In September 2026, Atlassian revised its usage-based pricing. Some things that were free or cost little now count toward your allowance. If that's not enough, you can use extra, but for several features, extra usage becomes billable in December. In many organizations, everyday setups nobody paid much attention to now risk becoming additional rows in their Atlassian bills. This article covers 8 of them, with suggestions on how to spot each one and what you might change.
TL;DR
- Automation now bills by the step. Every trigger, condition, action, branch, and loop counts each time it runs, so branches and loops multiply the cost of a single run.
- Using subtasks for checks costs more. A flow that creates detailed subtasks spends a step per subtask. If the checks don't need separate work items, an app like Smart Checklist for Jira can keep them in one work item and add them without automation steps.
- Runs that do nothing still cost. Broad triggers, global flows, and scheduled flows can burn steps without any visible result.
- AI actions in flows bill Rovo credits, not steps. An admin who watches only automation steps won't see this cost.
- Both Rovo Chat and enriched Rovo MCP calls use credits. Asking Rovo Chat for a summary the built-in feature gives for free, or letting Claude or ChatGPT pull enriched context across Atlassian apps through Rovo MCP, both consume Rovo credits.
- Tests aren't free by default. Sandbox automation and Rovo usage count toward your allowance, and Customer Service Management (CSM) agents tested in live channels cost $1.00 per resolution.
- Monitor where your allowance goes. The Usage tab in automation settings, performance insights, and Platform usage show which flows and features use the most.
What the Latest Atlassian UBP Update Changed: Key Points
Atlassian announced a usage-based pricing (UBP) extension on September 1, 2026. It covers five metered capabilities in Atlassian Cloud: Rovo credits, Automation steps, Assets objects, Bitbucket, and CSM AI agent resolutions. UBP works alongside seat-based pricing. Here are the key changes:
- Extra usage is on by default. For most organizations, usage continues past the included allowance on a pay-as-you-go basis, billed at published rates. If you buy usage packs, they're used up before extra usage starts. Organization and billing admins can set a monthly limit or turn extra usage off in Atlassian Administration.
- Allowances are pooled and reset monthly. Your allowance is shared across all users and subscriptions in your organization, not split by app or person. Enterprise plans and Collections get higher allowances than standalone apps on the same tier. It refreshes every month, even on annual plans, and unused allowance doesn't carry over. With extra usage off, metered features may be limited once you reach 100% of your allowance, until the next billing cycle.
- Extra usage billing starts December 3, 2026. This applies to most meters, but timing can depend on the capability and your billing cadence. Until then, you have time to review usage and set limits. The usage calculator can help you estimate what your organization will need.
- Automation counts steps instead of flow runs. Every trigger, condition, action, branch, and loop counts as one step each time it runs. Extra steps cost $0.50 per 1,000.
- Only admins get alerts. Organization and billing admins are notified at 80% and 100% of the allowance and of any limit you set. End users don't receive billing notifications.
Cost Drivers to Check Before December 2026
Complex Flows with Branches and Loops
For years, a rule (flow) run counted once, no matter how many steps it had. Under the new model, a flow's cost depends on how much it does on each run. Every trigger, condition, action, branch, and loop counts as a step each time it runs. A flow that looked cheap then can become one of your largest step consumers.
Branches and loops now require extra care, since they multiply steps. For instance, when a flow branches on related work items, the sub-branch runs against each work item. A loop repeats its actions for every value in a list, such as each label or component on a work item.
Consider the flow below as an example. It moves all the subtasks to Done when the parent is transitioned to Done. For a task with 5 subtasks, the flow will cost 7 steps: trigger with attached condition + one branch + 5 transition actions. More subtasks mean more steps spent.
How to spot it: Open Space Settings -> Automation and click the Usage tab to see which flows consume the most steps, and audit those.
In the automation audit log, examine each run to see the steps executed.
How to optimize your usage:
- Put the most selective condition inside the trigger, so runs that don't qualify stop before the branches.
- Whenever possible, Atlassian recommends narrowing each branch with JQL, so it acts only on the work items that need the change. In the example above, if 6 of a story's 8 subtasks are already done, a branch limited to open subtasks runs 2 transitions instead of 8.
- Remove components that no longer change anything.
Flows That Turn Each Check Into a Subtask
Say every Story should match the Definition of Done with 12 checks, from code complete and code coverage to a security review. A plain list in the description is easy to ignore: different people may handle different checks, and nothing stops the Story from moving on. So you set up a flow that turns each check into a subtask with an owner whenever a Story moves to In Progress.
Given the new automation consumption rules, creating extra work items costs more. The Create sub-tasks action can create all 12 subtasks in one step, each with only a summary. Adding a description or an assignee turns each subtask into a separate Create work item action. The flow still looks simple, but each run now costs 13 steps instead of 2.
How to spot it: In the Usage tab, sort the flows by automation steps and open the top ones. If one of them creates subtasks, that flow is the one to rethink.
How to optimize your usage: Keep subtasks for checks that need their own owner, workflow, or place on the board. If the checks are simply steps inside one work item, a checklist may cost less. For instance, Smart Checklist can add all items with one Import Checklist Template action, or without Jira Automation at all by using built-in automation capabilities.
Flows That Find Nothing, Fail, or Stop Halfway
Under the old model, only runs with a successful action counted, so empty runs were free. Now, Atlassian charges for every step that finishes in any of these states: it did its job, it ran but changed nothing, it hit an error such as a misconfigured field, it stopped before finishing, or the trigger found no matching work item.
That means your automation allowance is consumed with no visible result by:
- Flows with broad triggers like work item updated that can easily fire accidentally and find nothing to do.
- Global flows doing useful work in some spaces and firing uselessly in others.
- Scheduled triggers that fire on every interval, nights and weekends included, whether or not they find anything.
How to spot it: In the Usage tab, look for flows with many runs but only 1-2 steps per run. That pattern suggests runs that stop at the trigger or the first condition. Also, open performance insights and see the stats on runs by status. Flows with high No action or Some errors counts are the ones to examine in detail. The automation audit log will help you understand why a specific run failed.
How to optimize your usage:
- Narrower triggers. Replace Work item updated with Field value changed for the fields the flow actually reads. Add a condition inside the trigger - if the flow is triggered accidentally, it will consume 1 step (trigger) instead of 2 (trigger + condition).
- Narrower scope. Limit global or multi-space flows to the spaces that need them.
- Fewer chain reactions. Turn off Allow flow trigger on flows that don't need to react to other flows. Add a work type condition inside the trigger, so creation flows skip subtasks.
- Fewer scheduled runs. Run scheduled flows less often, or only during working hours with a cron expression. For flows that search with JQL, turn on Only include work items that have changed since the last time this flow executed.
- Cleanup. Fix or disable flows that rarely do anything. Check flows that reference removed fields, archived spaces, or deactivated users, since these may fail on every run.
AI Steps in Automation Flows That Consume Rovo Credits
AI-powered automation actions such as Use Rovo, Use agent, and Jira Coding Agent actions don't use steps, but they still cost credits. Jira Coding Agent sits in the Premium tier, where the credit cost depends on the computational effort. A flow can look cheap in steps while drawing on the same Rovo credits pool as your team's chats and agents.
Consider an example. Say a flow calls a basic Rovo agent that costs 10 credits per invocation. 5 runs use 50 credits, the monthly allowance of two Jira Standard users (2 x 25 credits per user per month).
How to spot it: In the Usage tab, sort the flows by the Rovo credits column to see which ones consume the most credits. If they eat up a significant portion of the allowance, reach out to owners to find out how important AI output is in their workflows.
How to optimize your usage: Ensure the output of costly AI actions is actually useful. Audit triggers and conditions to prevent avoidable runs.
Using Rovo Chat for Tasks That Free AI Features Already Cover
Not all AI capabilities cost credits. Rovo Search, definitions, and summaries in Atlassian cloud apps don't use Rovo credits. Rovo Chat Quick answers cost 10 credits per billable event, and Think deeper chats cost more. Thus, the same task can be free or paid, depending on where you run it. Using Search to find a Confluence page with product requirements (screenshot on the right) is free. Asking Rovo Chat to do the same (screenshot on the left) will cost 10 credits.
How to spot it: In Atlassian Administration, open Insights > Platform usage > Rovo credits and export the CSV. Filter the rows where Feature is Rovo Chat, then group them by Actor ID, sum Usage Count, and find top consumers. To match Actor IDs to names, export the list of users with their IDs as a CSV (Administration > Directory > Users > Export users). The latter is necessary, as neither Administration nor CSVs reveal how exactly those top users employ Rovo Chat, so you’ll have to interview them.
How to optimize your usage: Ensure the team knows which AI features are free. Point them to search and built-in summaries for quick lookups, and keep Chat for questions that really need to go deeper.
External AI Tools Using Rovo MCP
When AI tools outside Atlassian connect through Atlassian Rovo MCP, some of their calls consume Rovo credits. This is new. Before the credit model changed, Atlassian's docs listed the Rovo MCP server among beta features that didn't count against the Rovo credit allowance. Now, enriched Teamwork Graph calls consume allowance, with most of them costing 1-10 credits each. Still, installing Rovo MCP and any lookup or update inside a single Atlassian product are free.
This is easy to miss because the calls come from AI assistants like Claude or ChatGPT, not from Jira or Confluence. For example, a manager asks Claude to summarize what the team did this month. To answer, Claude may search Jira, pull related Confluence pages, and request extra context several times. Each of those enriched calls can use credits.
How to spot it: In Atlassian Administration, open Insights > Audit log and filter Activity by Invoked tool, under Atlassian MCP user actions. Each entry shows the date, location, actor, and activity, so you can see who uses external tools and how often. Then compare those users' Rovo credits consumption in Platform usage with the rest of the team.
How to optimize your usage: Ensure the people who connect AI tools know which requests are still free and which aren't. In Atlassian Administration, go to Rovo > Rovo MCP server and allow only the AI tools your teams use. Finally, set an extra usage limit for Rovo credits, so external tools can't push extra usage past a budget you choose.
Sandbox Activity That Draws From Your Allowance
Rovo credits and Automation steps used in a sandbox count toward your allowance. Assets objects in a sandbox don't. Bitbucket meters, such as build minutes, Git LFS storage, and packages storage, don't apply, since Bitbucket doesn't use sandboxes. Atlassian explains that AI and automation use the same compute resources in both environments. Note that you can't tell sandbox and production usage apart on your usage dashboard.
How to spot it: Compare usage spikes in Platform usage with the dates of your sandbox tests. Then check the sandbox automation audit log and sync with the team to spot the flows that run without anyone testing them.
How to optimize your usage: Disable flows copied into a sandbox unless you are testing them. Keep AI tests in the sandbox scoped and planned.
CSM AI Agents Billed From the First Resolution, Live Tests Included
Customer Service Management is part of Service Collection, together with Jira Service Management. Before the September announcement, Atlassian described the CSM AI agent as free to use on all plans, with per-resolution charges planned for later. The "later" now has a specific date: December 3, 2026. After that, each successful resolution will cost $1.00, since this feature has no allowance. A resolution counts when the agent gives a complete answer, and the system's LLM judge rates the conversation as resolved.
How to spot it: Track resolutions in Atlassian Administration > Insights > Platform usage. Each one costs $1.00, so this number is your bill for the month. Resolution packs bring the price down at volume.
How to optimize your usage:
- Decide how much you want to spend on AI resolutions per month, and set the extra usage limit to match. The default of 25,000 resolutions allows up to $25,000 a month.
- Keep all tests in the test panel and evaluations, which are free. Testing in a live channel is billed like any resolution.
- If the resolution rate looks satisfactory, but the message feedback chart in Reporting shows many thumbs-down ratings, review those conversations. Some of them may still count as resolved, so you'd pay for answers customers didn't find useful. Then refine the agent: fill the gaps in its knowledge or guidance. Optimize can help by analyzing past support interactions and suggesting content for missing knowledge. However, it needs at least 500 support interactions to run, and it is still rolling out, so check your release notes.
Conclusion: Audit the Behaviors, Not Just the Configurations
Misconfigured flows that spit errors and do nothing, a subtask for every check, AI steps on busy triggers, chatting away with Rovo like it is ChatGPT, experimenting freely with Rovo MCP, testing a CSM agent in a live channel. The problem with those goes beyond fixing existing inefficiencies. It means changing the mindset of both admins and users to avoid the same mistakes.
Flows that run and do nothing are more than log clutter, they're a liability. Complex automation and AI should not just produce some result, but bring value proportional to their cost. Using the right tool for the job is more pressing than ever. An experiment should have a carefully chosen scope and place. Treat your metered features allowance like any other budget: know where it goes, and spend it where it pays off.
A few practical habits help make these principles stick:
- Educate users on how features use allowance. Many costs start with everyday actions, not admin settings. Explain which Rovo features use credits and which are free, that AI tools connected through Rovo MCP can draw credits too, and that flows built in their spaces spend automation steps. End users don't get billing alerts (unless they are admins), so they won't notice on their own.
- Review flows before you enable them. Check the trigger and scope, and ask how often the flow will fire.
- Keep flow owners current. Every flow has an owner, usually its creator, and the Usage tab shows who it is. When people leave or change roles, transfer their flows so someone can still answer whether a flow brings value.
- Set limits to your testing. Use the CSM test panel and evaluations for AI agents, and switch off sandbox flows once a test is done.
- Review usage monthly. Check the Usage tab and Platform usage, and together with the owners of the top consumers, confirm whether the value matches the cost.
- Set extra usage limits. A cap per meter keeps an unexpected spike from turning into a surprise bill.
