E-Commerce Team Reduces Support Ticket Volume 58% With Claude Agents

A mid-market e-commerce company was drowning in support tickets. 800-1,200 tickets per day during peak season, 60% of which were tier-1 questions: “Where is my order?”, “How do I return this?”, “Can I change my shipping address?”

The support team of 12 was fully occupied. Average response time was 4 hours during business hours, 18 hours overnight. Customers were churning over support quality.

The Problem

They’d tried chatbots before. The experience was bad: rigid decision trees, frustrated customers asking for a human, zero ability to handle anything slightly outside the script.

The appeal of Claude agents was different: actually understand the question, access real data, give a real answer. An agent that could look up order status, explain the return policy in context, and process a simple return request without human involvement.

The risk: these agents needed access to customer data. Order history, shipping addresses, payment information. The wrong policy could expose data it shouldn’t. The wrong action could process an incorrect refund or modify an order that shouldn’t be touched.

They needed a way to give agents enough access to be genuinely useful, with hard limits on the most sensitive operations.

The Solution

Sentrely with a carefully layered policy for their support agents:

Tier 1 (fully autonomous): Order status lookups, shipping tracking, standard return policy questions, product information. The agent reads from their order management system and responds. No human in the loop.

Tier 2 (agent handles, human can review): Processing standard returns within policy (item returned within 30 days, in original condition). The agent initiates the return, sends confirmation to the customer, and logs the action. Support leads can review the queue but don’t need to approve each one.

Tier 3 (requires human approval): Anything involving a refund over $200, address changes on shipped orders, any dispute where the customer claims the agent made an error. These route to a Slack channel where a support lead can approve or modify with full context.

The policy ensures the agent can never touch billing information, cannot issue refunds without approval on high-value orders, and cannot modify orders that have already shipped without human sign-off.

The Results

58% of tickets handled autonomously — the entire tier-1 volume plus a significant portion of tier-2. The support team reviews the tier-3 queue, which is far more manageable than the original full volume.

3-minute average response time. Autonomous responses are nearly instant. Tier-3 escalations are resolved by support leads within 15-30 minutes during business hours. The 18-hour overnight wait is gone.

120 hours saved per week — allowing the support team to focus on complex cases and customer relationship work that actually requires human judgment.

NPS improved by 28 points in the quarter after deployment. Faster response time and more accurate answers were the primary drivers cited in customer feedback.

The VP of Customer Experience: “We were skeptical about AI in support because the previous chatbot was so bad. What’s different with Claude agents and Sentrely is that the agent actually reads the situation and the controls mean it can only do what it should do. Our agents feel like a smart colleague who happens to work 24/7.”