Woohee Byun




The Machine


PROBLEMThe Machine, an AI-powered operating system built for the Stagwell network, launched its MVP in October 2025. However, the research identified a discrepancy between the current design and the users’ usage patterns.


OUTCOME
The team has addressed the gap by translating research findings into actionable tasks, including a new information architecture, redesigned navigation, and automated, actionable suggestions for users.
CLIENT
Stagwell


TIMELINE08.2025 - 06.2026


SCOPE
Product Experience, UX, AI Interaction





FINAL DESIGN

Left-Side Vertical Navigation


Persistent left-rail structure to improve wayfinding for deep hierarchies and support faster access to core areas of the product.


Actionable Feed Cards


Surfaces personalized, actionable cards based on predicted next steps, recent activity, and role-relevant tasks. It helps users quickly resume work, discover relevant agents, and move directly into high-value actions.

Unified Global Chat


Unified – previously separate – chat experiences help users understand where to go to ask questions, improve continuity across workflows, and reinforce chat as a central interaction model.







CONTEXT

What is The Machine?

The Machine started from the pain point that information was “scattered” across platforms, marketing tools, and systems that didn’t communicate with or integrate. The operating system connects brand strategy, creative production, and media performance, enabling new campaigns to account for past learnings.







PROBLEM

MVP launched – but assumptions didn’t match behavior

MVP shipped in October 2025 to a small group of Stagwell employees, including Code & Theory and Crispin. For the first time, we’ve heard from a group of users across the network about what they think.






REFRAME

Users knew what the tool was, but not how to use it

Across sessions, we found that users could articulate the product's value but consistently struggled to act on it. Navigation patterns revealed that users were backtracking, skipping core features, and resorting to workarounds — signaling a gap between the mental model and the actual architecture.


How might we redesign the Machine to grow with users, 
surface the right actions proactively, 
and become the connective tissue of their work?

🧭

Navigation that scales with complexity
As users take on more projects and integrations, the overall interface needs a structure that makes complex IA feel simple.
🚥

Insights that tell you what to do next
Project data should speak for itself. The interface should translate what’s happening into clear, actionable next steps without users having to dig.
💬

A chat experience worth coming back to
Chat is already where users spend their time. It needs to be powerful enough to hold context, surface outputs, and support the way people actually work.




COMPETITIVE AUDIT

Learning from the field before designing for it

Before exploring solutions, we looked at how other AI platforms structure their navigation. The patterns we found shaped five distinct models — each testing a different hypothesis about how The Machine's nav should grow with its users.


ClaudePerplexitySana AI




EXPLORATION

Finding a structure that scales without overwhelming

With the hierarchy defined, the team explored different navigation models — each making a different trade-off between simplicity, discoverability, and scalability. The goal was to find a structure that felt lightweight on day one but could hold up as projects and integrations grew.






SOLUTION 1

Unified Global Chat

Users were surprised by chat behavior that felt like automated “thinking” and found streamed or verbose responses hard to scan and control. Users expected clearer controls and more scannable output, similar to those of familiar consumer chat models.



 Separated chats confuse users and scatter the history and context. Unified chat keeps context and memory in a single place.



SOLUTION 2

Left-rail Navigation

Users were surprised by chat behavior that felt like automated “thinking” and found streamed or verbose responses hard to scan and control. Users expected clearer controls and more scannable output, similar to those of familiar consumer chat models.



 Breadcrumb is used for site navigation, which prevents users from easily accessing other pages. Side navigation, a common pattern for LLM, is introduced to support broad, growing IA.



SOLUTION 3

Actionable, custom-populated cards

Users were surprised by chat behavior that felt like automated “thinking” and found streamed or verbose responses hard to scan and control. Users expected clearer controls and more scannable output, similar to those of familiar consumer chat models.


 Plain text gives limited information about the project. Card surfaces in-depth information with actionable items to feed and get better results over time.














OUTCOME & IMPACT

A foundation built to grow with the product

The redesign now aligns with how the backend actually works — making The Machine a single source of truth where users can connect their tools, return with intent, and always know what to do next. Rather than a place to check in, it becomes a place to act.
©WOOHEE2026