Introducing Zendesk Outcome AI™: better outcomes wherever service begins

When most AI in service was still judged by how many conversations it deflected, we argued it should be judged by whether it actually solved the problem. That became the standard, and last year Zendesk resolved more than 4.5 billion service issues.

Resolution was the right standard for that moment. It was never meant to be the finish line.

Sitting at the centre of billions of service moments shows us where the bar is moving next. A ticket can close while the customer still leaves, or the employee still can’t get back to work. Customers feel that gap: 83% believe customer experience should be far better than it is today, and 88% say the quality of service they receive strongly influences what they buy.

Leaders feel it too. 79% of CX leaders say traditional metrics alone can’t capture the full impact of AI-driven service. They’re no longer asking whether AI closed the ticket. They want to know what outcome it delivered: whether the customer stayed, the order went through and the new hire was up and running on day one.

A trusted way into service, wherever work starts

How people work is changing fast. Teams increasingly work inside AI tools like ChatGPT, Claude and Copilot, and people are starting to hand everyday tasks to personal AI assistants. Nearly two-thirds of consumers already use AI more now than they did a year ago. Soon, they’ll hand off their service problems too.

That’s becoming the new norm, and it means service has to show up where the work happens. Teams need Zendesk inside the tools they already use, and customers’ assistants need a trusted way to find the right company. 

“At Zendesk, this is the shift we’ve been preparing for: evolving service to meet people where they already are, and making sure the trust that used to live inside one front door for service travels to every door a customer might use. We think that’s the most exciting rethink our industry has had in a decade.” – Ram Sangireddy, VP of Platform Product and Engineering

Zendesk Bridge makes that possible. It helps businesses deliver better service outcomes by bringing Zendesk’s specialised tools, context and AI agents to the agents, apps and tools that customers and employees already use today. A support lead, for example, can investigate a payment issue in ChatGPT or Claude using a Zendesk ticket. A customer can get help from an AI agent inside a customised app. And an employee can get an onboarding question answered in Microsoft Copilot.

Bridge also helps businesses stay in control. Administrators choose which agents can connect, who can use them and what they can do. Existing user permissions still apply, and teams can see which agent acted and what happened. 

The Open Service Directory extends that same trust to customers’ own AI assistants. It helps compatible personal AI assistants find a brand’s verified support destination. Brands approve their listings, and participating assistants can connect customers to official support with permission. That way, customers get a trusted way to reach the right company, rather than having to sort through search results and guess which route is legitimate.

Agents built to do the deeper work

Access is only the start. The next test is whether the work gets done. An answer that leaves someone chasing approvals, switching systems or coordinating the next step is only part of the job. That’s the gap between a resolution and an outcome.

That’s especially clear in employee service. A request for software access or an employment letter can involve policies, approvals and records spread across IT, HR and Finance. Employee Service AI agents, powered by Advanced Action Flows, can coordinate those steps across connected systems. An employee might request access in Slack, for example, and the agent can route the approval and carry out the next steps without making the employee figure out who to contact or where to go.

“Too often, employees lose time chasing approvals and navigating internal systems instead of doing the work they were hired to do. That’s the promise of an agentic employee experience: less friction for employees and more time for the work that matters.” – Shashi Upadhyay, President of Product, Engineering and AI

Early results show how much manual work Employee Service AI agents can take on. In their first two months of customer use, 300 customers executed 9 million workflows, work Zendesk estimates would otherwise have taken 750,000 hours to complete manually.  

The same principle applies to customer service, though the work varies by business. A return, a healthcare claim and a billing question each depend on different policies and systems. Industry Agents bring specialised knowledge to common service needs, while businesses can use agent builder to create customised agents for their own workflows and exceptions.

And when a conversation moves from digital channels to a phone call, voice AI agents can help carry context with it. Real-time translation and proactive follow-ups help customers keep moving without having to start over.

The measure isn’t how many tasks an agent can take on. It’s whether customers and employees get the outcome with less waiting, fewer handoffs and less effort. 

Improve the work, not just the volume

A good outcome matters. So does making the next one easier to achieve. As AI takes on more service work, patterns across those interactions can show teams where the process still breaks down. Repeat contacts may mean the first answer didn’t solve the problem. Unanswered questions may point to gaps in knowledge. Escalations may reveal a policy that needs human judgement. Across customer and employee service, these signals help teams see where effort is still getting in the way.

Self-improving Zendesk helps teams use that evidence to make service better. Agentic analytics helps them understand what to fix and why. Admin copilot can help turn what they learn into a proposed change. AI agent launchpad helps teams discover automation opportunities from past conversations, generate knowledge and procedures and test AI agents before they go live. Simulations show where agents succeed and where they need improvement, while A/B testing helps compare versions to understand which delivers better outcomes.

People review and approve changes before they go live. Afterwards, teams can measure what changed and decide what to improve next with a quality score. 

The work is a continuous feedback loop: find the pattern, make a change, test it and see whether it helped. Teams can track whether customers had to come back less often, whether issues were resolved sooner and whether service quality held up.

Price the work by the value it delivers

If outcomes are what matter, they should also be what businesses pay for. A per-seat model can make that harder, especially when a request needs someone beyond the core services team.

Outcome-based pricing ties spend to the work AI completes, rather than the number of people involved. As Lucy Field, VP, Product, explains: “We’re giving customers more freedom to automate with AI and bringing more people in to help resolve customer issues. Unlimited seats and a flexible outcome budget help teams resolve issues their way, with the visibility and safeguards they need to stay in control.”

Spend is tied to two kinds of outcomes: conversational outcomes, when an issue is successfully addressed within a customer interaction, and operational AI outcomes, when an automated workflow is completed across business systems.

The model builds on two years of live enterprise leveraging, hundreds of AI customers and billions of automated outcomes. As software takes on more meaningful work, the value and the way organisations pay for it should reflect what gets accomplished.

The outcome is the experience

The measure of AI isn’t how well it talks about the work. It’s whether people get the outcome they came for.

That’s what the Outcome AI™ Platform is built for. Bridge gives AI access to the work, wherever it starts. Specialised agents take on deeper, more complex work. And self-improving Zendesk learns from every result, so the next outcome is better than the last.

We helped make resolution the standard for AI in service. Now we’re setting a higher one.