An Action Flow is a task inside your product that Adoption Agent performs for a user instead of explaining how to do it. You build the sequence of clicks and inputs once. When a user asks for that task in the Agent chat, the Agent offers to run it, and the user confirms.
Action Flows are part of Adoption Agent. Whether Adoption Agent is included in your subscription depends on your plan — check your billing settings.
How an Action Flow differs from a Flow
Both are built from your product's UI, and both are surfaced by Adoption Agent when a user's question matches. The difference is who does the work.
| Flow | Action Flow |
|---|---|---|
What the user sees | Step-by-step guidance layered on your app | The task being carried out for them |
Who performs the clicks | The user | Adoption Agent |
Best for | Tasks users should learn to do themselves | Repetitive tasks where learning isn't the point |
Reversible | Yes — the user controls every action | No, once a step completes |
When to use an Action Flow
Use an Action Flow when the task is repetitive, has a predictable shape, and there is no value in the user learning to do it. Adding a team member is a good example: most users don't need to know where that screen lives, they just need the teammate added.
Use a Flow instead when learning your product is the point. If a task is central to how your product works and users will do it repeatedly, guiding them through it once is more valuable than doing it for them.
Tasks to avoid automating
Do not build Action Flows for destructive or irreversible actions. Frontend actions cannot be undone once a step has completed, and the user may not realize what happened until afterward.
Specifically, avoid:
Billing changes — subscriptions, payment methods, plan changes
Deletions of any kind
Anything that sends external communication the user hasn't reviewed
Anything that needs a confirmation step to be safe
For these, build a Flow that guides the user to the right screen and lets them make the decision themselves.
What the user experiences
The user asks Adoption Agent for something in chat.
If an Action Flow matches, the Agent replies with a card showing the flow's public name and the note The action will be performed on your behalf.
The user clicks the card. Nothing runs until they do.
The Agent works through the steps in your app while the widget stays open alongside. The widget shows Execution in progress with a Cancel button.
Where a step needs information only the user has, the Agent asks for it in the chat, confirms what it received, and continues.
If a step can't be completed, the run stops and the user is told the step couldn't be finished.
When the Agent also has a knowledge-base answer for what the user asked, it leads with the answer and offers the Action Flow as a call to action below it. The user can read the steps, hand the task off, or both. This is the default, and it works the same way for Flows.
The user can click Cancel at any point during execution and the run stops there. Steps that already completed stay completed.
What Action Flows don't do yet
An existing Flow can't be converted into an Action Flow. The two are built separately.
An Adoption Agent recommendation is the only way an Action Flow starts. They can't be launched from an Announcement, a Checklist, a Resource Center block, or any other surface.
A completed action can't be reversed from Userflow. Cancelling mid-run is the only stop.
Credits
Action Flows use the same AI message credits as any other Adoption Agent response — there's no separate charge for running one. The Agent uses credits to understand the question and recommend the flow. Once the user confirms and execution begins, following the steps you built doesn't consume additional credits, no matter how many steps or inputs the flow contains.
See Adoption Agent usage limits and billing for how message credits work.
Data and privacy
Action Flows can collect information from users mid-run, and that input is handled the same way as any other Adoption Agent input. Userflow does not use AI inputs to train, fine-tune, or otherwise improve public large language models.
As with all Adoption Agent content, we recommend building Action Flows that don't require sensitive or private information, and instructing your users not to submit any in chat. Refer to our Terms of Service for additional details.