What a real account audit checks, and where AI helps

The word "audit" doesn't mean much any more. In plenty of places, it means someone opened the account, took ten screenshots and found three settings that weren't ideal.

A serious review doesn't look like that, and it doesn't start in the account.

The four layers you have to check

Measurement. This comes first, because everything else depends on it. What counts as a conversion, whether anything is counted twice, whether there's a value, how long the window is, whether the numbers in the account match reality. If this layer doesn't hold up, you can't say anything about any other number in the account.

Structure. Whether brand and non-brand traffic are kept apart, whether there are exclusions, whether campaigns overlap, and whether the spend sits where the revenue is.

Offer and landing page. Whether the page delivers what the ad promised. This is the most common silent loss. The campaign's fine, the ad's fine, and the page is about something else.

Business context. What margin you work on, what a customer is worth, who you don't want. Without that, an audit can only tell you the account is tidy, not that it's profitable.

Where AI does speed things up

Working through the data and finding patterns. Categorising thousands of search terms, sorting ad copy by theme, collecting feed errors, spotting outliers. That saves hours, sometimes days.

Where it doesn't

In the judgement calls. A machine can tell you a campaign's cost per acquisition is above average. It can't tell you it's because customers from that segment are worth twice as much. For that, you need to know how the business works, and that information doesn't live in the account.

What to expect from an audit

Not a list of settings. Three things: where you're losing money, how much is at stake, and what order to fix it in. If there's no order at the end, the document is an inventory, not a plan.