Building Gorgias' AI Agent: from business case to 1,000 customers in one year

At the end of 2023, ChatGPT changed what customers expected from software. For companies building in customer experience, the question was not whether AI would reshape the category. It was whether you would move fast enough to define it or spend the next 3 years catching up.

At Gorgias, we moved. We set an ambitious goal: design, build, and launch an AI Agent in 5 months. I served as the GM of the product, owning the go-to-market strategy from launch through scale. Here is how we did it, what worked, and what I would do differently.

Making the case to move fast

Before any product work started, we built an internal business case. The question was not whether AI was coming. That was clear. The question was whether the investment was justified and whether we could execute fast enough for it to matter.

We structured the case around 3 things: what the product needed to do in order to be differentiated, what it would cost to build inside our existing platform, and what adoption would look like in the first 12 months if we got it right. The answer was unambiguous. The window to define the category was open, but it would not stay open. We committed to shipping in 5 months.

Building with customers, not for them

The first 5 to 6 months were spent in close collaboration with a small group of customers who knew what was at stake. We did not build behind closed doors.

Alpha: 10 customers. We ran structured conversations, watched how they interacted with early versions, and pushed hard on the question that matters most in early product development: what do customers actually need versus what they say they want. Those sessions shaped the first real version of the AI Agent.

Beta: 25 customers, 10 of them enterprise. Adding enterprise accounts early was deliberate. Enterprise customers have stricter requirements around reliability, configurability, and edge case handling. Hitting that bar early made the product better for everyone downstream.

During the beta phase, I went deep into both the usage data and the customer conversations. I found a pattern that changed how we thought about prioritization: 5 question categories accounted for roughly 80% of the questions customers received. If we could automate those 5 categories well, we would move the needle for almost every merchant on the platform.

That insight became a product decision. I created templates designed to automate those 5 question types, built around each brand's own policies so the AI responses felt accurate and on-brand. No customer had told us to build this. I found it by looking at the patterns across all the conversations. Customers describe their problems; they do not always know the right solution. The job is to understand the underlying need, then design something better than what they asked for.

The launch

The June and July 2024 launch was the largest product launch I have led. Gorgias was a 400-person company at that point, with sales, partnerships, customer success, and marketing all moving on different timelines. Launching something this significant is as much an alignment challenge as a product one.

Internally, I focused on enabling 3 groups:

• Sales: a full pitch deck built from scratch, covering every stage of the conversation: discovery questions, product demo flow, objection handling, and pricing. This was not just messaging guidance; it was a tool the team could walk into any meeting with and use end to end.
• Partnerships: making sure the ecosystem of agencies and technology partners could speak the same language about what the AI Agent was and who it was for.
• Customer success: a clear, repeatable implementation path that could work across different merchant types and setups.

On the implementation side, we created a program called 50 in 50: help customers reach a 50% automation rate within 50 days. We packaged it as an implementation offering, which did 2 things at once. It gave customers a clear path to results, which drove adoption. And it created a commercial structure that rewarded the work we were investing in implementation. Running 50 in 50 required mapping every step of the customer journey, identifying the highest-impact features, and sequencing them in a way that got merchants to the automation threshold efficiently. That program became one of the most effective things we built that year.

On the marketing side, we built a waitlist of around 700 potential participants before launch. We rolled out progressively, sharing features, collecting reactions, and refining the approach. By the end of 2024, the AI Agent had crossed 1,000 paying customers.

Scaling through the year

After the launch, the work shifted from acquisition to refinement. The second half of 2024 was about collecting implementation feedback, improving the product, and building go-to-market tactics tailored to different customer segments.

The key strategy was segmenting customers by where they were in their adoption journey, then designing different interventions for each stage. A customer who had set up the AI Agent but not yet seen results needed something different from a customer with early success who was not expanding usage. Understanding those differences, and acting on them systematically, is what drove adoption rates up through the second half of the year.

The results

The AI Agent reached 1,000 paying customers in under a year, with $8 million in ARR by the end of 2024. Through 2025, the product continued to scale: multiple thousands of merchants and multiple tens of millions in revenue. The automation rate across the customer base grew from near zero at launch to a meaningful share of all support conversations.

These are not just growth metrics. They represent a shift in what Gorgias is: from an AI-powered helpdesk to a platform where AI resolves entire customer conversations end to end.

What I learned

Large launches require more preparation than they seem. At a 400-person company with existing customers, partners, and a complex go-to-market motion, a product launch is not just about shipping. It is about the alignment, the enablement, the sequencing, and the readiness of every team that has to carry it forward. What took the most time: finalizing the positioning language, acting on beta feedback quickly, designing the assets, stress-testing every implementation path. Planning explicitly for that time is what separates launches that land from those that trail off.

Customer obsession is necessary, but not sufficient. Being close to customers is the most important thing in early product development. But it does not mean building exactly what they ask for. Customers describe their problems; they rarely design the best solutions. The 5-question-type insight is the clearest example: no one asked for it. I found it by reading patterns across hundreds of support threads. The discipline is understanding the underlying need, then designing something that works better than what they imagined.

The 80/20 principle is a product strategy, not just a productivity tool. There is always more to build than time allows. The discipline is figuring out which features move the most needles for the most customers, and building those first. We applied this ruthlessly, and it is what made it possible to ship something meaningful in 5 months and scale it to 1,000 customers in the 6 months that followed.

Customization compounds. Different customers come from different starting points. Designing the product and the implementation experience to meet customers where they are, rather than offering a single path for everyone, is what drives adoption at scale. The merchants who reached 50% automation in 50 days did so because the program was built around their specific setup, not a generic template.