AI Knowledge Base
Introducing AI-assisted self-resolution for Tier Light users, connecting an existing knowledge base to new AI interactions and human support when issues remain unresolved.
Client
Confidential enterprise support SaaS
Role
Product design
Year
2026
Status
Shipped · anonymised
At a glance
- Business goal
- Enable AI-assisted self-resolution for Tier Light users
- Product intervention
- Introduce AI assistance into the existing knowledge-base experience.
- Design decision
- Propose a support ticket after three cumulative escalation events if the issue remains unresolved
Making self-resolution work for Tier Light
Tier Light users have more limited access to human support. The brief was to introduce AI-assisted self-resolution into a platform that did not yet offer AI. I designed the new AI-assisted experience around its existing knowledge base, article consultation and exit feedback.
That made trust a practical requirement: users needed to inspect the evidence behind an answer, judge whether it helped, and reach human support when self-service remained unresolved.
One unresolved support journey
The walkthrough follows one authentication problem from a question to a source-backed answer and a ticket offer within the conversation. It illustrates one route through the experience, not every escalation event.
The new AI chat connects to the existing knowledge-base journey, so users can move between answers and articles without starting again. AI exchanges and negative article ratings contribute to the same count, rather than separate funnels.
Trust comes from inspectable knowledge
Self-resolution depends on more than a plausible response. Users can check the cited article, ask a contextual follow-up, and receive a streamed answer without losing the source they are evaluating.
Recognising unresolved moments
The platform already asked users whether an article was helpful when they left it. I used that existing interaction as a natural escalation point and connected it to related-article suggestions.
A negative article rating contributes to the shared count; simply opening or reading an article does not. Related articles let users continue seeking an answer, with subsequent negative ratings adding to the same count as AI exchanges.
Text alternative: The existing article-exit feedback asks whether the article was helpful. A negative rating contributes one escalation event. Related articles offer another route to self-resolution; opening or reading them alone does not increment the progression. Further negative ratings and AI exchanges can accumulate in any order toward the shared threshold of three.
When self-service should become human support
I proposed a shared threshold of three cumulative escalation events and designed how it connected the new AI interactions with existing article feedback. Each AI exchange and each negative article rating contributes one event. Users can alternate between them in any order.
If the issue is resolved, the journey ends. Otherwise, self-service continues below the threshold. Once the shared count reaches three, an unresolved issue leads to a ticket offer. In a chat-only journey, this happens after the third AI exchange, when the assistant asks whether the problem is resolved.
The system proposes a ticket; the user chooses whether to open it. The handoff carries forward the question, consulted sources and feedback, rather than asking the user to reconstruct the problem.
- Enter support and choose either path
- AI: each exchange adds one to the shared count.
- Knowledge base: read an article or follow related articles. Reading alone adds nothing; each negative article rating adds one.
- Keep one shared count
- Move freely between AI and articles, in any order. Both counted actions contribute to the same threshold of three.
- After reading, continue with related articles or AI without adding to the count unless you rate the article negatively.
- Is the issue resolved?
- Yes, before or at the threshold → end without a ticket.
- No, with fewer than three counted actions → return to either path.
- No, when the shared count reaches three → propose a ticket.
- Propose a ticket
- This is an offer, not automatic creation.
- User opens the ticket
- Ticket created with context
Text alternative: Users can alternate freely between AI exchanges, articles and related articles. Each AI exchange and each negative article rating adds one to the same count; reading an article adds nothing. Resolution at any point, including the third counted action, ends the flow without a ticket. If unresolved below three, the user returns to either path. If unresolved when the count reaches three, the system proposes a ticket. The user must then open it before a ticket is created with the prior context; creation is never automatic.
What I contributed
My contribution was to design AI-assisted self-resolution within an existing support platform, connecting new AI interactions to knowledge-base journeys, resolution checks and conditions for human support.
The shipped design shows how the new AI interactions connect to existing support journeys. The three-event threshold was a design proposal; its effect on ticket volume and self-resolution has not been established here.
The interface shown is a faithful reconstruction of the shipped design. The client identity, realistic data, and brand colours have been changed for confidentiality; the UI, flows, and functionality are unchanged.