Usage of SMG’s Knowledge Agent was declining. Customer feedback identified two problems: the dialog-based interface constrained sustained analysis, and unreliable answers reduced trust.
As a senior product designer on the AI feature team and design-system lead, I redesigned both the experience and its underlying architecture. The work transformed a service-coupled dialog into reusable conversation components and responsive layouts that could support AI across the product.
SMG is a customer-experience platform used by multi-location brands to collect and analyze customer feedback. Customers include: McDonalds, Chik-fil-A, Wendy's, Academy Sports, Crunch Fitness, Dominos, Pizza Hut, and more.
Analyze Usage and 1,400 Chats→ Define Requirements → Redesign the Experience → Refactor the Architecture → Validate in Beta
The original Knowledge Agent opened from a floating action button into a small dialog. Long responses frequently exceeded the available space, forcing users to scroll within a constrained workspace.
Analysis of 1,400 conversations using LangChain revealed additional needs, including filter scope, chat history, visible context, interruption controls, and recoverable error states.

We replaced the dialog with a persistent conversation panel that gave users more room to analyze data without losing access to the underlying page.
The new workspace accounted for the full conversation lifecycle: starting, loading, failing, stopping, resuming, restoring context, and returning to previous chats.

The original implementation embedded service dependencies inside a single feature component. We separated the interface into smaller conversation components and moved service-specific logic downstream.
This allowed the same foundations to support the Knowledge Agent, survey conversations, and future chat experiences.

AI Chat and Survey Chat served different purposes but shared many interaction patterns. We created a common workspace component that both experiences could configure without duplicating their foundations.

We placed the agent in a responsive push-and-pull layout so users could analyze a conversation alongside the underlying page. Shared layout components coordinated navigation, page content, and the conversation panel instead of making them compete for space. The architecture included ForgePage, ForgeMainContent, ResponsiveMainAsideLayout, PageWithPanel, and PageFixedLayout.

We designed the conversation model to persist across product surfaces while preserving context and consistent behavior. This established a foundation for agents that could support workflows throughout the application rather than living on a single page.

Within one month, we designed and released the new Knowledge Agent experience through a controlled LaunchDarkly beta. The release introduced a persistent workspace, clearer conversation states, reusable components, and a responsive page architecture while reducing implementation debt.
The resulting system gave SMG a shared foundation for extending conversational experiences across the product.
