Self-Portraits: Madeline Kleiner

Great products don’t happen by accident. They’re shaped by a deep understanding of people. Microsoft has over a billion customers worldwide, and experiences must navigate different behaviors, cultures, accessibility needs, and expectations. For over ten years, UX Research leader Madeline Kleiner has been translating customer insights into frameworks and design principles that guide how we build for billions worldwide.

As a holder of over 10 patents in the AI productivity space, Madeline’s work has influenced everything from interaction patterns to long-term experience strategy. Her research helps teams design with greater empathy, clarity, and confidence. In her own words, she talks about how her work supported a recent shift in how Copilot shows up across Office apps.

AI experiences can fail for many reasons. They’re hard to use. They aren’t useful enough. Or they simply don’t feel worth the effort. With each failed interaction, trust gets broken, and we knew it was critical to think about the moment-to-moment product signals that either maintain or interrupt flow. Simplifying the in-app design for Copilot meant creating patterns that feel intuitive inside the tools people already rely on.

I lead a research team that studies how people build trust and momentum with AI in everyday workflows. At the intersection of human behavior and platform-scale design, my work focuses on one question: how does AI become a continuous thought partner?

Ramping up to legacy behaviors

To answer that question, we embedded with customers, following how they scan a page or where they move their cursors. Microsoft 365 is a mature ecosystem with decades of learned behavior, and the smallest changes can disrupt flow and productivity. Undo, Save, and copy & paste between apps are mental models that people almost unconsciously engage with, but when experiences evolve faster than those habits, confidence drops, cognitive load spikes, and people disengage.

We also know people don’t experience apps as isolated tools. They first experience intent, then action, and finally, outcome. What became clear was that Copilot needed to be designed to feel supportive while meeting people in the flow of real work. That meant our insight had to move beyond observation and into shaping product decisions.

Our researchers partnered closely with design, product, engineering, and applied science to test how people would find Copilot and build confidence using it. We explored a wide range of entry points in the Office apps including ribbon, skittle, search, top-of-document placements to understand the tradeoffs between discoverability and decision fatigue.

To get beyond opinions (including our own), we mixed methods and watched behavior.

We built a mouse heatmap tracking tool so we could see where people searched for Copilot when they wanted to use an AI thought partner to achieve their goals. We ran 1:1 sessions to hear how individuals described Copilot in their own words and where uncertainty showed up. We used surveys to validate patterns at scale, especially around confidence and in “I’m not sure what to click” moments.

From insight to a unified system

What we learned was that when Copilot is broken into isolated interactions, people end up re-stating their intent, re-checking context, and re-building confidence every time they switch surfaces. What people wanted was a continuous thought partner. That meant Copilot’s visual consistency had to support cognitive continuity. This was reflected in sessions where people told us that they could relax when they trusted Copilot understood three things:

• What they were working on
• Where they were working
• Why they were working on it

When that trust was established, the language changed. People stopped describing Copilot as “something extra” and started describing it as “with me.” That became our north star: create one predictable Copilot that’s easy to find, then express it contextually without fragmenting attention. In practice, that meant pairing a stable “home base” with in-the-moment guidance and on-canvas interactions, so Copilot can show up where you are, while still feeling like the same Copilot everywhere.