Customer service doesn’t need faster agents. It needs better matches.

AI is changing how customer problems get solved. AI Agents can resolve more interactions autonomously. Agent Copilot can make agents more effective. But there is another opportunity sitting upstream of the work itself: using AI to decide where that work should go in the first place.

Routing is often an invisible part of customer service, but it shapes the work that follows. Before  an agent answers a customer, a consequential decision has already been made: Who should get this problem?

Get the match wrong and everyone downstream has to work harder. Get it right and the same workforce can resolve the same problems with less effort.

Predictive Routing changes how that decision gets made: from rules based on assumptions about who should handle what, to a model that learns from what actually happens and adapts as your team changes

AI Agent Copilot can make individual agents more capable. Layer in AI Predictive Routing, and we can make the entire operation more intelligent.

Routing rules freeze an organization in time

Service teams have traditionally used rules to decide where customer problems go: which agents handle billing, which handle technical issues, which queues take priority.

Those rules are necessary. The problem is that they freeze an organization in time.

Agents learn. Products change. New kinds of issues emerge. Expertise shifts. But manually configured skills, groups, tags, and routing rules only change when someone updates them.

“The fundamental weakness of this approach is not that the rules are wrong,” writes Zendesk Senior Director of Product Management Barry Neary in Zendesk’s white paper on Predictive Routing. “It is that they are static in a dynamic environment.” Rules describe who should be good at something. Learning systems can observe who actually is.

“Who should get this ticket?” is harder than it sounds

Predictive Routing uses machine learning to answer a specific question for every eligible agent: How long would it take this agent to handle this ticket?

The model is trained on cross-industry data from millions of real agent-customer interactions. As account-specific data accumulates, the predictions become increasingly tailored to that operation.

Match point: where routing creates value

Zendesk tested Predictive Routing through randomized A/B tests across six enterprise deployments. Four saw statistically significant reductions in agent engagement time of between 9 and 21 percent.

At one company, 55 agents handling roughly 2,000 tickets a week recovered an estimated 98 hours of agent time every week. That's two full-time agents worth of capacity recovered — without hiring, without process changes, without asking anyone to work harder.

Those gains didn’t come from adding headcount, changing processes, or shifting more work onto the fastest agents. Workload distribution remained statistically unchanged.

“The efficiency gains came from better matching, not from concentration of work,” Neary writes. The agents didn’t become faster. Predictive Routing got better at putting problems where they could be solved faster.

“The efficiency gains came from better matching, not from concentration of work,” Neary writes.

Make the whole operation more intelligent

Predictive Routing happens before the customer sees the work, but its impact can shape everything that follows. It creates the most value in operations with genuine agent diversity, sufficient ticket flow to generate enough volume for the model to learn your team's patterns, and routing complexity that creates real decisions about where work should go.

By getting each problem to the agent best placed to solve it, Predictive Routing can reduce the effort required to resolve it, create capacity across the team, and ultimately get customers to an answer faster.

At Zendesk, we’re applying intelligence across the service journey to create value for both the customer and the business. AI Agents can resolve more interactions autonomously. Copilot can make agents more effective on the problems that reach them. Predictive Routing can improve where that work goes in the first place.

Each capability tackles a different part of the customer service problem. Together, they point to a bigger opportunity: making the entire service operation more intelligent.

The future of customer service will depend on getting better at deciding how—and by whom—each customer problem should be solved.

For service leaders, the opportunity is to look beyond what happens during the customer interaction. Some of the most consequential decisions happen before the work begins—and better decisions can mean less effort, more capacity, and faster resolution.