G2A's Autonomous AI Agent Dave Resolved 14,400 Support Tickets in 63 Days — What It Means for E-Commerce

What Happened: G2A Launched Its Own AI Agent, Dave, Globally

At the end of July, G2A.COM — one of the world's largest gaming marketplaces — announced the global launch of Dave, its proprietary autonomous AI customer support agent. After a two-month trial period, the agent resolved more than 14,400 seller support tickets in 63 days: an average of 228 tickets per day, running 24/7 across 180 countries.

For context: a typical support operator handles 30–50 conversations a day, and even that is at capacity. Dave works several times faster, without breaks, in multiple languages and — most importantly — at a consistent speed even under peak load. That is exactly the kind of spike caused by major game releases: a single blockbuster launch (the press release names GTA 6) can generate thousands of inquiries within hours, more than any human team can physically keep up with.

One important caveat: the information was published as a company press release (distributed via TechnologyWire), so all figures are self-reported by G2A and have not been independently audited. That does not diminish the core point, though: this is a concrete, measurable case of an AI agent deployed in real operations — which is still rare today.

How Dave Works — and Why It Is Not a Typical Chatbot

In recent years, chatbots have become a habit: a customer writes, the bot replies with links, and the case still ends up with a human. Dave works differently. It connects directly to the seller's transaction history and verifies payment status, order state and the seller's policies in real time — so it can answer questions immediately.

This is a fundamental difference from a plain generative chatbot: the agent does not base decisions on generated text alone, but on verified data. That is precisely what reduces the risk of hallucinations — AI's biggest danger in customer-facing communication. According to G2A, Dave also handles complex cases: vouchers, game keys, software licenses and refunds.

Equally important, the agent never leaves the customer in a dead end. If a case requires the seller's manual intervention, the seller and their team receive a full analysis and summary of the case — so the customer never has to repeat themselves. This is the kind of detail that genuinely changes the customer experience.

The Results in Numbers: What the Metrics Say

Volume: roughly 15,000 tickets processed in 63 days (the headline figure is 14,400) — more than 228 per day on average. This cut the number of tickets reaching sellers' support teams by 30%, and reduced the volume of tickets requiring human intervention by 27–30%. The agent fully resolved 54% of cases.

Quality: the average response quality score is 91.2%, with only 0.81% receiving negative in-chat feedback. Every response is automatically evaluated across 13 quality and safety dimensions. That last detail matters most: the agent constantly evaluates its own work and does not let errors accumulate.

Customer satisfaction: an average score of 4.16 out of 5, and 74% of users rated the process easy or very easy. Globally, the agent now manages roughly 10,000 conversations per week.

Reach and accuracy: 24/7 operation, 180 countries, multilingual support, and 93.8% routing accuracy — cases are correctly distributed between sellers and the platform's own support, in line with marketplace policy.

"Autonomous" Means Full Resolution, Not Just Routing

The term "autonomous agent" is often used loosely. Many products marketed as "AI support" only do classification and routing. In Dave's case, autonomy means the agent processes the ticket end to end: it receives the request, checks the data, makes a decision, writes the answer and, when needed, performs the action itself.

Paweł Wróbel, CGO at G2A.COM, explains exactly this logic: demand spikes in the gaming ecosystem — like a major game launch — appear within hours, and scaling by "hiring more operators" is inefficient. In his words, Dave is a step toward building an "AI-native organization": "AI is no longer just a productivity tool or a competitive advantage. The real advantage comes from embedding AI into the core of how a company operates."

That is the key takeaway for marketplaces and e-commerce: AI agents are most effective when they have access to real business data — orders, payments, policies — and can make independent decisions, while humans remain the layer of control and complex-case handling.

Lessons for Georgian Businesses and E-Commerce

It may seem that the experience of a global marketplace with 35 million users is far from Georgian reality. But the principles are the same — only the scale differs.

E-commerce is growing in Georgia, yet support remains one of the most expensive operational costs for small and medium businesses. An online store processing hundreds of orders a week must either hire an operator, or the owner answers messages late at night themselves. Seasonal peaks — New Year, flash sales, promo campaigns — triple or quadruple demand, and that is exactly when support grinds to a halt.

Dave's approach can be adapted at a much smaller scale. An AI agent connected to an order database can answer a large share of inquiries: order status, delivery times, payment confirmation, refund procedures, product usage. In practice, that is 60–70% of all incoming questions — and that is where the economics of automation live.

You do not need hundreds of developers for this. A modern stack — LLM + database + simple integrations (CRM, payment system, courier service) — is enough to get a first agent running within a few weeks. The key is to start not with technology but with analysis: which questions repeat most often, which can be answered automatically, and where human intervention is mandatory.

The economic logic is simple: an AI agent's monthly cost is significantly lower than an operator's salary, and it works 24/7 — weekends and holidays included, exactly when online shopping is most active. For marketplaces it matters twice as much: support quality directly affects ratings and sellers' revenue.

A Word of Caution: What Not to Take at Face Value

This story deserves healthy skepticism. First: it is a press release — company-published data without independent verification. Second: the definition of a "resolved ticket" varies between companies — the text itself uses two different framings: "resolve" and "streamline." This means some tickets may have been efficiently processed and routed rather than fully closed.

Third: the hallucination risk does not disappear simply because the agent is connected to a database. G2A made the right architectural choice — verification based on data — but that requires constant monitoring. That is exactly why the company scores every response across 13 dimensions: this level of governance is now a necessary condition for AI support, otherwise mistakes are inevitable.

A practical recommendation for Georgian businesses: start with clear metrics (resolution rate, customer satisfaction, error rate), set escalation rules — when a case goes to a human — and roll the agent out in stages: first drafting answers, then autonomous decisions.

Conclusion: AI-Native Operations as a Competitive Edge

G2A's case matters not only for the numbers but for the precedent: a large, real-market company demonstrated an AI agent's ROI in its own operations — closed tickets, reduced load, measurable quality. "AI-native organization" may sound like a marketing cliché, but Dave fills the concept with concrete meaning: AI is not a standalone chatbot in the corner of a website — it is part of operations, with access to data and accountability for results.

The timing is favorable for Georgian business: AI tools are getting cheaper and simpler, while competition is growing. Those who embed agents into their support and sales first will gain the cost and speed advantage that is already part of today's competition — not a tomorrow opportunity.

📖 Source