In the Age of AI, XLAs are a Leadership Move

[#1]

When AI agents are involved, the person who used to notice the watermelon is gone. It’s time for the service management experience to move front and center.

---

Every SLA on the board is green. And yet CSAT and NPS are down.

Anyone who has run a service desk knows the feeling. The report says the team hit every target. The people on the other end tell a different story. Barclay Rae named this years ago and the name stuck: the watermelon SLA. Green on the outside, red on the inside.

That gap between the metric and the reality was survivable for one reason. A person sat in the middle and quietly closed it. An experienced agent heard the frustration under a “resolved” ticket, reopened it, and made the outcome right before it became a complaint. The dashboard never showed the save. The person made it anyway.

That person is now being automated away.

XLAs Aren’t New. What Changed Is Who’s Watching.

The experience level agreement is not a 2026 invention. Giarte and Marco Gianotten were making the case for it more than a decade ago, rooted in a simple idea: measure whether the service felt good to the person using it, not just whether the provider hit its internal targets. The instrument has been sitting on the shelf, well understood, for years.

Most service organizations treated it as a satisfaction upgrade. A nicer survey. A softer number to sit next to the hard ones. Optional.

That read was defensible when a human was absorbing the gap. It is not defensible anymore. What changed is not the instrument. It is the operating conditions around it.

Automate the Routine, the Early-Warning System Goes With It

Here is the mechanism most of the current conversation skips.

When you automate the routine tier of service, you do not just remove steps. You remove the person who used to notice when green did not match reality. The agent hit its handling-time target. No human heard the frustration underneath it. Nobody reopened the ticket. The watermelon is still a watermelon. There is just no longer anyone in the room to cut it open.

This is not speculation. In one large field experiment on agentic customer service, the AI cut average handling time and lowered customer satisfaction at the same time. Human agents preserved quality specifically in the cases where customers had voiced frustration. Take the human out of those cases and you take out the one who caught the problem.

Efficiency metrics stay green. Experience erodes. And nobody notices by accident anymore, because noticing by accident was a human function.

That is the shift. Experience now has to be measured on purpose, because it is no longer being measured for free.

An SLA Measures Our Numbers. An XLA Commits to Their Outcome.

The distinction is worth stating plainly, because it drives everything that follows.

An SLA measures the provider. Uptime, resolution time, first-response time. Did we hit our operational targets? An XLA commits to the recipient. Did the service help the person get their work done?

There is a second difference that matters more in an automated world. SLAs are built around sanctions. Miss the target, pay the penalty. XLAs are built around rewards and partnership. Reach the experience level, and the relationship improves. That flip is the point. Committing to an experience, rather than defending a number, is how a service organization signals it cares about the outcome and not just its own compliance record.

None of this means abandoning SLAs. Mature organizations run both, and anyone who tells you to throw out your SLAs has never carried a pager. The operational targets still matter. The XLA sits on top of them and answers the question the SLA cannot: did any of this land.

This Is a Leadership Move, Not a Metrics Program

Now the part that separates the leaders from the rest of the field.

As agents take over the routine work, everyone’s efficiency numbers start to converge. Handling times drop across the board. Ticket volumes to humans fall for everyone. When speed becomes table stakes, it stops being a differentiator. Experience is what is left.

There is a tension worth naming here. The vendors shipping the most autonomy are the quietest about experience as a governance layer. The platforms racing to hand more work to agents are not, for the most part, leading the conversation about how you know the agents are serving people well. That silence is an opening.

Committing to an experience level is a design decision, not a support-desk afterthought. It belongs in how a service is planned and built. As agents take over delivery and operations, that intent has to be set deliberately upstream, because there is no one downstream left to improvise it.

The service management leaders who move first, who commit to an experience level, instrument it, own it, and tie agent behavior to it, will distinguish themselves. Not because XLAs are novel. Because they will be the ones who can prove their automation is helping rather than quietly hurting, while everyone else is still admiring green dashboards. That is a competitive advantage for their organizations, and it is a credibility advantage for them.

Where to Start

The good news is that starting does not require a new platform or a big program. It requires a few disciplined moves, and they are the same whether you run internal IT or external customer service.

Start with a question, not a metric. Every XLA should begin with a plain statement of the experience problem you are trying to solve. If you cannot state it clearly, you are not ready to set a target. The number comes last, not first. And answer that question early, in planning and delivery, not at the support desk. Define what a good experience looks like before the service ships, so operations is measuring a commitment you set on purpose rather than one you reconstruct after the complaints arrive.

Instrument before you commit. You cannot set an experience target without knowing where you stand. Measure experience continuously, at the point of each interaction, and give it enough time to establish a baseline before you promise anyone a level. Practitioners who do this well talk in terms of a couple of months of data, not a couple of weeks.

Measure experience before and after each automation. This is the discipline the AI age demands. When you launch a bot, a virtual agent, or an autonomous workflow, watch what happens to happiness and lost time on either side of the change. This is exactly how you catch an automation that improves your handling time and hides a worse experience underneath it. It is the watermelon warning, turned into an operating practice.

Set targets as levels to reach, not thresholds to punish. Reward the team for getting to an experience level. Do not penalize them for missing one. The goal is to improve the experience, not to defend a line on a report. That posture is the whole difference between an XLA and one more SLA wearing a friendlier label.

Blend hard signals with soft ones. Technical and digital-experience telemetry on one side, sentiment, task completion, and lost time on the other. Neither tells the whole story alone. The value is in correlating them, which is precisely what modern tooling now makes feasible at full population rather than by sampling.

For teams in the Atlassian estate, the on-ramp already exists. Jira Service Management ships customer satisfaction surveys switched on by default, with results in the built-in satisfaction report, and richer experience reporting is available through Marketplace apps. CSAT on its own is a snapshot, necessary but not sufficient, so treat it as the first rung and not the finished agreement. It is enough to start baselining today.

On the customer service side, Atlassian is moving. The Customer Service Management app went generally available inside the new Service Collection at Team ’25 Europe, and the roadmap shows genuine momentum: live chat in open beta, omnichannel channels arriving, and Atlassian actively asking the Community what to build next. Atlassian’s “purpose-built, AI-first” language is directional, and it points the right way. The field read is more measured. The teams who have evaluated CSM closely are landing on promising, not yet ready to bet the operation on. So if you are running customer service on JSM today, the pragmatic move is not to wait. Watch the direction, and start measuring experience where you already are.

The Watermelon Is Still a Watermelon

The fruit never changed color. Someone was slicing it open before it reached you, catching the red under the green and making it right off the record.

That someone is being automated out of the loop, one routine tier at a time.

The instrument for replacing them has been on the shelf for a decade. The only open question is who picks it up first.

📚Further Reading

This article is part of a “Beyond ITSM” series. Stay tuned for more.

See also…

• Watertight not watermelon ‘SLAs’ — Barclay Rae on the concept he originated, and why green-on-red SLAs miss the outcome that matters, Barclay Rae
• The Watermelon Effect in Jira: How to use experience metrics to uncover true Customer Satisfaction — Bringing the watermelon problem, and experience metrics, into Jira Service Management, Atlassian Community
• Creating a Happier Service Desk: Combating Technostress via Leadership and XLAs — Why XLAs are a leadership matter, from the IT service management professional body, Service Desk Institute
• Introducing Atlassian Service Collection — Where Customer Service Management, JSM, Assets, and Rovo now sit together, Atlassian Community
• Track customer satisfaction feedback trends — How to watch experience move over time in JSM, which is the trend view baselining depends on, Atlassian Support

[Gary Blower]

Great article @Dave Rosenlund _Trundl_ - AI service automation is so challenging to measure effectively if it is purely based on transactions and SLAs.

ITXM and XLAs are very much a journey in that you have to carefully measure that starting baseline. The main mechanisms for measuring experience tend to be survey based and should avoid being too transactional -- in other words avoid a 'CSAT' or 'NPS' for every ticket -- as the initial experience may not reflect the long term sentiment and people get fed up with endless "how did we do" questions.

Organisations that have been successful with ITXM have utilised "voice of the customer" style interactions with carefully designed XLAs and (x-data) experience metrics.

[Dave Rosenlund _Trundl_]

Thanks for the addition, @Gary Blower.

I think you're spot on that ITXM and XLAs are a journey, not something you can measure by flipping on a new metric. I also like the way you've framed the risk of relying too heavily on transactional surveys.

One thing I think organizations will wrestle with is exactly what you're describing. If they lean too hard on ticket-level CSAT or NPS, they end up measuring how someone felt about a single interaction instead of how they experienced the service over time. That's where the voice-of-the-customer programs you mention really earn their keep. The best approaches I've seen treat CSAT as a tactical signal and XLAs as the strategic lens, with a deliberate connection between the two.

What's encouraging is that the technology is finally catching up to that way of thinking. We can now correlate sentiment, lost time, and experience data with operational telemetry at a scale that just wasn't practical a few years ago. But as you point out, none of that matters if the XLAs themselves haven't been designed with intent. Otherwise, you just end up with more data instead of better insight.

Appreciate you bringing the ITXM perspective into the discussion. It reinforces that XLAs aren't just about reporting differently. They're about defining the experience you want people to have, then measuring whether you're actually delivering on it.