward. I had to realign the plan to elicit the actual KPI needs by asking questions about their business goals day in and day out with respect to the reports and business values.
Why focus groups: Because a group of people used the dashboards and I wanted to see the possibility of reducing the redundancy and create a one stop solution as per the persona needs and role hierarchies.
The KPIs were redesigned for the users and a third round to validate the KPIs and to order them as per priority or criticality of use took place.
“I use this report for the claims volume KPIs and I know other people using it but from other dashboards.”
“I use the custom created report from a folder and the other day I do not know which one to open. It is exhausting!.”
The Logic Bridge
This was how the KPIs actually transitioned from user goals, acknowledging the manual process and getting rendered in the reporting system. I used SQL to audit the data transformation layer, ensuring our design was grounded in technical reality, not just user assumptions.
Automated the 'Calculator Gap' by translating manual user workarounds into scalable DAX logic
Risk Mitigation
To mitigate the risk of rejection during the transition from legacy spreadsheets, I conducted iterative usability testing with 15 power users. This led to a 40% reduction in 'time-to-insight' and ensured the KPI logic matched the mental models of the actuarial team.
The usability testing helped in validating the navigation structure, person-driven KPIs and the overall time to insight time.
The System: Inclusive Intelligence
The Inclusive Design system
The existing dashboards had no consistent design system. The colors were all over the place. Due to the inconsistent categorization of the data points a data visualization as simple as a pie chart was not only overused but had a rainbow like structure. The more the data points the more the colors used. The more confusing and challenging it was for aged users and color blind users struggled. The cognitive load was high due to inconsistent visualizations that looked forced to get all data points and the data story was a big miss.
I had to come up with an inclusive design system for these users.
It wasn't for aesthetic reasons but the new design system reduced the cognitive load and was comfortable used by the color-blind and aged users.
Old Design System: Rainbow Dashboard with high cognitive load, inaccessible
“I do see the pie chart has 10 legends but now I cannot differentiate between the dark and light shades. I hover over and then I discover the split.”
- A color-blind user
Designed to show all data
New Design System: The Accessible System with Strategic use of color, high contrast, and status for risk
Color blind users were now not only able to differentiate between the color contrasts but also found it intuitive to understand the data with the simplified visualization approach.
In post-launch feedback, the same user who couldn't distinguish between chart segments now described the redesigned dashboards as 'finally readable.'" If you have that quote, use it. If you don't, paraphrase the outcome.
Designed to tell a story to the relevant user with required information to achieve their goals
Along with calculating measures and calculated columns, DAX was used to create symbols and render card designs that stand out. Row Level Security was implemented to ensure the right personas view the right dashboards and reports.
The Inclusive Color Palate
I designed the color palate by running combination tests and how colors worked together. I used Copilot to validate the final palate to ensure that all combinations are color blind friendly.
Note: The below combinations need to be used in a certain way. Some can only be used for bar charts and not for pie charts. Also, all colors follow the WCAG 2.1 AA standards.
The Design Strategy
3 pillars of my design strategy were:
-
Provide a unified navigation for using the dashboards
-
Provide the KPIs that mattered the most on priority
-
Keep a door open for people to self-serve or request new views
Owning the product implementation direction
I owned the data story and influenced the way data marts were built - based on the priority reports and the way the reports information architecture was built across the 7 user bases with hierarchy of use.
By stepping out of the "Design" silo and into "Product Strategy," I successfully influenced both the Data and Business teams. My points were validated: the Data Marts were adjusted to follow User Goals, and the top-priority KPIs were built into the first release. This shift ensured the final product wasn't just a "new look" for old data, but a functional tool for business growth.
The "AI" Philosophy
Users, especially executives complained about cluttered and data heavy visuals. They had business decisions to make which would get delayed when just because they could not find the right data point for their argument.
Copilot was a feature that came with Power BI and was well practiced by me and my team. It looked like executives would follow this scenario in real life: They had goals, they would make some decisions looking at the dashboards and then take an action to impact the business in some way. To make it simple and for them to look at what data needs attention in the given report context, using executive AI summary and quick question answer chat felt like a helping hand to help them look at the obvious and also focus on problem areas.
The Showcase: The Tiered Ecosystem
Executive Summary Dashboard
This dashboard was created by targeting an executive user who need s quick summary of what is happening in the claims world and if there are any standout insights that need attention.
Following are some important design decisions made during the process.
High-level KPI cards with month-over-month (MoM) variance
Architected a 'Primary Health' layer using standardized MoM variance indicators. This allows Executives to identify high-risk anomalies in under 5 seconds without manual data mining
A dedicated natural-language box for automated insights
Bridging the Insight Gap
Integrated an AI-driven summary layer that translates raw fraudulent claim rates into actionable monitoring recommendations, reducing the cognitive load of data interpretation by 30%
Clean, tiered data table with cross-filtering capabilities / Structured Drill-Downs
Developed an interactive hierarchy mapping 1,000+ attributes into a regional performance matrix. This provides the 'Analyst' tier with the granular root-cause data needed for settlement adjustments
Profitability Analysis Layer
Engineered a dual-axis visualization to track the relationship between Claims Paid and Premiums Earned. By calculating the Loss Ratio in real-time, I provided Executives with a direct indicator of portfolio health, moving the needle from 'data tracking' to 'underwriting strategy
A side-panel conversational AI /Conversational Intelligence
Engineered a Copilot interface to handle complex ad-hoc querying. I designed the conversation beyond 'what is' and added depth to the conversation to show 'why and what can be'
The Outcome: Business Impact & Reflection
Decision Velocity
Reduced average "time-to-insight" by 25%, allowing executives to identify and address loss-ratio spikes in real-time rather than at the end of the fiscal month.
Cognitive Load Reduction
Improved insight discovery by 35%, enabling users to identify insurance risks without manual data mining.
Standardizing Quality
Established a new usability benchmark (3.7/5) for enterprise reporting, surpassing legacy satisfaction baselines.
Operational Efficiency
Executed a consolidation strategy that decommissioned 33% of redundant legacy reports, reducing maintenance overhead.
It was quite a platform. The one thing I keep going back to is the way users surprised me and my strategy which made them best at achieving their goals. I learned how important it is to advocate for the right data for our users. Advocating for users does not end when a feature is launched in the favor of users but also when the right data and calculations are in place. This was one of the most complex projects of my career — not because of the technology, but because the real problem kept hiding behind the obvious one. I walked in thinking the dashboards were broken. I left knowing the users had never been given the language to ask for what they needed. What I learned — and what I'll carry into every data-heavy project — is that advocating for users doesn't end when a feature ships. It extends into the data layer, the calculation logic, the way a KPI is defined before it ever reaches a dashboard. That's where the real design work was. That's where it always is.