80% of Level 1 tickets — password resets, access requests, 'how do I' questions — could be auto-resolved, but they still hit the service desk queue, eating 60% of agent capacity and frustrating users with wait times.

Utilyx models self-service as an executable workflow: the AI chatbot resolves common issues from the KB and catalog, and escalates to a human with full context only when confidence or policy requires it.

The workflow is structured as a decision tree with 3 branches (Auto-resolve, Catalog fulfillment, Escalate to agent) and 6 steps end-to-end. Each step is executed by a named actor (End user, AI Agent, call-connector, KB search, email service, Service desk agent) and every action is timestamped.

• Step 1 (Start — End user): the user opens the self-service portal or chat widget and types their request in natural language. The session captures: session_id, user_id, query_text, language, channel (portal/chat/Teams).

• Step 2 (AI Agent): the ai-agent classifies the intent (incident, request, how-to, feedback) and extracts entities (target system, error code, asset). It checks the user's entitlements via sql-query (does the user have permission to request this?).

• Step 3 (Routing): three branches — Branch A (how-to intent, KB match > threshold): auto-resolve by surfacing the KB article and confirming the answer helped; Branch B (request intent, catalog item exists): trigger catalog fulfillment via call-connector (provisioning API); Branch C (incident intent, or confidence < threshold, or policy requires human): escalate to service desk agent with full chat transcript attached.

• Step 4A (ai-agent): the chatbot presents the KB article in the user's language, asks 'did this solve your issue?' and if yes, closes the session and logs a deflected ticket (no agent involved). If no, it re-classifies and either tries another KB article or escalates to Branch C.

• Step 4B (call-connector): for catalog requests (e.g. 'I need a new laptop'), the call-connector triggers the provisioning API and the user receives an email confirmation with ETA — same as the catalog workflow, but initiated by chat.

• Step 5C (Service desk agent): the agent receives the escalated ticket with the full chat transcript, the AI's classification, the KB articles already surfaced, and a suggested resolution. The agent picks up from there — no re-explaining by the user.

• Step 6 (Closure): on auto-resolve, the session is logged in deflected_tickets with resolution_source='chatbot'. On escalation, the ticket follows the incident or request workflow. Weekly, the AI reviews deflected sessions to identify KB gaps (sessions that escalated after KB miss) and flags them to the KB owner.

Teams that adopt Utilyx report 80% of Level 1 tickets auto-resolved by the AI chatbot without agent involvement, agent capacity freed by 60% for higher-value L2 work, and end-user satisfaction up 30% because answers come in seconds in the user's own language.

The chatbot escalates with full context so the user never re-explains, the catalog fulfillment via chat removes the portal navigation step, and the weekly KB-gap review turns missed deflections into new articles — making the chatbot smarter every week.