Product
AI Support Pipeline
A full pipeline: it classifies tickets, flags duplicates, routes each one to the right team and person, builds an FAQ automatically, and answers questions about the ticket history.
Pilot project · AI Support Operations
Dispatch
Built and validated an AI-powered support-routing prototype for a US fintech, after mapping exactly where tickets were getting stuck. It covered classification, deduplication, routing, and workload balancing across 2,000+ monthly tickets.
Fractional
product leadership
2025
Overview
Validated AI routing in one week
Client & context
US Fintech
Fast-growing company processing 2,000+ support tickets per month, with a 10-person support team spread across multiple specialized squads.
My Role
Fractional product leadership
One-week engagement: identified the bottleneck, designed the system, built a working prototype with OpenAI API, and validated a pilot scenario with historical ticket data.
The problem
2,000 tickets a month, all routed by hand
Manual triage
One person sorting everything
Every ticket had to be manually categorized, then assigned to the right team, then balanced across agents. One person spent a significant chunk of their time just routing, not solving customer problems.
Downstream waste
Duplicates, imbalance, no self-serve
Duplicate tickets clogged queues. Some agents were overloaded while others sat idle. Many questions already had answers buried somewhere, but there was no FAQ, no knowledge base, nothing.
Product loop
A four step pipeline for automatic ticket routing
Each ticket passes through a multi-step AI pipeline: classify, deduplicate, route, and balance, then the same data feeds a live FAQ and a conversational query layer.
1
Classify and deduplicate
OpenAI API scans each incoming ticket, assigns a category from the company's taxonomy, detects near-duplicates, and flags tickets that already have known answers.
2
Human review fallback
If the model's confidence score for classification drops below 95%, the system holds the ticket in a queue for a human to check before it moves on.
3
Route and balance workload
High-confidence tickets are distributed evenly across specialized squads based on current load, preventing one person from drowning while others wait.
4
Generate FAQ and answer queries
Recurring patterns automatically update a structured FAQ. A conversational tool lets the team query ticket history directly.
Key decisions
Three choices that made the automation work
Mapping bottlenecks
Fixing triage before response
We observed support workflows end-to-end and realized the main delay was sorting and routing tickets rather than writing answers. We focused our scope on automating triage first.
Evaluation dataset
Testing accuracy on past data
Rather than guessing if prompts worked, we built a dataset of 200 past tickets and ran test scripts to measure classification accuracy and eliminate errors before launch.
Building agent trust
Designing transparent reasoning
To help support agents trust automated decisions, routing rules displayed clear citations showing why a category was chosen, along with easy one-click manual overrides.
Standalone pilot
Validating before live integration
We built and validated a working prototype on historical tickets in five days. Testing standalone allowed us to refine routing logic before connecting to live customer support channels.
Results
Proven time savings in pilot testing
15h / week
Manual triage eliminated
Saved an estimated 15 hours per week by routing tickets automatically in seconds instead of three minutes of manual sorting per ticket.
92%
Classification accuracy
Automated categorization matched existing company guidelines in simulation runs across 10,000 historical support tickets.
30%
Duplicate tickets flagged
Flagged 30% of incoming tickets as repeat requests, reducing queue clutter and automatically generating self-serve FAQ answers.
Retrospective
What I learned
Prototype standalone before integrating live systems
Building a working prototype on historical data is faster and safer than modifying live software on day one. Showing concrete results on past data makes team buy in much easier.
Treat internal tools with the same product rigor as consumer apps
The biggest risk with AI automation is adoption. Involving support reps from day one to design routing rules ensures the tool fits real daily workflows and gets enthusiastic adoption.