How I Used an AI Copilot to Redesign 100 Enterprise Dashboards (Without Drowning in the KPIs)

Hello, I’m Abhinee. As a Product Designer, I anchor my design work in deep product strategy to drive measurable, data-driven outcomes for complex systems. I hope you find this article helpful.

Photo by Firmbee.com on Unsplash

Enterprise UX design is rarely a blank canvas. More often, it is a rescue mission.

Recently, I stepped into a project as the Lead Product Designer for a major insurance client. The landscape I inherited was a classic enterprise tangle: 100 legacy dashboards overflowing with duplicate reports, conflicting KPIs, and complex workflows. Instead of empowering users, the system was actively hindering them. Executive leaders, in particular, were drowning in data noise, making it incredibly difficult to track high-level business goals or take swift, informed action.

Our team’s mission wasn’t just to redesign the user interface. We needed to realign the entire organization’s KPI framework, streamline role-based workflows, and bring clarity to executive decision-making.

Here is how I navigated the strategic crossroads of this massive redesign, and how I leveraged Microsoft Copilot as a research specialist to scale our UX process.

The Strategic Crossroads: Patchwork vs. From Scratch

After analyzing the organization’s structure and mapping user groups against a complex, multi-phase insurance lifecycle, everything seemed clear on paper. But when I began looking at how KPIs were distributed across the user hierarchy, the real challenge emerged.

Redefining these metrics was like trying to find a needle in a haystack. As an experience consultant, I realized I was standing at a strategic crossroads with two potential paths forward:

1. The Incremental Route: Take the existing, legacy KPIs, validate them with users in their current format, and patch in new ones where necessary.
2. The Ambitious Route: Toss out the legacy assumptions and redefine the entire KPI framework from scratch, mapping them strictly to user personas, role hierarchies, and actual executive goals.

The first option was safer and faster. The second option was highly ambitious and ideal. As a consultant, my responsibility is to advocate for what truly matters to the end user. I chose the ambitious route. But I began the journey from what already existed.

Gathering Ground-Truth Inputs Through Focus Groups

To build a new framework from scratch, I needed a deep understanding of current usage patterns. I began by organizing a series of targeted focus groups.

The decision to leverage focus groups rather than just individual interviews was intentional. I needed to see the friction points in real-time — specifically, how different roles within the same organizational hierarchy interacted with the exact same cross-functional dashboards and with each other on the related work or tasks.

During these sessions, I took meticulous notes on how users relied on existing metrics, where their workflows stalled, and what goals they were actually trying to achieve at each phase of the insurance lifecycle as an individual and also as a team. The dependency of one team or role on another was uncovered in these conversations.

Onboarding the AI Assistant: Security, Iteration, and Guardrails

With focus group insights in hand, I faced a mountain of data. Every phase of the insurance lifecycle contained multiple dashboards, and each dashboard housed dozens of microscopic reports. To synthesize this safely, I brought in Microsoft Copilot to act as my research assistant.

When handling enterprise data, security is paramount. I operated entirely within a secure, client-approved environment. With all proper permissions in place and complete data security guaranteed, I fed the massive, complex Excel sheets containing the client’s current data structures that I created meticulously into Copilot.

I gave the AI deep context regarding the client’s organizational structure and operational functions. Initially, Copilot acted like an overly eager industry expert, returning a massive, idealistic “wish list” of KPIs. It was completely overwhelming.

I quickly realized that AI is only as valuable as the constraints you place on it. I iterated multiple times, establishing strict guardrails that limited Copilot’s focus strictly to the dashboards and operational areas that were within our project scope. To ensure absolute accuracy, I continuously cross-referenced Copilot’s refined outputs with human insurance Subject Matter Experts (SMEs). It was all worth it.

Scaling the Research: 20+ Executive Card Sorting Sessions

The true turning point of the project came during the testing phase. Many of our core dashboards contained upwards of 50 KPIs. If you have ever tried to run a traditional card-sorting exercise with a busy, high-level executive using 50 separate elements, you know it is a recipe for user fatigue and disengagement.

To solve this, I used Copilot to pre-group and categorize the KPIs based on our earlier lifecycle mapping.

This completely transformed the energy of our research. Instead of forcing 20+ senior executives to sort through granular rows of data, we presented them with intelligent, AI-assisted clusters.

The nature of the conversation shifted entirely. We stopped arguing at a microscopic level about individual data points and started discussing high-level strategy. The card-sorting sessions became focused on:

• KPI-specific goals: What does this cluster of metrics tell you about your business unit?
• Actionable insights: What specific operational action do you take when this dashboard metric changes?

Because the AI helped us organize the noise beforehand, it became incredibly clear which metrics were vital and which could be discarded. Ruthlessly removing the KPIs that didn’t make sense was the most critical step in saving executive cognitive load. But I made sure that I was examining every group date and card before making a decision and not letting AI make it for me. Also, I rearranged some groups and ordered them to accommodate the existing mental models of the users. For example, starting off with the obvious KPI groups that were already conveyed to us by the users and not asking irrelevant or noncontextual questions so that users feel comfortable and in the loop with all the decisions.

The Takeaway: AI as an Equalizer in Enterprise UX

By pairing deep qualitative research (focus groups and card sorting) with the analytical speed of an AI Copilot, I successfully transformed a chaotic ecosystem of 100 duplicate dashboards into a streamlined, role-based experience.

For product designers tackling complex enterprise workflows, the lesson is clear: AI doesn’t replace the UX research process; it scales your capacity to handle complexity.

Copilot didn’t talk to our users, and it didn’t make the strategic design decisions. What it did was crawl through the data trenches for me, freeing up my time to lead high-value, strategic conversations with stakeholders and design a system that truly drives business outcomes. I was able to reduce almost 30% of my analysis time during this research sprint.