Technology & SaaS · 2025
An AI support copilot for a fast-growing SaaS company
58%
tickets resolved by AI
- Client
- B2B SaaS platform (demo)
- Industry
- Technology & SaaS
- Services
- AI & Generative AI, Web Applications
- Year
- 2025
01 Project overview
A SaaS company with 15,000 customers needed to scale support quality without linearly scaling headcount.
02 Business challenge
- Support volume doubled in a year while response times slipped.
- Knowledge lived across docs, past tickets, release notes and Slack.
- Leadership required accuracy guarantees before anything reached customers.
03 Solution architecture
How it was built.
Knowledge
Hybrid search over docs, tickets and release notes with permission filtering.
Agent
Tool-using agent that can check account status and draft cited replies.
Evaluation
Automated groundedness and helpfulness scoring on every release.
Experience
In-app widget plus agent-assist panel inside the help desk.
Technologies used
- Claude
- LangGraph
- TypeScript
- PostgreSQL + pgvector
- Redis
- Next.js
- Zendesk API
04 Implementation approach
Delivered in clear phases.
- 01 · Evaluate
Golden dataset of 1,200 real questions with expert answers.
- 02 · Assist
Agent-assist mode for human agents first to build confidence.
- 03 · Automate
Customer-facing deflection for high-confidence intents.
05 Results & impact
- Tickets resolved by AI
- 58%
- First response time
- -63%
- Groundedness score
- 94%
- Net promoter score
- +12
“The evaluation framework is what convinced us. We know exactly how good the assistant is, every release.”
— Director of Customer Experience (Demo testimonial)
More case studies
72%
faster claims intake
Generative AI cuts claims processing from days to minutes
An insurer processing 40,000 claims a month needed to reduce manual document handling without compromising accuracy or regulatory compliance.
41%
lower infrastructure cost
Zero-downtime cloud migration for a high-traffic retail platform
A retailer with 3M monthly visitors needed to exit an aging data center before peak season without risking revenue.
Want results like these?
Ready to raise support quality without growing headcount at the same rate? Let’s start with your evaluation set.