An e-commerce site reduced support tickets by 70% with an AI agent

This D2C e-commerce brand specializing in fashion accessories had experienced rapid growth. In two years, they went from 3,000 to 15,000 orders per month. But this growth had a downside: the support team, composed of four people, was drowning in requests. Every week, 800 tickets landed in their inbox, and the pace was only accelerating.

Detailed ticket analysis revealed a recurring pattern. Sixty-five percent of requests fell into three categories: order tracking ('Where is my package?'), returns and exchanges, and sizing questions. These requests, while simple, each required between 5 and 15 minutes of manual processing: opening Shopify, checking the status, copying the tracking number, drafting a personalized response. Multiplied by hundreds of tickets, this represented a massive time sink.

Average response time had climbed to 18 hours on weekdays and regularly exceeded 48 hours on weekends. Customers, accustomed to Amazon's instantaneity, couldn't understand these delays. The CSAT score had dropped to 3.2/5, and negative reviews on Trustpilot were starting to impact the site's conversion rate. Management was considering hiring two additional people, an annual investment of 80,000 euros.

Our proposal was different: rather than adding hands, add intelligence. We deployed a conversational AI agent capable of handling recurring requests autonomously. The agent is connected in real-time to Shopify (order statuses, purchase history), the carrier (live tracking), and the product database (size guides, availability, recommendations).

Designing the agent was precision work. We analyzed 5,000 historical tickets to identify conversation patterns, customer phrasings, and edge cases. The agent doesn't just respond to keywords: it understands context. A customer asking 'where is my order' instantly receives the status with the tracking link. A customer saying 'I want to return the blue sweater' sees the agent automatically identify the relevant order, check return eligibility, and generate the return label.

Intelligent escalation was a crucial deployment point. The agent knows how to recognize situations requiring human intervention: an angry customer, a dispute over a damaged product, a request for a commercial gesture. In these cases, it transfers the conversation to a human agent with a complete context summary, preventing the customer from repeating their problem. This hybrid approach ensures AI handles the volume while humans focus on high-value interactions.

The Gorgias integration created a unified workflow. Human agents see the complete history of exchanges between the customer and the AI, with annotations on detected sentiment and estimated priority. They can pick up a conversation where the AI left off, without any friction. The monitoring dashboard allows real-time tracking of satisfaction metrics by request type, and identifies topics where the AI needs improvement.

Results were spectacular. Within four weeks of deployment, the AI agent was handling 70% of tickets without any human intervention. Average response time for these tickets dropped from 18 hours to 12 seconds. CSAT jumped from 3.2 to 4.6/5. And most importantly, hiring two additional people was no longer necessary, saving 80,000 euros per year. The existing support team, freed from repetitive tasks, now focuses on complex cases and proactive loyalty initiatives.