The automation that quietly duplicates your data (and how to catch it)

An automation triggers, partially completes, and then fails or times out. The retry logic, built in by the platform or added deliberately, reruns it from the start. If the workflow isn't designed to recognize it already did half the job, it does the whole job again: a duplicate contact, a double-counted order, a lead scored and routed twice.

This is one of the most common, least visible failure modes in automation, because nothing looks broken. The workflow shows a green checkmark on both runs. The problem only surfaces weeks later, as a sales team wondering why the pipeline count doesn't match reality, or a customer getting the same welcome email twice.

The fix is a concept borrowed from distributed systems: idempotency. Every workflow that creates or modifies a record should check first whether that exact action has already happened, an idempotency key, a duplicate-check query, something that makes rerunning the workflow safe rather than merely rare.

It's a few extra minutes of build time per workflow, and it's the difference between a system that's safe to retry and one where every failure quietly seeds new data problems that surface somewhere else, weeks later, attributed to the wrong cause.