In one sentence
An audit trail is a record of the actions in a process: what happened, when, by whom, and on what basis.
The simple model
Every time something meaningful happens in a workflow, a decision, a change, an action, the audit trail adds an entry. Over time it becomes a history you can read to reconstruct exactly what occurred, in order.
The key property is that it adds rather than overwrites. You can see that a value was changed from A to B by someone at a certain time, not just that it is now B. This is often called append-only.
Why you care
An audit trail is what lets you answer "what actually happened?" with evidence rather than memory. That question comes up constantly: a customer disputes a decision, a result looks wrong, a regulator asks, or you want to learn from a failure.
For AI workflows it is essential. When an AI step produces a wrong result, the trail should show what it received and what it produced, so you can tell whether the problem was the input, the model or the rule around it. Without that, you are guessing. It is what makes reviewing failed cases and documenting incidents possible.
A small example
A pricing workflow kept only the current price. When a customer was overcharged, nobody could see when or why the price changed. Adding an audit trail, old value, new value, who, when and why, meant the next such question was answered in minutes, and a recurring cause was spotted and fixed.
What this does not solve
An audit trail records what happened; it does not make what happened correct, and it can hold sensitive information. Capture what you genuinely need for review, protect it appropriately, and remember it is a tool for trust and learning, not a substitute for good checks.
