The Customer

Gorgias provides a product called AI Agent, the conversational AI for e-commerce brands that drives sales and automates repetitive support work through personalised, on-brand conversations.

Gorgias’s AI Agent helps drive growth for ecommerce brands:

• AI that sells and streamlines: AI Agent acts as a Support Agent and Shopping Assistant, automating up to 60% of repetitive support tasks and driving 62% higher conversation rate for ecommerce brands.
• Maintains brand excellence at scale: AI Agent maintains the same level of customer excellence brands are known and loved for. Following the brand’s tone of voice, sales and support instructions, every message is personalised and accurate.
• Always available, at every touchpoint: AI Agent is always available, across key sales and support channels, never missing a chance to convert or the right moment to provide support.

The Challenge

In 2024, the Gorgias engineering team incorporated agentic AI into one of their customer service products, which helped manage all customer support questions. The first agent they deployed was reactive. It could only respond to questions over email, like, “Where’s my order?” or “What’s your return policy?” This agent followed a simple pipeline of steps to retrieve data.

After initial success, the team wanted to expand to proactive use cases. The agent needed to trigger actions, like canceling orders. This required a more complex system with more components, like external HTTP calls to warehouses.

They knew they needed:

• Retry policies with backoffs for the external APIs
• Saga patterns with distributed transactions for complex steps, like returning an order
• Support for asynchronous steps like finding tracking IDs
• The ability to pause in the middle of a workflow and wait 10-20 seconds (i.e. “sleep”), then send a message asking the customer in chat if they’re still there

Building these capabilities on their own would be challenging and time-consuming, so they searched for a framework.

Once you understand what a Workflow is and what an Activity is, it’s really easy to build something that works in production with Temporal, and is easy to iterate on. So it was a perfect fit for us. –Romain Niveau, Senior Engineering Manager

The Solution

The engineering team attended Replay 2024 in Seattle to learn more about Temporal. By the end of that week, they’d built a POC to validate Temporal would support their use case. They also did a spike on their initial agentic use case with a Temporal certified services partner, SpiralScout, to help accelerate learning and establish clear best practices. The team found Temporal easy to use, build with, and iterate on, and decided to move forward.

In addition to solving their technical requirements, Temporal provided:

• Support for both Python (used by the Gorgias’ ML team) and Typescript (used by Gorgias’ workflow orchestration team)
• Temporal Cloud—the team didn’t want SREs to waste time hosting the Server, as they wanted to move quickly

At the end of the first week, we had built a POC to check if Temporal could fit our needs or not…. One month later, we had a version in production, and we were able to run the support agent on Temporal for some customers.

The Results

The team rapidly built their first Temporal-supported agent and deployed it to production within a month.

In their customer support architecture, each agent is its own Temporal Workflow. There are three agents: 1) a support agent that handles support issues. 2) A sales agent that searches for upsell opportunities, like asking the customer if they want to buy a new pair of shoes after returning the existing ones. 3) A manager or “synthesizer” agent that sits in front of the other two. It synthesizes responses and chooses the appropriate response channel (email, SMS, or chat).

In addition to getting to production faster, Temporal lets the team:

• Avoid dealing with complexities like retry policies
• Design more successful systems—the concept of Activities forces developers to think more carefully about architecture and makes it easier to align stakeholders around architecture

We're able to work with the agent again and again. It was really painful to do this without a workflow approach. I think for us, a workflow approach with Temporal was good because in the end, all LLM use cases are workflows.

The Takeaways

Temporal is helping Gorgias revolutionize customer service for e-commerce brands small and large. In the future, they plan to expand Temporal to additional use cases, both within AI and outside of AI. With Temporal’s use-case agnostic approach and polyglot language support, it will be a powerful tool to support these initiatives.

We wanted a framework that’s not only focused on AI, because agents are just the first use case. We also want to expand usage of Temporal within the company, and not everyone is doing AI work.

As Gorgias continues scaling, they’re hiring across teams. You can find out more here.

Looking to improve your AI agents just like Gorgias did? Start today with a free trial of Temporal Cloud and $1,000 in credits.