An AI support agent that resolves 60% of tickets

Challenge

Demand outgrowing the team

Volumes kept climbing while response times slipped. Most requests were repetitive, but every one still waited in the same queue as the hard cases.

Objectives

What we set out to achieve

Resolve routine requests instantly and give people more time for the complex ones.

• 50%+Automated resolution
• <10sResponse time
• 0Drop in satisfaction

Strategy

Automate the top intents first

We analysed six months of conversations, ranked intents by volume and risk, and launched with the ten that covered most of the demand.

Solution

An agent that can actually do things

Grounded in policies and connected to the CRM, booking and billing systems, the agent completes tasks — not just answers — and hands off with a summary when it should.

Technology

The stack behind it

Chosen for reliability, speed and easy ownership by the client team.

• OpOpenAI
• ClClaude
• TwTwilio
• HuHubSpot
• NoNode.js
• PoPostgres
• ZeZendesk
• LaLangSmith

Implementation

8 weeks, four phases

Weekly demos kept the client team involved at every step.

1. Weeks 1–2Conversation analysisIntent mapping, policies, success metrics.
2. Weeks 3–5Agent & integrationsTools, knowledge base, hand-off flow.
3. Week 6TestingRed-teaming and real-transcript replays.
4. Weeks 7–8Launch & tuneGradual traffic ramp and weekly tuning.

Results

Faster answers, happier customers

-45% Support Costs in the first 90 days, while satisfaction scores held steady or improved.