Cleo
EDI / Enterprise Support

RAG-Powered Past Ticket Intelligence for Zendesk

Ingested historical Zendesk tickets into a knowledge base and built an in-app Zendesk widget that analyses open tickets, cites relevant past tickets, synthesises root cause, diagnostic steps, and resolution — with AI chat for engineers.

RAGZendeskKnowledge BaseVector SearchSupport AI
Build Something Similar

Impact Metrics

-65%
Engineer Research Time
Time spent researching resolution approaches
-40%
First Response Time
Faster initial substantive response to customers
+78%
Resolution Consistency
Improvement in cross-engineer consistency scores
-35%
Unnecessary Escalations
Escalations for previously-solved EDI issues
The Challenge

Institutional Knowledge Trapped in Closed Zendesk Tickets

Cleo's EDI support organisation accumulated enormous institutional knowledge through their Zendesk ticket history — thousands of resolved EDI issues representing tested solutions to real problems. Yet this knowledge was effectively inaccessible: buried in closed tickets that could be searched by keyword but not semantically queried or synthesised.

When a new ticket arrived, an experienced engineer might remember resolving something similar months ago and know exactly where to look. A newer engineer, or one unfamiliar with that specific EDI configuration, would start from scratch — researching documentation, asking colleagues, or escalating unnecessarily.

The result was inconsistent resolution quality across engineers and shifts, longer resolution times, and repeated escalations for issues that had been solved multiple times by different engineers who never shared their solutions systematically.

Key Pain Points

Historical Zendesk ticket knowledge inaccessible beyond keyword search
New engineer onboarding taking weeks to build institutional EDI knowledge
Inconsistent resolution quality across engineers and shifts
Unnecessary escalations for previously-solved EDI problems
Knowledge siloed in individual engineers' memories
The Solution

Zendesk-Native Knowledge Base with RAG Synthesis and AI Chat

We built a specialised RAG system that ingested Cleo's historical Zendesk tickets — including ticket descriptions, resolution notes, engineer comments, and time-to-resolution data — creating a living knowledge base that makes this institutional knowledge semantically searchable.

We then built a Zendesk app (widget) that sits inside the Zendesk interface. When an engineer opens a ticket, the app automatically analyses the open ticket, retrieves the most relevant past tickets, and presents a RAG synthesis that includes: root cause analysis, diagnostic steps to verify, and resolution steps to take action — all citing specific past tickets as evidence.

The system also includes an AI chat interface within Zendesk, allowing engineers to ask follow-up questions directly against the knowledge base. Engineers can drill deeper into specific aspects of the resolution, ask about edge cases, or get clarification — all powered by the historical ticket knowledge. The system continuously improves as new tickets are resolved and ingested.

Our Approach

1
Historical Zendesk ticket ingestion pipeline building a living knowledge base
2
Semantic embedding of ticket descriptions, resolution notes, and engineer comments
3
Zendesk app (widget) providing in-context AI assistance on open tickets
4
RAG synthesis generating root cause, diagnostic steps, and resolution actions
5
Citation of specific past tickets with confidence scoring
6
AI chat interface for engineers to ask follow-up questions from ticket knowledge

Key Features Delivered

Zendesk ticket ingestion pipeline building a living knowledge base
In-app Zendesk widget analysing open tickets against past resolutions
RAG synthesis: root cause, diagnostic steps, and resolution actions
Citation of relevant past tickets with confidence scoring
AI chat for engineers to ask follow-up questions from past ticket knowledge
Ranked resolution candidates with evidence from specific past tickets
Continuous learning pipeline as new tickets are resolved
Analytics dashboard tracking recommendation adoption and accuracy
Technology Stack

Built With

LangChainOpenAI EmbeddingsPineconeFastAPIZendesk APIZendesk App FrameworkPostgreSQLRedisDockerAWS
Results

Outcomes Achieved

By making Cleo's historical Zendesk ticket knowledge semantically accessible through an in-app widget with RAG synthesis and AI chat, the system transforms every engineer into an experienced one — eliminating knowledge silos and providing actionable root cause, diagnostics, and resolution steps for every open ticket.

-65%
Engineer Research Time
-40%
First Response Time
+78%
Resolution Consistency
-35%
Unnecessary Escalations

Want Similar Results?

Let's discuss how we can build a similar solution for your organization — with the same certified quality and production-grade delivery.