A founder-dependent support desk, rebuilt as an AI ticketing system

38,000

tickets migrated into a structured knowledge base

Migration scope is documented: 38,000 tickets / 85,000 messages / 123,000 data points. Automation is a Target KPI — 80% of routine queries — not a realized result: the engagement is active (Phase 2), deploying against the 38,000-ticket backlog, so a final deflection rate is not published yet.

The problem

A support operation buried in 38,000 historical tickets and 85,000 messages, with the CEO acting as the manual translation layer between scattered shadow systems.

How it worked before

Support agents granted concessions through unstandardized channels. A heavily customized vTiger CRM had drifted into a dozen shadow systems, and onboarding knowledge was scattered across Mizito. Routine answers depended on the founder.

What we built

A distributed-agent ticketing system with a semantic memory layer (PGLite / pgvector). 'Smart ticketing' surfaces similar resolved tickets at submit time and is built to deflect routine queries before they reach a human.

What changed

The support desk moved from founder-dependent and improvised toward an AI-augmented model, with the historical ticket corpus turned into a searchable knowledge base instead of tribal memory.

System shape

Inputs

38,000 historical tickets / 85,000 messages

Processing

Semantic memory + classification over a vector store

Outputs

Pre-submit ticket deflection + suggested knowledge-base answers

Method

1Map: AI-led audit of the support operation to surface the CEO bottleneck and shadow systems
2Structure: a 4-level documentation framework; 38,000 tickets moved into a structured knowledge base
3Implement: distributed-agent architecture over a PGLite / pgvector semantic memory layer
4Adapt: an exception-handling rule protocol so the system retrains on edge cases instead of escalating them

Constraints

Persian / RTL language throughout
123,000 distinct data points migrated securely
Local data-security compliance (AFTA certification)

Our role

Transformation lead and strategic advisor to the CEO

Why this is anonymized

Client identity is withheld under engagement confidentiality. The metrics are real and verifiable in a private reference call.

Have an operational document you want analyzed?

We do not run standard sales calls. Send your most complex operational document and an engineer will return a clear model of what automating it is worth.

Request an operational diagnostic

All work