AI Assistant
Designing Intelligent Escalation in SaaS Support
ROLE Product Designer DURATION 10 weeks STATUS Shipped YEAR 2024
The Challenge
Support teams need automation that feels capable without creating dead ends. This project explored how an assistant can clarify intent, show confidence, and know when to introduce a human expert.
The Outcome
A calmer escalation pattern that keeps people oriented, puts handoff in context, and gives agents the detail they need before they join the conversation.
Fallback Conversation Paths for Unresolved Queries
When the AI couldn’t resolve a query, the user hit a dead end. No path forward. No human escalation.
Why This Mattered
A support bot that says “I can’t help with that” and stops is worse than no bot at all. It creates false hope and then abandons the user. For enterprise support, an unresolved query can block critical work.
How I Approached It
I designed structured fallback paths: a route for recognized edge cases and a seamless handoff to Salesforce Live Chat for unresolved queries, with the full conversation context pre-loaded for the agent.
Intuitive Escalation with Clear Design and Progress Status
Users didn’t know whether they were talking to a bot or a human — or when that transition happened.
I introduced a persistent status indicator at the top of the chat showing “AI Active”, “Transferring to Live Support”, and “Agent Connected”. Each state had a distinct color and label, plus an estimated wait time for live connections.
Clear Cross-Communication with Escalation Color Paths
The handoff between AI and live agent was invisible to the user — and confusing for agents.
I designed a color-coded conversation history agents see in Salesforce: AI messages in one treatment, user messages in another, and a summarized “AI Attempted” section at the handoff point. The summary extracted the topic, the user’s core question, and what solutions had already been attempted.
Projected Outcomes
The AI Assistant launched in late 2024 with full Salesforce integration. Early monitoring showed meaningful changes across support volume and escalation quality.
• ~40% reduction in support ticket volume — the bot resolved queries that previously required human agents.
• 65% faster resolution time for escalated cases — agents had full conversation context at handoff.
• 78% containment rate for tier-one queries — above the original target.
• ~30% fewer duplicate tickets opened during live-agent wait time.
OTHER PROJECTS