For years, customer service has been measured on activity. Tickets handled, average handle time, first reply time, deflection rate. These numbers were easy to count, so they quietly became the job.
AI is breaking that model. When an AI agent can answer thousands of questions at once, activity stops being scarce, and so it stops being valuable. At the same time, AI has a running cost that grows with every conversation, every summary and every automated step. Leaders are now asking a fair question: what are we actually getting for that spend? The only honest answer is outcomes. Did the customer's problem go away?
We agree with the direction. But after years of implementing Zendesk for Danish and Nordic companies, we have learned that the hard part is not deciding to invest in outcomes. It is knowing what an outcome actually is.
A closed ticket is a claim, not a fact
Every service system has a status called solved. It is set by an agent, a trigger or, more and more often, an AI agent. It means someone decided the conversation was over. It does not mean the customer agrees.
We see the gap every day. A customer asks about a delayed parcel, gets a polite and accurate answer, and the ticket closes. Two days later the same customer calls, more frustrated, because the answer was right but the parcel still hasn't arrived. In the reporting, that is two solved tickets. For the customer, it is one unsolved problem.
Example
Two solved tickets. One unsolved problem.
When people did all the work, this blind spot was tolerable. Experienced agents could sense when something was off. When AI handles the first line at scale, nobody senses anything. If "solved" is the outcome you reward, you will get a lot of solved, and you will pay for every one of them. You will not necessarily get satisfied customers.
Three things outcome-based AI demands
1. Define the outcome from the customer's side. Before you switch on an AI agent, write down what a good outcome looks like for your three or four biggest contact reasons. "The customer knows when the parcel arrives" is different from "the customer received the tracking link". The first is an outcome. The second is an activity dressed up as one.
2. Measure the gap, not just the score. CSAT and NPS are useful, but a score alone tells you little. In our own work we compare how the customer rated a conversation with how the agent would have rated the same conversation. Where the two disagree, there is almost always something to learn: a policy that sounds fine internally but lands badly, an answer that is correct but not helpful, a process that forces the customer to come back. That perception gap is where outcomes are won or lost, and it applies to AI agents just as much as to people.
3. Look at the whole journey, not one channel. Many outcomes fail between channels. The bot answers, the customer calls anyway, the phone call never makes it into the ticket history, and the frustration ends up on Facebook or Trustpilot. If your AI only sees the chat, it will report a success that never happened. Calls, social mentions and reviews belong in the same picture as email and chat. Otherwise your outcome data is only part of the truth.
People become the owners of outcomes
There is a fear that outcome-based AI simply means fewer people. Our experience is different: the role changes. Agents spend less time typing the same answer for the hundredth time, and more time on the cases where judgment, empathy and authority matter. Team leads stop managing handle time and start managing quality, reviewing where the AI resolved and where it only closed, and feeding that back into knowledge, rules and processes.
That is a better job. But it does not happen by itself. It takes training, clear ownership of the knowledge base, and leaders who are willing to look at numbers that are less flattering than a deflection rate.
Start smaller than you think
You don't need a perfect outcome model to begin. Pick one contact reason. Agree on what resolved means for the customer. Measure it for a month, with AI and without it. Then read the conversations where the customer and your system disagree. You will learn more from those few hundred cases than from any dashboard of handled tickets.
Try it
What does a problem that is actually resolved cost?
Move the sliders using your own numbers. The calculation is indicative and shows how repeat contacts change the price of a real outcome.
Lowering repeat contacts is often the fastest route to better AI economics, not more automated answers.
At Available we like to say that we connect cables and hearts. Outcome-based AI sits exactly where the two meet. The technology can now do a remarkable share of the work. Whether that work actually helps the person on the other side is still something we have to define, measure and take responsibility for. That is the real currency.
Go further
- Take the AI readiness test to see whether your data, knowledge and processes are ready for AI agents.
- Compare your customer service with your industry in our CX benchmark.
- Read our take on CX Trends 2026, which this article builds on.
Want to measure what your AI actually resolves?
At Available we help companies define the outcomes that matter, connect every channel in Zendesk and measure the gap between solved and resolved. Let's talk about where to start.
Contact us