Adding AI to service is no longer enough. The next step is building an Intelligent Service platform that can act on the right work, validate every outcome, and continuously correct what isn’t working.
Thomas Verschoren
Director, AI Product Evangelism
최종 업데이트 2026년 9월 24일
For years, customer service technology has been built around one central question: how do we help agents move faster? That question still matters. But it's no longer enough.
AI has changed the ambition of service. The goal is not simply to make agents more productive, resolve a few more tickets automatically, or add a chatbot in front of an existing support process. The goal is to build a service operation that can understand what customers and employees need, resolve more issues automatically, measure whether those resolutions were right, and improve itself over time.
That is what separates an AI feature from an Intelligent Service platform.
At Zendesk, we think about this through a simple operating model: act, validate, correct.
The platform needs to act on behalf of the customer, employee, agent or admin. It needs to validate that those actions are accurate, safe and successful. And it needs to help the business correct what is not working, so the service operation becomes better with every interaction.
This is the difference between adding AI to service and building service around AI.
Act: automate the right thing
The first job of an Intelligent Service platform is to act. That sounds obvious, but this is where many AI strategies become too narrow. In customer service, AI is often still treated as a smarter reply engine. It can understand a question, generate an answer, summarise a conversation or suggest the next best response. Those are useful capabilities, but they are only the beginning. The real value of AI comes from what it can do, not just what it can say.
In a modern service operation, the right action can take many forms. Sometimes it’s a fully automated resolution by an AI Agent. Sometimes it’s an assisted action suggested by Agent Copilot, where AI helps a human agent move faster and make better decisions. And sometimes the right action is handled by a Specialised agent built for a specific workflow, with the context, permissions and steps needed to complete that job reliably.
That’s why our recent launch of Specialized Agents is key. Service is not one generic process. It’s a collection of different workflows, intents, edge cases and business rules. A return is different from a billing dispute. A password reset is different from an account change. An IT request is different from a payroll question. Each may require a different path, different data and different safeguards.
An agentic workforce reflects that reality. Instead of one generic bot trying to answer everything, businesses can deploy AI agents that are specialised for the work they are meant to perform. They understand their domain, follow the right workflow, and know when to act, when to ask for more information, and when to hand it over to a person.
This is the platform shift. Automation isn't about deflecting as many tickets as possible. It’s about automating the right thing, in the right context, with the right level of control:
For simple, low-risk requests, the platform should resolve the issue directly.
For more complex issues, it should assist the human agent with context, recommendations and actions.
For specialised processes, it should use the right agent for the job.
And when the issue requires judgement, empathy or exception handling, it should escalate cleanly with the full context preserved.
This is how AI becomes fully operational. Not as a layer that sits beside service, but acting as part of the service workforce itself.
Automation isn't about deflecting as many tickets as possible. It’s about automating the right thing, in the right context, with the right level of control.
Validate: prove AI is working
The second job of an Intelligent Service platform is to validate.
If AI is going to act, businesses need confidence in what it’s doing. They need to know which issues were resolved, which ones were escalated, which answers were used, where customers were satisfied, where agents had to intervene, and where automation created friction.
This is especially important because service is not only a volume problem. It’s a trust problem.
Customers and employees don’t care that an interaction was automated. They care whether the issue was handled correctly. Businesses therefore need more than automation rates. They need to understand quality, accuracy, containment, resolution, customer sentiment, agent effort and business impact.
That is where analytics, reporting and QA become foundational. They’re not after-the-fact dashboards. They’re part of the AI operating model.
Validation gives teams the confidence to scale automation responsibly. It helps leaders understand which workflows are ready for more AI, which need better knowledge, which should stay human-led, and which require changes to policy or process before they can be automated safely.
It also changes the conversation around success. The question is not simply, “How many tickets did AI touch?” The better question is, “Which issues did AI resolve successfully, and how do we know?”
That is why trust also needs to show up commercially. Outcome-based pricing is powerful because it aligns the platform with the outcome customers actually care about. Businesses should not pay because AI was present. They should pay because work was resolved.
When validation is built into the platform, AI becomes measurable. Teams can see what’s working, inspect what’s not, and make better decisions about where automation belongs.
Correct: move from resolving issues to preventing them
The third job of an Intelligent Service platform is to correct.
This is the part that turns AI from a productivity tool into an improvement engine.
Every service interaction contains a signal, for example:
A customer asking the same question again and again may point to missing knowledge.
A surge in billing tickets may point to a confusing invoice. A repeated handover may reveal a broken workflow.
A long resolution time may show that agents do not have the right permissions or context.
A poor AI answer may expose a gap in the underlying content.
Traditional service operations often struggle to act on those signals because teams are too busy resolving today’s tickets. They know there are root causes to investigate, articles to improve, workflows to redesign and policies to clarify, but the queue keeps moving and the same issues keep coming back.
AI changes that balance. By resolving more of the work that can be resolved today, it creates time and room for teams to analyse why the work is being created in the first place. The future of service isn’t just about faster resolution. It's about prevention: finding the reasons customers and employees need help, fixing the underlying issues, and reducing the need for future tickets.
Correction closes the loop. The platform acts on the issue in front of it. The organization validates whether that action worked. The system learns where the operation can improve. Then the platform helps make those improvements real, whether that means recommending a knowledge update, identifying a broken process, improving routing, suggesting a new automation, or helping an admin design a better workflow.
This is where the role of the admin starts to change.
Historically, Zendesk admins have needed to do three things at once. They had to understand the business, understand the service process, and understand how to translate that into platform configuration. That last part often required deep product knowledge: knowing which settings, triggers, automations, workflows and integrations to build.
Admin Copilot changes the product experience by helping admins turn intent into configuration. Instead of starting with a blank screen and building manually, admins can describe what they want to achieve, review the proposed setup, and approve the implementation.
But the bigger shift is the emergence of the Service Architect.
A Service Architect is not simply a traditional admin with better tools. It’s a shift in the role itself. Less time spent building individual configurations by hand. More time spent designing the service model, deciding how work should flow, defining what should be automated, setting the right guardrails, and making sure the operation reflects the needs of the business.
In that future, the most important skill isn't “know Zendesk”. It's “know your business”.
The platform should help with the building. The human should focus on the decisions.
The future of service
Act, validate, correct is not only a model for customer service. The same shift is happening across employee experience or your backoffice processes.
Employees also need answers. They also need requests resolved. They also get stuck in fragmented processes across HR, IT, finance, operations and internal support teams. And businesses face the same challenge: how to automate more work without losing accuracy, trust or control.
An Intelligent Service platform needs to work across that entire landscape. It should help customers get support. It should help employees get help. It should help agents and internal teams work faster. And it should help admins and service leaders continuously improve the operation behind the scenes.
This is the larger promise of AI in service.
Not just faster replies. Better action. Not just more automation. More trusted resolution. Not just fixing tickets. Preventing the next ones.
The companies that get this right won’t treat AI as a feature added to the side of their service operation. They’ll use it as the operating layer for how service improves. They will build systems that act, validate and correct. And over time, those systems won’t just resolve more work. They’ll help the business understand why that work exists, where the experience is breaking down, and how to make service better for everyone.
Thomas Verschoren
Director, AI Product Evangelism
Thomas Verschoren is Director of AI Product Evangelism at Zendesk, where he translates the platform’s rapid AI evolution into clear narratives for customers, go-to-market teams, and product leaders. He writes Internal Note, a strategic blog that connects individual Zendesk releases into the bigger story of modern AI-powered resolution platform
Before joining Zendesk, Thomas spent years as an implementation partner, designing and deploying the platform for organizations across industries. That hands-on experience grounds his work today: showing what is possible now, where the practical limits are, and where the platform is heading next.
Move from adding AI to building intelligent service
See how Zendesk helps businesses build a more capable, measurable, and trusted service operation.
