How to Run a Paid Media Audit (The 40-Point Method)

A paid media audit is a structured review of an ad account in four layers — measurement, structure, creative, and efficiency — run in that exact order, because each layer decides whether the numbers in the next one mean anything. Our version checks 40 points across those four layers, takes one to two weeks on a live account, and ends in one of two verdicts: remediate the account you have, or rebuild it deliberately. Run it before any scaling decision, because scale multiplies whatever the audit would have caught, waste included.

What does a paid media audit actually check?

Forty points, four layers, fixed order. Measurement (points 1–10) asks whether the account's numbers can be trusted at all. Structure (11–20) asks whether the architecture concentrates enough signal for the platform algorithms to learn. Creative (21–30) asks whether the auction is being fed enough variety to stay competitive. Efficiency (31–40) — deliberately last — asks whether the outputs are fairly priced against market benchmarks.

The order is the whole method. An account with broken conversion tracking will happily report a beautiful ROAS on phantom data. A fragmented structure makes strong creative look mediocre, because no ad set accumulates enough conversions to exit learning. Audit efficiency first and you spend two weeks optimizing a mirage. This guide leads the diagnostic set in our growth marketing guides collection, and everything below is the sequence we run on real engagements.

Layer 1: Can you trust the numbers?

Measurement fails more audits than the other three layers combined, which is exactly why it goes first. Five checks carry most of the weight:

1. Conversion definitions reconcile with reality. Pull platform-reported conversions and your backend or CRM truth for the same 30 days. Disagreement within a tolerance you chose deliberately is fine; disagreement nobody can explain is finding number one.
2. Attribution overlap is acknowledged. Platform-attributed revenue summed across channels routinely exceeds real blended revenue, because every platform grades its own homework with its own windows. The blended guardrail is MER — total revenue over total ad spend — and its absence from reporting is a reliable sign decisions run on flattering numbers. The ROAS glossary entry unpacks why the platform number and the bank account disagree.
3. Signal loss is quantified. Ad blockers and Safari's tracking prevention strip a meaningful share of client-side events, and server-side tracking typically recovers 15–30% of otherwise-lost conversions. An account bidding without it has been training algorithms on partial data, usually visible as smart bidding drifting toward the segments that track cleanly.
4. Windows and revenue definitions match. A 7-day-click campaign compared against a 28-day-click campaign is an apples-to-oranges error hiding in plain sight, and gross-versus-net-of-returns revenue quietly changes every efficiency number downstream.
5. UTM hygiene passes a spot check. Sample twenty live ads. If tagging conventions drift by channel or by manager, analytics has been misfiling revenue for months.

The one-afternoon version of this layer has its own runbook; the audit version goes deeper on reconciliation. Done looks like: platform and backend within tolerance, MER on the dashboard, signal recovery scoped and priced.

Layer 2: Does the structure let the algorithm learn?

Modern ad platforms are machine-learning systems that trade data density for performance, so structure is really a question about signal concentration.

• Fragmentation. Twenty campaigns at small daily budgets split conversions that four campaigns would concentrate, so every ad set idles in learning. Consolidation is the single most common fix our audits prescribe.
• Brand and non-brand separation. Branded search runs 15–30%+ CTR against the 6.42% cross-industry median (WordStream/LocalIQ, 2024), and it harvests demand you already earned. Blending it into account-level ROAS is how mediocre prospecting hides inside a great-looking average.
• Explicit channel roles. Search captures demand that already exists; social manufactures demand for people who never searched — the division of labor our Google Ads vs Facebook Ads comparison covers in depth. Accounts fail this check when both channels chase the same bottom-funnel conversion with no stated role for either.
• Waste controls. Negative keyword lists touched within the quarter, placement exclusions on display and PMax inventory, geography settings that match where the business actually ships or serves.
• Budget follows marginal return. Money tends to sit where returns were good last year rather than where the next dollar performs best now. The audit flags every budget line nobody has re-justified in six months.

Layer 3: Is creative getting enough at-bats?

Creative explains the majority of paid-social performance variance in platform and agency studies — more than bidding, more than audience settings — yet most teams give it a fraction of the attention they lavish on bid strategies. This layer checks volume, variety, and freshness:

• Velocity. Count net-new creative concepts from the last 90 days. Accounts shipping two per month get out-learned by accounts testing dozens, because the auction rewards the portfolio with more attempts.
• Format spread. UGC and native-feeling formats cut CPAs 20–50% versus polished static in head-to-head tests (platform and agency studies). An account running only studio assets has an entire price tier it has never tested.
• Fatigue signals. Rising frequency with decaying CTR on the same audiences is creative wearing out on schedule. The fix is a production pipeline rather than a pause button.
• Hook diversity. Ten ads opening with the same value proposition are one ad with nine understudies.
• Competitive context. Our free Ad Library Explorer pulls what competitors are running and how long each ad has survived — longevity is the closest thing the public libraries give you to a performance signal.

Layer 4: How do the metrics compare against benchmarks?

Only now do the efficiency numbers earn attention, because only now do they describe something real. Benchmarks calibrate rather than judge: the useful question is whether your costs sit inside the plausible range, remembering that the spread between average and top-quartile accounts on the same channel runs 2–4x.

Three adjustments keep the comparison honest. Compare year over year with inflation in mind — CPCs on the major auctions rise roughly 10% a year (directional, WordStream year-over-year studies), so a flat CPC is quietly a win. Compare in-season, since Q4 swings CPMs by ±30% or more. And convert everything to margin terms before judging: break-even ROAS is 1 divided by contribution margin, so a 3x target means opposite things at 30% and 60% margins. Our free ROAS calculator runs the break-even math on your real margin, and the full channel-by-channel dataset lives in our Paid Media Benchmarks report.

When do you rebuild instead of remediate?

The audit ends in a verdict, and the deciding question is whether the account's accumulated history helps or hurts from here.

Remediate when measurement problems are recent, structure needs consolidation rather than re-architecture, and the conversion data underneath is fundamentally sound. Most accounts land here: the fix list is long and the foundation is usable.

Rebuild when the algorithms have been learning from broken signal for months — wrong conversion events, double-counting, a pixel fed by bot traffic — or when fragmentation runs so deep that fixing it means rebuilding anyway. Rebuilt accounts eat a learning-phase dip measured in weeks, which is the honest price of clean signal. Budget for the dip explicitly; teams that rebuild without warning stakeholders tend to abandon the rebuild mid-dip and inherit the worst of both worlds.

The tiebreaker is time-to-trust: if remediation cannot produce trustworthy numbers within a quarter, rebuilding is usually cheaper than governing on fog for another two.

What do you do with the findings?

Rank every finding by estimated recovered dollars, then fix in audit order: measurement first, because nothing downstream is provable until the numbers are real, then structure, then the creative pipeline, then efficiency pruning. A useful 90-day rhythm: reconciliation and tracking repairs in the first month, consolidation and channel-role definitions in the second, creative velocity and benchmark-driven pruning in the third.

Audits also reliably free budget, and the follow-on question is where it goes. The efficiency layer often shows owned channels underfunded relative to paid — if email carries a weak share of revenue, our deliverability runbook is the highest-leverage next project, with a domain warmup schedule if a dedicated sending domain is part of the plan. The newest surface deserves a line item too: assistants now answer a real share of buying questions, and getting cited by ChatGPT and AI Overviews is the visibility play with the least competition for the freed dollars today.

Running this method end to end is the first week of any engagement with our paid media practice — the audit is where the margin math, the measurement reality check, and the scaling plan all start.