Webex Control Hub AI Assistant
I led the evolution of Webex Control Hub’s AI Assistant from a simple help bot into a full conversational layer—adding threaded chats, Smart Search, and AI-powered report analysis—which boosted monthly usage by 120%, proving that thoughtful system-level design can directly drive efficiency and adoption.
Webex Control Hub is the single pane of glass where IT administrators deploy, configure, and monitor the entire Webex Suite—Messaging, Meetings, Calling, and Contact Center. As the surface area of the platform grew, finding settings, interpreting reports, and diagnosing issues became increasingly time‑consuming.
In June 2024 the team introduced the first cut of the Control Hub AI Assistant (CHAI)—a help‑only chatbot. I joined just after that launch with a mandate to evolve CHAI into a strategic layer
I partnered with PM, engineering, and data science to build and ship the next chapters on our roadmap
Thirty‑five percent of setting queries in Control Hub returned nothing, forcing admins to hunt manually. I crawled configuration APIs, generated vector embeddings, and paired them with LLM intent detection to match natural‑language queries to precise setting pages.
A “tunnel” affordance surfaced suggested follow‑up questions that opened CHAI for deeper help. The update slashed zero‑result searches by 80 % and now drives 14 % of total assistant entry points.
In Control Hub, we provide reports, but getting insights from the reports required downloading CSV files and manual analysis, which is time-consuming and inefficient.
I embedded “Ask AI” shortcuts directly into each report row. Clicking the sparkle icon button reveals a shortlist of context‑aware suggested questions—or admins can type their own.
Either path launches CHAI, which pulls the underlying report data, answers in plain language, and, when helpful, renders sparklines or bar charts inline. Admins can now surface insights without exporting a single CSV.
Research shows that admins expect AI help most when incidents strike, yet existing dashboards bury the signal in noise.
CHAI will cluster impacted users by root cause, summarize findings in natural language, and recommend next steps. I am prototyping multi‑signal embeddings and partnering with the Devices BU to embed CHAI directly on device pages.
I am also working on the northstar vision of using AI Assistant to process complex data and troubleshoot meeting and calling.
Trust through Transparency – show data citations, confidence, and fallback channels.
Progressive Disclosure – reveal complexity only when the admin asks for more.
System First – build reusable conversation + card components for rapid feature onboarding.
Outcome‑Aligned – measure every release against admin task completion and ticket deflection.
Business alignment turns design into growth—framing each feature in terms of ticket deflection made exec buy‑in easy.
Explainability is the new usability for AI: admins only trust answers they can audit.
