How to avoid duplicate records in an AI workflow

A form can send the same event twice. If a workflow creates a new record on every arrival, a follow-up may be duplicated even though the person submitted only once. Decide what identifies the record before turning on automation.

Choose a key that matches the job

For a test, use a fictional identifier that stays the same when the same event is repeated. Decide whether two events with that identifier should update one record or represent separate transactions. Avoid using a person’s email as a universal key if one person can legitimately have multiple orders.

The APX flows guide describes a Save to table step: saving under the same key changes only the columns that step fills. Read table can retrieve one row by key or the latest rows. This is a building block, not a guarantee that every outside action is deduplicated.

Test both repetition and change

Send a harmless sample event, inspect the row, then send it again and confirm the intended fields. Try a second key and a missing value. Check the run log and the actual table, not just the success indicator.

If the flow also sends email or calls an external API, inspect that destination separately. A repeated event can still repeat an external side effect. Define retry and recovery rules before enabling live inputs.