Thousands of tickets a day about refunds, bugs and lost passwords buried the support team, and every one meant an agent hopping between the billing systems and the developers' tracker to work out a single player's status. Players waited, got frustrated, and left, and the churn showed up straight on the revenue line.
With millions of active players, the studio's support queue never emptied. Thousands of tickets a day arrived about refunds, bug reports and account access, and each one landed on a human who had to sort it, judge it and act on it by hand.
The real drag was the swivel-chair work. To resolve one ticket an agent moved between the billing systems and the development team's issue tracker, piecing together the player's purchase history and the status of whatever they were complaining about. It was slow, it was repetitive, and it did not scale with the player base.
Players felt every minute of it. Long waits on something as simple as a refund pushed people out of the game, and in a business where a churned player is lost revenue, a support backlog was quietly a growth problem.
Route every ticket, settle the safe refunds within the rules, and turn a pile of bug reports into one clean ticket for the developers.
The first agent reads and classifies each incoming ticket the moment it arrives: refund, bug, account access or something that needs a person, so nothing sits in an undifferentiated queue waiting for a human to open it and decide what it even is.
A second agent runs with read-only permissions, checks the player's purchase history, and issues a credit automatically only when the request meets pre-defined business logic. It can approve a clean case in seconds but it cannot invent an exception, so speed never comes at the cost of control.
A third agent groups the many near-identical bug reports players file about the same issue into a single consolidated, structured ticket, and opens it directly for the development team in their tracker, so engineers see one clear signal instead of hunting through a hundred duplicates.
Splitting the work across purpose-built agents, each with its own narrow job and permissions, made the system easier to trust and to reason about than a single model trying to do everything, and let each part be tuned and constrained on its own.
Anything outside the first tier still goes to a person, and the support staff no longer buried in routine tickets were moved onto personal, high-touch care for the studio's highest-value players, where a human genuinely changes the outcome.
The agents are wired across the billing and development systems that players fall between, so the purchase history, the ticket status and the fix all connect automatically instead of a human stitching them together tab by tab.
Two thirds of the routine queue cleared itself, and the people were freed for the work only people can do.
The system resolved 65% of first-tier tickets with no human involvement at all. Simple refunds and common issues closed in seconds instead of sitting in a queue, and the basic support team shrank sharply as the routine load disappeared.
An immediate, smooth support experience cut player churn and protected revenue, the remaining agents moved to personal care for high-value players, and consolidated bug tickets reached the developers faster, so the problems behind the complaints got fixed sooner too.
An automated refund is only safe if it physically cannot over-reach, which is why the refund agent is read-only and bound to pre-set rules, and why the risky calls still route to a person. The value is a set of narrow, constrained agents wired into the real systems, not one unchecked bot with the keys. It is the same agentic-automation discipline behind our insurance automation work, aimed at a game studio's player base and its developers rather than a contact centre.
A delivery platform's agents juggled four systems to resolve one late order. AI triage with a ready credit and a hard guardrail cut handle time 15%, lifted first-contact resolution 30%, and handled 1.5× the tickets.
Read the case → Community safety · GamingToxic chat was driving players off a gaming platform while regulators circled. Context-aware moderation across 30+ languages told banter from abuse, cutting toxicity 45%, lifting new-player retention 20%, and dropping false blocks 75%.
Read the case →
Reef TRH