The Meta Ads account audit: the ten checks that recover spend

A Meta Ads audit is supposed to answer one question: where is this account leaking money, and what is the cheapest thing to fix first? Yet most audits bury that answer under a screenshot tour of Ads Manager. This is the version that leads with impact: the checks that actually recover spend, in the order a senior buyer runs them. It is the Meta twin of how to audit a Google Ads account: same discipline, different platform.

Start with the signal you can't trust

Before you judge a single ad set, judge the measurement. On Meta more than anywhere, the number you optimise toward is only as good as the pixel and Conversions API feeding it. A misconfigured event, a browser pixel firing without a CAPI match, a purchase event that double-counts: any of these makes ROAS and CPA read like fiction, and no amount of budget tuning fixes a broken signal.

So the first check is not performance. It is whether performance is even measurable. Open Events Manager, confirm events fire once per action, check event match quality, and make sure the browser pixel and CAPI events deduplicate on a shared event ID. It is the same discipline behind leading with the verdict: say what you can stand behind, flag what you can't, and never manufacture confidence.

The ten Meta audit checks, ranked by recoverable spend

Ten families of check, ordered so the money moves first and the hygiene comes last. Each is a place spend hides on Meta specifically:

1. Pixel and Conversions API health. Events firing once, with the right value, on the right action? Confirm browser and CAPI events deduplicate, and check event match quality against Meta's Conversions API setup. A broken signal corrupts every optimisation downstream.
2. Attribution setting. Is the account on 7-day click, or the default 7-day click / 1-day view? Ad sets judged on different windows are not comparable, and view-through credit inflates the wrong winners. Judge it on settled data, too: an ad set read on day-one data looks pessimistic and gets paused before it has finished reporting.
3. CBO budget leakage. Under Campaign Budget Optimisation, Meta pools the budget and often starves a proven ad set to feed a cheaper but worse one. Check where the money actually landed against where the conversions did.
4. Audience Network placement drift. Spend quietly drifts to Audience Network and Reels at a fraction of feed performance: the Meta twin of the placement drift that hides inside a blended average.
5. Frequency and creative fatigue. Rising frequency with climbing CPM is the account telling you the creative is spent. CPM is the early creative-fatigue signal, well before ROAS drops.
6. Audience overlap. Ad sets bidding against each other in the same auction drive up your own CPMs. Run the overlap tool on your prospecting sets.
7. Retargeting exclusions. Are purchasers and recent converters excluded from prospecting and retargeting? Missing exclusions mean paying full CPA to re-buy customers you already have.
8. Catalogue and feed health. For commerce accounts, a stale or partly rejected catalogue silently caps Advantage+ Shopping reach. Check feed freshness, rejection rate and required fields.
9. Learning-phase status. Ad sets stuck in learning, or endlessly re-entering it after edits, never stabilise. Count how many are learning-limited and why.
10. Ad-account structure. Too many overlapping ad sets fragment the signal below the conversion volume Meta needs to learn. Consolidation is often free performance.

Rank, don't just list

The difference between a useful audit and a report nobody acts on is ordering. Every finding should carry an estimate of recoverable spend so the account owner knows what to do on Monday morning, not a flat list where a $10,000-a-month leak sits next to a cosmetic naming warning.

• Attach a number to each finding. “Audience Network drift: about $6,000 a month” (an example figure) beats “review placements”.
• Sequence the fixes. Editing an ad set can reset the learning phase; order changes so you don't undo one fix with the next.
• Make every figure verifiable. Each number should point back to where to confirm it in Ads Manager. Trust is the whole point.

Red flags: what a neglected account looks like

Some findings you have to calculate. Others you can spot in thirty seconds: the tells that an account has been left to run itself. If you see any of these, you have found recoverable spend before you have done a single sum.

• Frequency above 5 on a prospecting audience. A cold audience seeing the same ad five or more times in a week is fatigued and getting more expensive to reach. Rising frequency with climbing CPM is the clearest early fatigue tell there is: act before ROAS confirms it.
• No retargeting exclusions. If purchasers and recent converters are not excluded from prospecting, you are paying full CPA to re-serve ads to people who already bought.
• A backend versus Ads Manager revenue gap wider than roughly 15–20%. That is our working threshold for investigate versus ignore. When Ads Manager claims, say, $40,000 in sales and your store or GA4 sees $30,000, someone is optimising on a number that is not real. Inside that band, call it attribution noise; past it, your pixel and CAPI setup is the first thing to fix, not your bids.
• Audience Network eating a big share of a placement-optimised budget. When automatic placements let Audience Network absorb a large slice of spend at a fraction of feed performance, the blended ROAS looks fine while the leak sits hidden inside the average.
• Ad sets still in the learning phase after five days. Learning-limited ad sets never stabilise, usually because conversion volume is too low or edits keep restarting the clock. If half the account is perpetually learning, the structure is fragmenting the signal.
• One giant CBO campaign with every audience inside it. Pooling prospecting, retargeting and lookalikes under one budget lets Meta chase the cheapest conversions, usually warm retargeting, and take credit for demand you would have captured anyway.
• Default attribution, and nobody knows which window. An account reported on 7-day click / 1-day view while the team quotes numbers as if they were click-only is comparing ad sets that were never on the same footing.

Manual audit vs continuous audit

The format of the audit decides what it can catch. A person running this checklist once a quarter and software reading the account through the API every day are looking for the same leaks, but they see very different amounts of them.

Neither wins outright. A manual audit brings judgment a script can't; a continuous one catches the leaks that reopen the week after a manual audit ships. The strongest setup is a standing analyst for the high-drift checks and a person for the strategy calls, which is what an AI media buyer is built to be.

The real fix is a standing audit

The catch with a one-off Meta audit is that accounts drift. Frequency climbs daily, placements leak again, creative fatigues, CBO re-pools the budget overnight. A PDF from last quarter describes an account that no longer exists. The real fix is not a better template. It is a standing analyst that re-reads the account continuously and brings what changed.

Adgent is a team of AI agents working toward your marketing goals, and its ten specialists each read the account from their own angle. Several line up with this list: Measurement Integrity with the signal check, the Budget Allocator with CBO leakage, the Placement & Network Optimizer with placement drift, the Creative-Fatigue Scout with fatigue, the Audience-Overlap Detector with overlap and exclusions, and the Account Architect with structure. Every connected Meta and Google Ads account is read every day, and every decision carries an estimated monthly amount, so the list arrives ranked with its evidence and next move attached.