How to Read Your PMax Audit Report

Total spend across all Performance Max campaigns in the last 30 days. This is your denominator — every percentage metric is calculated against this figure.

A composite score built from two penalty components: how much spend is wasted (waste penalty) and how much of your 90-day cumulative spend sits in unstable learning stages (maturity penalty). ACI = 100 minus both penalties. Below 50 = critical. 50–79 = moderate. 80+ = controlled.

The total spend that produced zero conversions in the 30-day window. This is the maximum recoverable amount. Your Recovery Potential (on the synthesis page) applies a conservative capture rate to this figure.

The Black Box Tax expressed as a percentage of total PMax spend. Above 50% means more than half your PMax budget produced zero conversions — that is a structural problem, not a budget problem.

PMax campaigns can take 6–8 weeks to fully exit the learning phase. A 30-day snapshot misses campaigns that are mid-cycle. The 90-day window captures the full distribution of stable vs. unstable spend.

The algorithm is establishing baseline signals. Spend in this stage is expected — it's the necessary onboarding cost for any new campaign or major change.

The algorithm is actively testing new audiences and inventory. High exploration spend raises your ACI maturity penalty. Some exploration is healthy — excessive exploration is a warning sign.

Performance is declining — the algorithm is losing signal quality, usually from creative fatigue, audience saturation, or major account structure changes.

Algorithm learning spend — testing audiences and inventory that didn't convert. Some exploration is genuinely necessary. When exploration exceeds 60% of total waste, the algorithm lacks guardrails and is exploring junk inventory.

Spend on asset groups with POOR or AVERAGE ad strength that produced zero conversions. The algorithm funds these because it has no better option — not because they work.

The percentage difference between Ads and GA4 conversions. This normalises the metric so high-volume and low-volume campaigns are comparable.

Ads and GA4 broadly agree. The bidding algorithm has clean signal. No attribution action needed.

Meaningful discrepancy. Common causes: duplicate conversion actions, mismatched attribution windows between Ads and GA4, or cross-device tracking gaps.

The algorithm is bidding on phantom conversions. This inflates CPCs account-wide, suppresses Quality Scores, and degrades ROAS across all campaigns. Every day this persists compounds the damage.

Conversions where the user clicked your ad before converting. This is the strongest attribution signal — the user took a deliberate action. Higher click-attributed % = more reliable performance data.

Conversions where the user saw your ad (an impression) but didn't click, then converted later through another path. PMax counts these by default. A high view-through % means the algorithm is optimizing on impression exposure, not purchase intent.

Conversions where the ad interaction happened on one device (e.g. mobile) but the purchase happened on another (e.g. desktop). These are modeled by Google using signed-in user data — they're legitimate but less precise than single-device click conversions.

PMax serves across Display, YouTube, Gmail, and Search simultaneously. Display and YouTube campaigns generate far more impressions than clicks, which inflates view-through conversion counts. If you're evaluating PMax efficiency based on all_conversions (which includes view-through), your ROAS and CPA numbers may be significantly more optimistic than reality.

Display inventory heavily includes low-quality mobile app placements — games, idle apps, click-farm sites. If you see high Display share here, expect mobile apps to dominate your Junk Placements table on page 3.

Shopping inventory is purchase-intent. A Shopping-heavy distribution is a positive signal for ecommerce accounts.

Video includes YouTube channel placements. Cross-reference with the Junk Placements table — toy unboxing channels, gaming streams, and similar content consistently attract PMax spend without converting.

Severely constrained by missing or weak assets. Any spend on POOR groups is high-risk — the algorithm defaults to low-quality placements because it has almost no creative variety to test.

One asset group is absorbing a disproportionate share of campaign spend. When the algorithm concentrates this heavily, it stops testing alternatives — diminishing returns accelerate.

Detected when an asset group's performance has declined despite stable spend — a signal that your audience has seen the same assets too many times.

Marginal ROAS is at or above average ROAS. Each incremental dollar returns at least as much as the historical average. Green light for budget increases.

Marginal ROAS has fallen below average ROAS. You have hit the efficiency ceiling.

Marginal ROAS has fallen below 1.5x — the breakeven threshold. Each new dollar spent is returning less than it costs.

Placements are categorised as Mobile Apps, YouTube Channels, YouTube Videos, or Websites. Mobile Apps typically dominate — particularly mobile games and idle apps that generate accidental clicks with zero purchase intent.

The specific apps, channels, and websites draining your impressions, ranked by volume. All entries had zero conversions in the reporting period.

Mobile app placements on the Display network are the primary source of wasted impressions. High volume here with zero conversions confirms junk placement bleed.

YouTube channel placements consume budget for brand exposure but rarely drive direct conversions in PMax. High YouTube share without conversion signal means budget is funding awareness, not performance.

The number of distinct placements within each network-type combination. A low count with high impressions means budget is concentrated on a few specific bad placements — easy to fix with targeted exclusions.

The hour (0–23) where the algorithm applies the highest bid multiplier. A 3.2x multiplier means it's paying 220% more per click at peak than average.

The hour with the worst conversion rate. A large gap between Peak Hour and Worst Hour means the algorithm is spending aggressively during low-intent windows.

Insufficient conversions for the algorithm to build a reliable model. Spend is occurring in a data vacuum.

CPA is elevated but conversion volume is building. The market may improve with more data — or may plateau.

Strong conversion rate, low CPA relative to account average, high location score. This market is performing above expectations.

Google's own recommendation engine suggests reducing budget on this campaign. The algorithm's projected spend exceeds the optimal level — your money would work harder elsewhere.

The campaign is budget-constrained — Google projects it would perform better with more spend. Cross-reference with the Marginal ROAS table on page 2 before increasing.

The daily dollar difference between current budget and Google's recommended budget, derived from the weekly estimated cost change divided by 7. Negative = you're allocated more than needed. Positive = you're budget-constrained.

These are your own customer data segments — past purchasers, email lists, remarketing audiences. When PMax targets them, you're paying to reacquire customers who already know you, instead of finding new ones.

Total spend on this audience segment with zero conversions in the 30-day window. This is recoverable budget that should be redirected to prospecting.

Exploit CPA = the CPA of your top-spending asset group (what the algorithm is most confident in). Explore CPA = blended CPA of all other groups. The ratio measures the cost premium for discovering new audiences.

Exploration CPA is more than 3x the exploitation CPA. Learning spend is being wasted on junk inventory.

The algorithm is paying a reasonable price to discover new audiences. This is the expected steady-state.

Exploration is nearly as efficient as your best asset group. Strong signal that budget scaling will be efficient.

Products without a category or brand attribute are invisible to Google's product taxonomy. The algorithm can't match them to relevant search queries or audiences, so they either get no impressions or get shown to the wrong people.

The percentage of your active products that have generated at least one conversion. Low coverage means a small number of products are carrying all the conversion weight while the rest consume budget without results.

Products that have received meaningful traffic (20+ clicks) but zero conversions. These are the zombie product candidates detailed in the next section. A high count here relative to total products signals a feed structure problem — not just individual product issues.

The 20-click minimum prevents false positives from products that accidentally received minimal spend. A product with 20+ clicks and zero sales is a genuine signal failure — not statistical noise.

Total spend on this product in 30 days with zero sales. This feeds directly into your Black Box Tax.

Gross profit minus ad spend. Negative = you lost money on every sale. The algorithm has no visibility into your margins — it optimises for conversion volume, not profitability.

You are paying the algorithm to erode your margins. PMax will continue scaling these if they convert well — it cannot see your COGS.

Impression distribution by age range (18–24, 25–34, 35–44, 45–54, 55–64, 65+, Unknown). Cross-reference with the CPA table below — high impression share with high CPA means the algorithm is over-indexing on a costly segment.

Impression distribution by gender (Male, Female, Unknown). A large 'Unknown' share means the algorithm is reaching users it can't classify — typically lower-quality inventory.

The cost per acquisition for each demographic segment. Compare segments to identify where your budget is most and least efficient. Segments with high impressions but few conversions have inflated CPA — the algorithm is wasting budget on them.

The 'Unknown' demographic typically has the highest CPA because it includes unclassifiable traffic — often low-quality inventory, bots, or incognito users. If Unknown represents more than 20% of impressions with a CPA 2x+ the average, it's a signal quality problem.

Recovery is not money that appears in your bank account. It's budget currently spent on zero-return activity that, when redirected to high-efficiency asset groups, is projected to generate conversions instead. Your total spend stays the same — your ROAS improves.

Real-world accounts recover between 55–85% of identified waste when structural fixes are applied. The pipeline applies a category-level recoverability model — some waste categories are easier to recover than others. The 70% baseline described in the methodology is the conservative floor, not a flat multiplier.

Every formula, threshold, and data source used in the audit is documented on our methodology page. Each metric links to a specific BigQuery table and SQL query.