Case Study

AI Copilot for Operational Intelligence

An AI assistant built into an existing enterprise platform, so users can ask questions about their data instead of digging for answers.

AI ProductB2B SaaSAnalyticsEnterprise UX
Sunnie AI Copilot docked panel with dynamic suggestion chips, shown alongside the Solargik multisite dashboard

Context

The platform was already changing. New dashboards, like the Loss Analysis view, were giving users a clearer picture of planned versus actual output. As AI tools spread across the industry, we wanted to add an AI layer that would sit next to those dashboards as another way in.

The result was Sunnie, an AI Copilot built into the platform. It helps users analyze data, pull out insights, and work faster without leaving the screens they already use.

My Role

Product Designer. I was the only designer on the project and owned it from concept to launch: product direction, UX, interaction design, and fast iteration with the developers.

What Users Actually Needed

Users weren't asking for a chat. They wanted the platform to be faster and easier to read, and AI was one way to get there. Even with the new dashboards, they were still dealing with:

Buried information
Fragmented navigation
Slow reporting workflows
Hard to find insights
Too many steps to answer simple questions

So the goal was clarity and speed, added as an extra layer on top of the tools people already relied on.

Before: digging through screens
Open Dashboard
→
Find Report
→
Filter Data
→
Cross-Reference
→
Get Answer
With Copilot
Ask a Question
→
Get Answer

From Chatbot Popup to Built-In Copilot

The first version was a floating chatbot. I redesigned it as a panel docked inside the platform, so people could keep working on the dashboard while asking the Copilot for help. That made it feel like part of the daily work.

AI Beside the Workflow

No disruption to existing habits.

Immediate Value

Familiar enterprise interaction model.

Comparison of the V1 generic chatbot popup versus the in-production docked Sunnie AI Copilot panel

V1: generic chatbot popup, floating over the dashboard.

In production: branded, docked Copilot panel with suggested prompts.

What Users Could Ask

The Copilot enabled users to ask questions in plain language and receive answers as text, tables, charts, or insights.

"Compare production between Site A and Site B"
"Why is Site A underperforming?"
"Show all issues from last month by severity"
"Build a chart of downtime trends"
"Summarize anomalies from this week"
"Which assets need immediate attention?"
Sunnie AI Copilot answering a natural-language question about energy production with a structured data table

Proactive Alerts

Some issues shouldn't wait for someone to ask. Sunnie also sends a daily email digest of things that need attention, like tracker stalls or homing errors, with a link that opens the full details in the Copilot.

Sunnie AI proactive email digest listing tracker stalls and homing issues, viewed on a laptop

Key UX Decisions

Docked Side Panel

Use AI while staying inside the platform.

Full Screen Mode

Expand for deeper analysis.

Conversation History

Return to previous findings.

Structured Outputs

Answers can come back as tables and charts too.

Persistent Context

Keep answers connected to product data.

Annotated screenshot of the Sunnie AI Copilot highlighting fullscreen mode, persistent navigation, conversation history, dynamic suggestion chips, and graph output

How We Worked

This project changed how we worked. The developers were using AI-assisted coding, so a classic design-first handoff would have slowed everything down. Instead, design and development ran in parallel.

Design (Product)
→
Test & Refine
→
Build (Engineering)
↻
Design (Product)

Outcome

The Copilot went to production quickly.

For Users

Questions that used to take several screens and filters could be answered with one prompt in plain language.

For the Business

It showed customers, and the team, that the platform was keeping up with the industry. The Copilot became a regular part of demos and conference presentations.

What I Learned

Start From the Problem

Users didn't need "AI." They needed clarity and speed. Building it into their workflow made it a tool they actually used.

Support Existing Habits

Good AI UX works with how people already work before trying to change it.

Design as Strategy

When AI makes building faster, the designer spends less time producing screens and more time deciding how the product should work.

Next Case Study

Energy Loss Analysis & Optimization

View next project →