Context

This project focused on embedding an AI assistant into an existing enterprise support portal used for ticketing and knowledge management.

The platform already included dashboards, structured knowledge content, and escalation flows within a multi-tenant SaaS environment.

The challenge was not to add AI as a feature, but to integrate it into the system without compromising structure, predictability, or trust.

Brief & Objectives

The goal was to reduce unnecessary ticket creation by strengthening self-service through AI-assisted knowledge retrieval.

The experience needed to remain:

• Structured and traceable

• Consistent across tenant configurations

• Aligned with backend and data constraints

The focus was clarity and system coherence rather than conversational novelty.

My Role

I led the interaction and UI design of the AI assistant across:

• Conversation model definition

• Escalation logic

• Knowledge Base refinement

• Visual hierarchy and layout exploration

• Failure states and edge case mapping

• Engineering alignment

The work required balancing conversational flexibility with enterprise-grade structure.

Designing AI as Part of the System

Embedding instead of overlaying

The assistant was integrated into primary support entry points: homepage search, knowledge consumption, and escalation triggers, reinforcing existing workflows rather than introducing a parallel chat layer.

A structured interaction model

The experience evolves in stages:

• Homepage: single query + compact response

• Escalation: transition to a dedicated workspace

• Workspace: two-pane layout (conversation + structured resources)

This structure improves scannability, traceability, and validation of AI output. The intent was not to simulate human conversation, but to support structured reasoning.

Making AI legible

Enterprise users need clarity about where answers come from.

To support this:

• Responses are separated from structured resource references

• Related sources appear in a dedicated panel

• Citations link directly to articles

• Content can be previewed side-by-side

Transparency was treated as a primary design principle.

Rule-based escalation

Escalation was designed to be measurable and predictable.

After a defined number of exchanges, the assistant prompts for resolution confirmation. If unresolved, ticket creation is suggested through a clear transition.

Within the Knowledge Base flow, existing feedback mechanisms were leveraged before escalation.

This reduces premature ticket creation while respecting real user behavior patterns.

Designing within constraints

The system operates in a multi-tenant environment with variable branding and data completeness.

Design decisions accounted for:

• Customer configuration variability

• API and data limitations

• Explicit capability constraints (e.g., non-multimodal AI)

UI patterns avoided implying unsupported functionality.