Workflow Reliability

Test with Sample Data

You can test a saved workflow with sample data, made-up values you give it, such as a contact named Dana Lee (action workflows.test_run). A test is a run labeled Test. The data tools and AI steps run for real on your sample, so you see what each step produced. An action runs only if it has a test mode, and then only in it. An action with no test mode is skipped and marked Skipped. Nothing is sent, charged, deleted, or changed anywhere. A wait is not waited for, and a step that would ask for your OK does not stop the test.

A test counts the same as any run: 1 for each step that did something, which here means an action that ran in test mode and each AI step. It is labeled Test in Run History and in what a workflow used. A test does not turn the workflow on, and it does not count toward a failure alert.

Run a Failed Step Again

When a run fails, you can run it again from the step that failed (action workflows.run_retry). The steps that finished stay finished and are not taken again. The step keeps the same key it had, so an action that already did its work knows it has it, and nobody is charged or emailed twice. It starts with its full retries again, and every try is still counted. Only a person does this, and only for a run that failed.

Automatic Retries

A step that fails is tried again after 1 minute, 5 minutes, 30 minutes, then 2 hours, up to the number of retries it has (3 unless the workflow says otherwise, at most 4). A "too many requests" answer from an outside service waits for the time the service asked for. A step that cannot work however often it is tried, such as a missing AI key or a value that is not a number, stops at once.

Failure Alerts

A workflow can tell you when its runs keep failing. Set how many failed runs in a row it takes (1 to 20, or 0 for never) when you save the workflow (action workflows.save). New workflows start with 3. Workflows made before alerts existed start with it off. When the count is reached, everyone who may manage workflows is told in Needs You Today, once, with the plain reason the last run stopped and a link to Run History. A run that finishes ends the count, so the next time it keeps failing you are told again. Tests and runs you stopped are not counted. Who is told follows the permission, so a custom role that has it is told, and a person who loses it is not.

Version History and Roll Back

Every time a workflow is saved, the version it replaces is kept with its name, steps, and starters, who saved it, when, and a plain sentence of what changed, such as "Added Format Text. Renamed it." (actions workflows.versions and workflows.version). Looking at an old version changes nothing. A workflow made before versions were kept shows its current version and keeps older ones from its next save on.

To roll back, pick a version (action workflows.rollback). The roll back is a new version that copies the old one, so the history only grows and nothing is removed. A workflow that is on stays on. If the old version uses something your account no longer has, such as an action it is not offered or a lookup table that was removed, the roll back is refused with a plain sentence and nothing changes.

Who Can Do This

People who may manage workflows (owners, admins, and anyone whose role allows it). AI agents, API keys, and assistants never use these actions.

Where to Find It

On a workflow, Test With Sample Data is beside Test Run, If It Keeps Failing is a card on the Build tab, and History is a tab with See It and Roll Back for each version. In Run History, a test shows a Test label, and a failed run has Run Again From the Failed Step.