From Account Audit To Live Campaign

The short version

Most audits produce a document and no change. This post covers the path that actually matters: how to audit an account, decide what the findings justify, and get a change live in a form you can read afterward.

It is written as a workflow, because the failure is usually between the steps rather than inside them.

Why Audits Stall

An audit produces a list of observations. A campaign requires a decision. The gap between them is where most audits die, for three reasons.

The list has no priority, so everything looks equally actionable and nothing gets picked.
Findings are stated as problems rather than as changes, so someone still has to design the response.
The change requires creative, structure, and budget decisions that the audit did not make.

An audit that does not name the first change to make has not finished.

Step 1: Establish What Is True

Before diagnosing anything, settle the numbers.

Which source is authoritative for spend, conversions, and revenue.
What the attribution window is, and whether it changed during the period you are reading.
Whether tracking was intact for the whole window.
What the sample size is per campaign, and whether it supports any conclusion at all.

Skipping this produces confident diagnoses of measurement artifacts. A large share of "performance problems" are reporting problems.

Step 2: Read In A Fixed Order

Read the account in an order that separates causes, rather than scanning for anything that looks wrong.

Account economics. What can you afford to pay for a customer, and what are you paying.
Structure. Is the account arranged so the platform can learn, or is budget split across fragments that never exit the learning phase.
Delivery. Is spend going where you intend, and is anything constrained.
Creative. What is running, what is fatiguing, and how much genuine variety exists.
Landing and conversion. Does the post-click experience match what the ad promised.
Measurement. Are the numbers you are optimizing against the ones that matter.

The order matters because a creative problem and a structure problem look identical in a performance chart, and fixing the wrong one wastes a cycle.

Step 3: Convert Findings Into One Change

Turn the list into a ranked set of candidate changes, then take the first.

For each candidate, state four things:

The change, specifically enough to execute.
What you expect to happen.
How you will know within a defined window.
What you will do if it does not.

A finding that cannot be written this way is not ready to act on. That is a useful filter rather than a formality.

Prefer the change with the largest expected effect that you can read cleanly. A change you cannot attribute afterward teaches you nothing, regardless of outcome.

Step 4: Launch It As A Readable Test

The most common way a good decision becomes worthless is being launched in a way nobody can read later.

Change one meaningful thing at a time, unless you are willing to give up attribution.
Keep the naming convention so the change is identifiable months later.
Give it enough budget and time to exit the learning phase, or accept that the result is noise.
Write down what you expected before you see the result.
Set the review date at launch, not afterward.

Step 5: Read The Result And Close The Loop

Come back on the date you set and answer three questions: did the expected effect appear, is the sample large enough to believe it, and what does that justify next.

Then feed the answer back into the ranked list. The second change is chosen with more information than the first, which is the entire point of running the loop rather than executing an audit document.

Playad runs this loop against connected Meta, Google, and TikTok accounts, and queues the resulting change for approval rather than applying it directly.

Common Mistakes

Producing an audit document with no ranked first change.
Diagnosing performance before confirming the tracking was intact.
Changing several things at once and losing attribution for all of them.
Judging a test before it left the learning phase.
Never returning to the finding list after the first change.

Practical Rule

An audit is finished when it names the first change, what you expect, and the date you will read the result. Anything short of that is an observation list.