How PopyTrade Put Their E-Commerce CX on Autopilot
Full before-sales and after-sales automation with Make.com, Zendesk, and an AI agent — from order confirmation to ticket resolution, with zero manual intervention.
Working with Dominic and ROCA on our Help Center customization was an absolute game-changer! They didn't just knock it out of the park — they redefined the standard. The team is incredibly competent and reliable, giving us total peace of mind from start to finish. We were particularly impressed by how fast they operated without ever compromising on quality. They turned our outdated platform into a stunning, on-brand experience. For anyone needing Zendesk customization: ROCA is the fast, reliable, and expert team you want by your side!
Operations that couldn't scale with demand
PopyTrade had the products, the traffic, and the order volume. But the manual effort required to process orders, send confirmations, and handle after-sales inquiries was not keeping pace with the business.
PopyTrade is a fast-growing e-commerce business based in Hungary, serving customers across German-speaking Europe. They sell a wide range of consumer products including flavors, raw materials, vaping supplies, and CBD goods.
With a large and diverse product catalog and an international customer base, managing the full order lifecycle — from purchase to delivery to after-sales support — had become increasingly complex.
They came to us with a clear goal: run more of their e-commerce operations on autopilot without sacrificing the customer experience.
We designed and implemented a full before-sales and after-sales automation stack built around their existing tools, with an AI agent at the center of the customer support flow.
A full automation stack, end to end
Three interconnected systems working together to remove friction at every stage of the customer journey.
E-Commerce Automation with Make.com
PopyTrade was already using Magento and Stripe, but the connections between their systems were largely manual. We mapped out their full order flow and built an automated pipeline using Make.com. When a customer places an order, the flow triggers automatically: Magento registers the order, Make.com instructs Stripe to generate a payment link, the customer completes payment, and the order is pushed to BaseLinker for processing and tracking. Once a tracking number is generated, Make.com passes all order and tracking data to Klaviyo, which sends a branded confirmation email — all within seconds of purchase, with zero manual intervention.
AI-Powered After-Sales Support
After-sales was where PopyTrade felt the most friction. We introduced a Zendesk-based support flow with an AI agent powered by OpenAI / Google Gemini. When a customer sends an email, Zendesk captures the ticket and fires a webhook to Make.com. Make.com then queries BaseLinker and Magento to identify the customer and retrieve their order data. That context — order details, tracking status, shipment information — is passed to the AI agent along with a carefully structured system prompt. The agent composes a personalised, accurate response and posts it back to the Zendesk ticket automatically. Most common after-sales questions are resolved without a human agent ever needing to touch the ticket.
AI Agent Prompt Engineering
The quality of an AI support agent lives and dies by its prompt. We built and iterated a production-grade system prompt that defines the agent's persona, tools, escalation logic, language behaviour, and financial guardrails. The result is an agent that handles the full spectrum of after-sales scenarios reliably — without ever going off-script.
What makes the agent reliable
Six key elements built into the system prompt to ensure accuracy, safety, and great customer experience across a multilingual market.
🌍 Language detection & mirroring
Automatically detects and responds in the customer's language — critical for a DACH customer base spanning German, Austrian, and Swiss markets.
🛡️ Sentiment guard
Switches to a neutral, professional tone when customers are frustrated or threatening, de-escalating tense interactions automatically.
💶 Financial guardrails
Strict rules preventing the agent from inventing discounts or generating financial data — protecting the business from costly errors.
📋 Action priority matrix
Clear priority order: returns, damages, payment issues, cancellations, order status — ensuring the most critical issues are handled first.
🔄 Loop prevention logic
The agent never responds to its own outputs or internal system messages, preventing infinite loops in the automation chain.
🤝 Human handoff failsafe
Automatically escalates to a human agent whenever the situation falls outside defined parameters — so nothing slips through the cracks.
A living system, not a one-time deployment
Deploying the agent was only the beginning. We worked with PopyTrade to establish a feedback loop: when the agent produces a wrong or suboptimal answer, that example is used to refine the prompt.
The prompt evolves with the business — new product lines, new policies, new edge cases all get documented and translated into agent behaviour.
This is not a set-and-forget solution. It is a living system that improves over time with consistent attention. We equipped PopyTrade's team with the process and tools to keep it sharp.
Want the AI agent prompt template we used?
Download the production-grade e-commerce AI agent prompt from this project — free at roca.work
Let's automate your customer experience
Book a free 30-minute strategy call and we'll map out exactly what's possible for your team — no commitment required.