Summary

  • Atlassian measures automation execution in steps, specified AI and context work in Rovo credits, and a completed Customer Service Management request in resolutions. The units are separate even when they belong to one software workflow.
  • In an Automation flow, an executed trigger or ordinary action can count as a step while a Use Rovo component is billed through the Rovo meter instead. Customer Service Management adds a third, outcome-based charge only when its agent resolves a request under Atlassian’s stated test.

The first billable event in an Atlassian automation can be a trigger that finds nothing to do. The trigger ran, so it counts as one Automation step even if it returns NO_MATCH. If the flow then invokes Rovo, that AI action does not count as a second Automation step: Atlassian places agentic components such as Use Rovo in the Rovo-credit meter. One run can therefore draw on two allowances, but the same AI action is not supposed to be charged as both a step and a credit event.

That distinction is easy to miss when “usage-based pricing” is treated as a single surcharge. Atlassian is adding a usage layer to per-user subscriptions, but the layer has several units. Automation steps count executed parts of a workflow. Rovo credits account for specified AI interactions and enriched Teamwork Graph context. Customer Service Management resolutions count a qualifying customer outcome. The price of a unit cannot be read without its definition.

A step is an execution event

Atlassian defines an Automation step as one executed trigger, condition, action, branch or loop. Its rules include more than successful changes. A step that ends with no action, some business errors, or an abort is still chargeable; so is a trigger that finds no matching item. A throttled run in which no step executes and system-side errors are excluded.

That makes the meter sensitive to workflow design and run frequency, not just to whether a user sees a changed Jira issue or Confluence page. A flow with several conditions and branches may execute several metered components before it reaches an action. The relevant count is the work the system ran, according to the documented status rules, rather than the number of useful results that the operator intended to produce.

The allowance is shared across eligible Atlassian apps at the organization level. Atlassian’s example combines 75,000 monthly steps from 100 Jira Premium users with 10,000 from 100 Confluence Standard users, producing 85,000 steps for the organization’s Automation meter. If the organization exceeds its allowance and permits extra usage, Atlassian lists a price of US$0.50 per 1,000 steps. If extra usage is disabled and the step limit is reached, Automation flows stop until the allowance resets.

Rovo measures a different input

Rovo credits are not another name for Automation steps. Atlassian says a basic Rovo event uses 10 credits, while premium interactions consume a variable amount. Calls that return enriched Teamwork Graph context may also use credits, with usage depending on the action and its complexity. Eligible paid subscriptions contribute credits according to app, collection and plan; credits from those subscriptions pool within the organization and reset monthly without rollover.

Atlassian lists extra Rovo usage at US$0.01 per credit. At that rate, ten credits correspond to US$0.10 only when the event is above the included allowance and is billed as extra usage. It is not a universal price for every prompt, agent task or automation run. A premium interaction can use a variable number of credits, while a flow’s ordinary triggers and conditions remain in the separate step count.

The split matters for an administrator setting limits. If a flow uses a Rovo component, a credit allowance can be consumed while the surrounding steps draw down the Automation allowance. The two balances are not presented as one transferable pool. Atlassian also documents meter-specific controls; a limit reached in one meter can affect service differently from a limit reached in another.

A resolution is a completed outcome

Customer Service Management uses a third unit. Atlassian says a resolution is chargeable only when its AI agent gives a complete answer, does not hand the conversation to the support team, and the system’s language-model judge decides the request is resolved. It is an outcome test, not a count of every message or every model call.

The subscription includes no allowance for this meter. Atlassian lists US$1 per successful resolution from the first one, with annual packs offering volume discounts and a lower published rate at larger commitments. The resolution charge is explicitly separate from Rovo credits and Automation steps. An AI-assisted workflow therefore does not automatically create a resolution charge: that applies only to a Customer Service Management request satisfying the documented completion test.

For monthly plans, Atlassian documents a default usage limit of 25,000 resolutions. When a customer has exhausted any resolution pack and reaches that limit, the AI agent stops providing direct resolutions, tells the customer it is raising a request, and creates a work item. The service changes from autonomous resolution to human follow-up. That is a different operational boundary from an Automation flow stopping when steps run out, or a credit-consuming Rovo action pausing under its own settings.

The date is known; customer bills are not

Atlassian’s support pages say extra usage billing for these meters begins on 3 December 2026. On 10 October, that date is still ahead. The product rules can be described; customer overage totals, usage concentration and responses to the new controls cannot yet be observed. Atlassian’s Q4 FY2026 shareholder letter reported that Rovo assisted actions grew more than 50% quarter over quarter, but an activity measure is not a credit total, a billed amount or evidence of future revenue.

The economic change is therefore a change in what the platform makes countable. A workflow execution, an AI/context event and a support resolution each have their own numerator. They draw on separate allowances and expose distinct choices about buying more capacity, enabling extra use or accepting a pause. The posted rates make those units legible; they do not make them interchangeable.

Sources