Team '26 Europe in Amsterdam: Jira Service Management sentiment analysis in triage queues
Mihai Perdum
Author
9 min readOctober 7, 2026
Key takeaways
Sentiment = Negative is valid JQL on both of our test sites, and Sentiment = Angry is rejected. Neither site had a single work item with a sentiment value, so we never saw the queue fill.
Jira Service Management sentiment analysis marks work items Positive, Neutral or Negative. It needs Premium or Enterprise with AI turned on, and older work items only get a value when the reporter comments again.
Atlassian's own queue example, updateDate < 14d, fails strict JQL validation. The form that works is updated >= -14d, which matched 4,693 of 23,280 work items on our test site on 7 October.
Jira Service Management allows 300 queues per service space per work category, and a queue's count stops at 999+. Atlassian's advice is several simple queues rather than one complex one.
Rob Hean's Configure queues for agent triage ran on Tuesday 6 October as a 60-minute, laptops-out workshop. It repeats on Thursday 8 October at 10:00 on Stage 3, Hall 2.
The team behind the show car in the hall has been called Atlassian Williams F1 Team since January 2026. Atlassian announced the title partnership on 11 February 2025, and Williams called it the biggest partnership deal in its 48-year history.
I went to Rob Hean's workshop on Jira Service Management queues at Team '26 Europe, and it was excellent, sentiment analysis included. That's the AI field that marks a request Positive, Neutral or Negative. So on 7 October we ran the sentiment part through Jira's strict JQL validator on our two test sites. Sentiment = Negative is valid, Sentiment = Angry is rejected, and neither site had a single work item with a sentiment value.
The only Atlassian text we found that says you can build a queue on sentiment is a 2024 Community post, written when the feature was in beta. The current doc doesn't mention JQL at all. And Atlassian's own example for keeping queues fast, updateDate < 14d, fails Jira's validation outright.
The clause is valid. What I can't tell you is whether a queue built on it ever fills. That depends on your plan, the per-space toggle and what your customers write.
The rest of Tuesday 6 October is below too: the Atlassian Williams F1 show car, a smoothie bike, the partners we met, a short Forge section and what we planned to watch in the Founder Keynote on Wednesday 7 October. This is part 3 of our series. Part 1 is the hub with dates and livestream times, and part 2 covered Partner Accelerate on Monday. The keynote announcements get part 4, on 8 October.
Five things we found that you can act on
Don't copy Atlassian's updateDate < 14d queue tip. It fails validation. Use updated >= -14d.
Split a slow, complex queue into several simple ones. There's room: the cap is 300 queues per service space per work category.
On Premium or Enterprise with AI on, turn on Customer sentiment analysis per space. Old work items stay blank until the reporter comments again.
Build one Sentiment = Negative queue, put it in a priority group so it refreshes on the regular sync, and count what it catches after a week.
If your Forge app adds JQL clauses, time them in a queue with real volume. Atlassian tells admins to reduce Marketplace JQL when a queue underperforms.
Configure queues for agent triage in Jira Service Management
At Team I sat in on his Configure queues for agent triage, Tuesday 11:00 to 12:00 on Stage 2, Hall 2. It's a Learning session, listed as "an instructor-led hands-on workshop - Laptop required". Sentiment came up too, though it isn't in the published description.
Rob Hean's workshop on Tuesday, from the back rows. Look at the tables: it was a laptops-out session, and the three agenda items are on the slide.
If you missed it, it runs again on Thursday 8 October at 10:00 on Stage 3, Hall 2. Otherwise, this is the part of Atlassian's docs I'd read first.
Queues are filters. Atlassian calls them "a kind of filter for your requests", by type, status or a JQL statement, and you need to be a space admin to create a new queue.
There's a hard cap: "The total number of queues is limited to 300 per service space per work category across all sections." And a queue's count refreshes up to 999 work items, then just shows 999+ (What are queues?, updated 29 May 2026).
What refreshes depends on where the queue sits. Queues in Starred and in your priority groups refresh on a regular sync. Queues under View all queues don't. One queue can sit in several groups. Starring only changes your own view: starred queues "appear only in your view and won't impact your team's work".
Rovo can also write the JQL. In Rovo Chat, /manage-queue generates the filter and the column layout. It can't delete queues or queue groups.
Atlassian's own date example doesn't validate
Atlassian's best practices for managing queues at scale is blunt, and I agree with most of it. "Often it is better to have multiple simple queues than one complex one." Text search is "the worst of all". And it says "JQL clauses provided by Marketplace apps installed on your site can sometimes be problematic. Reduce their use if a queue is underperforming." That applies to app vendors, us included.
Its tip to "Show only recent work items, e.g. use updateDate < 14d in your JQL" is the one I'd fix. Don't copy it. On our wolfaenpak test site, strict validation answers "Field 'updateDate' does not exist or you do not have permission to view it." The field is updated, alias updatedDate. Spell it updatedDate < 14d and it validates, but it matched all 23,280 work items, because 14d without a minus sign means 14 days in the future. On 7 October, updated >= -14d matched 4,693. That's the form I'd put in a queue.
Putting Jira customer sentiment into a queue with JQL
Atlassian's doc, View customer sentiment on work items (updated 29 July 2026), says the feature "uses AI to determine how customers are feeling based on the content of their request and their recent comments". The result is Positive, Neutral or Negative, shown in the Sentiment field on the work item. It's "only available for Premium or Enterprise customers that have AI enabled for Jira Service Management", and AI features aren't in the Atlassian Government environment.
Switching it on is per space:
Next to your space name in the sidebar, select More actions (•••), then Space settings.
Select Features.
Under Work item view, turn on Customer sentiment analysis.
New work items get a value automatically. Existing ones get one only "when the reporter adds a new comment". The doc names the reporter, not the agent. So on day one, your old backlog is blank. I'd tell the team that before they judge the queue. Languages? The doc doesn't say. I'd test yours before trusting it.
On status, the record is messy. Atlassian announced it in beta on 15 August 2024. From the week of 16 December 2024, Atlassian's Cloud changes blog listed Customer sentiment among the Jira Service Management AI features that "are now generally available", marked Rolling out. Today's doc carries no status label at all. That 2024 beta post is also the only Atlassian text we found that says you can "use the sentiment field in JQL to create custom queues". The current JQL fields doc doesn't list it.
So we asked Jira directly. Strict JQL validation on both test sites accepted Sentiment = Negative and Sentiment in (Negative, Neutral). The JQL autocomplete offers exactly three values: Negative, Neutral, Positive. A made-up value fails with "The following sentiment IDs are not valid: Angry". The field is searchable and orderable, with =, !=, in, not in, is, is not and the range operators.
Our wolfaenpak test site, 7 October. Top: Negative is a valid value and the search runs, with no results. Bottom: Jira rejects any value that isn't one of its three.
Now what we couldn't prove. Both sites returned 0 work items for Sentiment is not EMPTY. Neither is a live desk with customers writing in, and Jira's API reports the plan only as "PAID", not which one. So I won't read anything into that zero about how often sentiment fires. We also don't know whether it uses Rovo credits. Atlassian's Rovo credits page doesn't say.
One trap from our first run. A logged-out session got "Field 'Sentiment' does not exist or this field cannot be viewed by anonymous users". The field was there. The login wasn't.
On Premium or Enterprise with AI on, I'd do this. Turn sentiment on and make a queue with Sentiment = Negative AND updated >= -14d. Put it in your triage priority group. After a week, count what it has actually caught.
Illustration, generated locally with FLUX. Sentiment gives a frowning request a value a queue can filter on. Somebody still has to open the envelope.
The Atlassian Williams F1 car and the sim racing rigs
I saw the car in a glass-walled hall of the RAI. Atlassian's event page promised "the official Atlassian Williams F1 Team show car" and racing simulators, and both were there.
The show car at the RAI. The sign behind it, Step into the driver's seat, was for the simulators.
Atlassian announced the partnership on 11 February 2025, as "the Official Title Partner and Official Technology Partner" of a team "known from today on as Atlassian Williams Racing". Williams' own release adds Official Collaboration Software partner and calls the multi-year deal "the biggest partnership deal in Williams' 48-year history". Neither names a figure. In November 2025 Williams announced the new name: "From January we will become Atlassian Williams F1 Team". Its 2026 car is the FW48, driven by Alex Albon and Carlos Sainz.
The part I care about as an Atlassian admin is what the team runs on. Atlassian's Williams page says Williams "has moved from scattered tools and spreadsheets to Jira Service Management and Assets", and uses Jira Product Discovery to "connect ideas to impact". A December 2025 case study has them putting simulation knowledge in Confluence, with Rovo so engineers can "ask complex questions, surface relevant findings, and get recommendations". There was a Williams breakout on Tuesday too, Designing for flow, at 11:00 in Elicium 2. It ran at the same hour as the queues workshop, which is where I was.
One of the racing simulators, cropped to the wheel and the screen. The lap leaderboard is left out on purpose: it had other people's names on it.
Illustration, generated locally with FLUX. Atlassian's event page calls the partnership a demonstration of choreographed teamwork. Four mechanics, one stopwatch.
Bring a problem, leave with a plan (and a smoothie)
Atlassian Professional Services had an area with the sign "Bring a problem, leave with a plan", a whiteboard of diagrams and a bike that powered a blender. There's no official page for it that we could find.
The bike-powered blender at the Professional Services area. Pedal, and the blender runs.
Partners we met: VMotion (Cardinal Ruler), Kantega SSO and SmartBear (BearQ)
We talked to three vendors whose apps a Jira admin might actually install. What each one makes, from their own pages:
Cardinal Ruler comes from VMotion IT Solutions, an Atlassian Gold Solution Partner based in Limerick. It's for moving Jira configuration between sites under change control. You design a change on one site and it pre-flights it against the target. Approvers sign off the frozen plan, and the record goes into a Git repository you own. It runs on Forge. Jira Cloud only. Version 2.6.0 shipped on 2 October. Per its own site, workflows deploy today. Permission, notification and work type schemes are already packaged and pre-flighted, and deploy for them "arrives in early Q4'26", so it's due about now.
Kantega SSO, based in Trondheim, was set up as a subsidiary of the Kantega consultancy in 2018. Its single sign-on apps (SAML, OpenID Connect, Kerberos) are for Data Center. On Cloud, its apps include the User Management & License Optimizer, whose Jira version 7.91.0 shipped on 21 September. Per its listing it "aids sysadmins by analyzing user activity, auto-configuring access rules, and managing temporary access for external users".
SmartBear had BearQ with them, and I thought it was excellent. SmartBear launched it on 18 March 2026 as an agentic QA system. On 16 September it became an assignable agent in Jira. You can assign work to it, mention it in comments and add it to workflow transitions, and it can record tests and results in Zephyr. Teams set the autonomy level "to meet their desired level of human oversight". The assignable agent is what I'd look at first. You can try it now.
We also had a short chat with the team from Spirit Experts.
What Forge builders can take from the queues session
We checked this section against Atlassian's pages on 7 October. An app gets into a JSM queue's contents through JQL. The jiraServiceManagement:queuePage module adds an item to the Apps section of the queues sidebar and renders its own page. It doesn't add anything inside a queue, and there's no sentiment hook. A queue's conditions are JQL, so a searchable Forge custom field, or a JQL function, is how your app shows up in triage. JSM UI modifications for the portal and agent views are Preview, not GA. We built on the portal side once, in our portal property panel tutorial.
And that same scale guide says Marketplace JQL clauses "can sometimes be problematic" and tells admins to reduce their use when a queue underperforms. If your app adds JQL, make it cheap.
Separate from the queues session, here are four Forge changelog entries part 2 didn't mention. Display conditions on the sub-pages and sections of Jira admin, global, project, project settings and personal settings pages landed on 29 September. The dashboard:filters module went into EAP on 30 September. Manual packaging for functions and UI Kit has been in Preview since 28 September, and the TypeScript bundler EAP it replaces has been removed. And on 7 October the refreshed tag, lozenge and badge look became the default, in UI Kit too, with no feature flag. EAP isn't open to everyone. Preview isn't GA. If you moved off Connect recently, also read how Cloud Fortified still probes your old Connect URL.
What we expect from the Founder Keynote
Written in the early hours of 7 October, before the keynote. The Founder Keynote, "Your business knows better", was listed for Wednesday 7 October, 09:00 to 10:15 CEST on the Main Stage. Mike Cannon-Brookes, Sherif Mansour, Tamar Yehoshua, Taroon Mandhana and Josh Miller of The Browser Company are listed. Before it started, the description promised "the evolution of Atlassian's platform, AI, and real-world stories", and the only post on Atlassian's blog on Tuesday was about its OpenAI partnership.
I'm watching for three things. Anything new for Forge and the Marketplace. Where Rovo agents land inside JSM. And any status change on the features above. Part 4 will list only what Atlassian publishes, with its label. The Founder Keynote stream needs no registration. Some other keynotes do. Atlassian's FAQ and home page disagree on whether it starts at 08:30 or 08:45 CEST, so check the hub for both.
