An autonomous support agent that resolves 70% of tickets
Client: TechStream
- 85%
- faster average resolution
- 70%
- tickets resolved autonomously
- $200k
- annual support cost savings
The challenge
A fast-growing SaaS platform was drowning in support volume — first response times had climbed past six hours, and hiring couldn't keep pace with ticket growth.
What we built
We engineered a multi-agent triage system grounded in their help center and product database. The agent classifies, researches, and resolves routine tickets end to end, escalating edge cases to humans with full context attached.
How we did it
- 01
Two-week discovery: classified six months of historical tickets to find the 70% that followed repeatable resolution patterns.
- 02
Built a planner-worker agent architecture: a triage agent classifies and routes, worker agents research the knowledge base and execute account actions through scoped, permissioned tools.
- 03
Constructed an evaluation harness from 400 real resolved tickets before writing a single prompt — every agent change was gated on eval scores.
- 04
Launched in shadow mode first: the agent drafted responses that humans approved, building trust and training data before autonomy was switched on.
- 05
Expanded autonomy tier by tier — password resets first, billing adjustments last — with full audit trails and instant human escalation paths.
The result
Seventy percent of routine inquiries now resolve without human touch, and the support team spends its time on the conversations that actually need judgment.
Stack
Services used
- AI Agents
- AI & GenAI
More work