Account audit template

An account audit is one row per finding. Each row separates what was checked from what was found, attaches the evidence, rates how much it matters, and names one person and one date for the fix.

The separation is the whole discipline. An audit that mixes observation with recommendation turns into an argument about the recommendation, and the observation — which was the useful part — is never settled.

Run it on a schedule rather than after a bad quarter. An audit commissioned because results fell is read as a search for blame; the same audit run every quarter is read as maintenance, and it finds the same things earlier.

Run it also before taking over an account, and before handing one over. Both are moments where what is actually true matters more than what was intended.

The most common way an audit fails is by mixing the two.

A row that says budgets are misallocated and should be restructured invites an argument about restructuring, and the argument never gets back to whether the budgets are in fact misallocated. A row that says prospecting carries four times the budget of retargeting while producing a third of the results, with the export attached, is a fact. What to do about it is a separate column and a separate conversation.

That separation is why this template has both a what-was-found column and a recommendation column, and why the found column should contain no verbs of instruction. It is also why the evidence column exists: a finding with an export and a date range behind it can be checked, and a finding that can be checked is one people argue about less.

The check-performed column asks what was actually looked at, written as something a person could do again.

This matters because the value of auditing compounds. One audit produces a list. Four audits, run the same way, produce a trend: which problems recur, which fixes held, which parts of the account are always fine and can be checked less often.

That only works if the checks are described well enough to repeat. Reviewed audiences is not a check. Listed every audience used in the last quarter, with its size, its source and the date it was last refreshed is.

Three levels are enough: something spending money wrongly right now, something distorting decisions, and something untidy.

The temptation is to inflate, because a finding at the top level gets attention. Doing it once works and doing it twice destroys the instrument — a list where everything is critical is read as a list where nothing is, and the genuinely urgent rows are lost in it.

The discipline is to reserve the top level for spend that is currently going somewhere it should not, and to be comfortable with a long tail of untidy findings that nobody will fix this quarter. Those still belong in the document, because the next auditor should not have to discover them again.

The impact column is what turns a finding into a decision. Where the cost can be worked out from the account, work it out and state the method: this much spend went to audiences that overlapped, over this window, calculated this way.

Where it cannot be estimated, write that. Some findings are real and unquantifiable — a naming inconsistency that makes reporting slow, an access risk that has not yet cost anything — and the honest version is more persuasive than a fabricated number. An invented figure is the fastest way for a good finding to be dismissed as exaggeration, and it takes the rest of the audit with it.

A recommendation that cannot be started on Monday is not a recommendation.

Write one action with a subject in it: reconcile the two conversion sources and document which is authoritative; remove the three audiences that overlap by more than half; rename the six campaigns that predate the convention. Not consider improving, not review the approach.

Then agree the owner and the date with the person who will do it, before the audit is circulated. An audit whose owners were assigned in their absence is a document that gets forwarded and not actioned, and the next audit reports the same findings as still outstanding.

The status column has three values, and accepted is the one that makes the list usable over time.

Some findings will not be fixed, for good reasons: the effort exceeds the benefit, the platform makes it impossible, the structure exists because of a constraint the auditor did not know about. Those rows should be closed, with the reason in the notes, not left open forever.

An audit where nothing is ever accepted accumulates a permanent backlog. Each new audit reports it again, everybody learns to skim the list, and the new findings are skimmed with the old ones.

Two moments deserve an audit outside the schedule.

Before taking over an account, because what you are inheriting is not what you were told, and the gap is worth documenting while you have no stake in it. And before handing one over, because an audit written by the outgoing owner is the most honest handover document there is, and it prevents the successor's first audit reading as an indictment.

In both cases the notes column earns its place. The history behind a decision is invisible in an account and impossible to reconstruct, and it is usually the difference between a finding and a misunderstanding.

Seven checks. Every finding separates what was found from what should be done. Every finding has evidence with a date range. Severity is honest, with the top level reserved for money going somewhere wrong now. Every impact is estimated with its method, or explicitly not estimated. Every recommendation is one action with a subject. Every row has an owner and a date, agreed with the person named. And every check is described well enough to be repeated next quarter.