The PMax Spend Leak: Why 25-40% of Performance Max Budget Goes to Queries You'd Never Bid On Manually

Performance Max is the only Google campaign type that bids on queries you never approved. The black box delivers conversions, the rep tells you to trust the algorithm, and somewhere in the middle of all that ROAS storytelling, a quarter of your budget walks out the door. The honest math is uncomfortable: most accounts leak 25 to 40 percent of PMax spend to intent the manager would have rejected in five seconds - if they had ever seen it.

Ranges from Optmyzr State of PPC 2024, Mike Rhodes / Agency Savvy PMax studies, Adalysis benchmarks. Account variance is high - DTC e-com tends higher, B2B lead gen lower.

1. Where the leak actually hides

PMax does not bid on keywords. It bids on signals: search themes you provide, plus a much larger pool the algorithm infers from your landing pages, feed, and asset group context. The leak is not malicious. It is the algorithm doing exactly what it was told - explore - without anyone constraining the exploration.

Pull the search terms report on any PMax campaign older than 60 days and the distribution looks roughly like this:

2. Be critical: not all "off-target" spend is waste

3. Why blacklist-only workflows lose the race

The default agency reflex is to scan the search terms report weekly and add the worst offenders to a negative list. It works on Search. It does not work on PMax. The algorithm generates new query patterns faster than a manual reviewer can disqualify them, and the off-target percentage keeps creeping up even on accounts under active management.

The fix is to flip the polarity. Instead of chasing every bad query after the fact, you whitelist the query patterns you want and auto-exclude everything else. That is the architecture behind Timo Adriaansen's open-source PMax Auto Excluder and what the ActCenter PMax engine industrializes for agency-scale management.

4. Whitelist-first vs. blacklist-only: the math

Side-by-side, the two approaches do not converge. Blacklist-only plateaus around 12-14% waste reduction because new query patterns keep emerging. Whitelist-first compounds because the universe of allowed intent is bounded by definition.

5. What 30% recovered PMax spend actually buys you

On a modest $5,000/month PMax campaign, recovering 30% of leak is $1,500/month back into the working budget. That money does not vanish - it goes back into queries the manager already validated, which means it compounds at the campaign's healthy CPA, not its blended-with-waste CPA. Over 12 weeks the cumulative impact looks like this:

Scale that across an agency portfolio of 20 PMax accounts and the math stops being a productivity story and becomes a margin story.

6. What we built into the ActCenter PMax engine

These are the concrete things the ActCenter PMax exclusion loop takes off the account manager every week, and the second-order advantage each one unlocks.

• 1
  Whitelist-first exclusion loop. Define the query patterns you want to show on. Everything outside the whitelist flows into a draft blacklist for one-click approval.
• 2
  Daily search themes export. Yesterday's queries land in a single sheet by 8 a.m., clustered by spend impact and converted/non-converted flag.
• 3
  Brand cannibalization detector. PMax queries colliding with your own Search brand campaign get flagged separately so you stop double-paying for the same click.
• 4
  Three-tier triage. Each query gets a recommendation: keep, cap (asset group rebalance), or exclude. No more binary keep-or-kill decisions.
• 5
  Auto-applied negatives at campaign level. Approved exclusions flow into the campaign without the manager touching the UI - with audit trail.
• 6
  Asset group health scoring. Each asset group is graded on conversion rate, off-target spend share, and feed match quality so under-performers get rebuilt, not punished with more exclusions.
• 7
  Feed × campaign waste reconciliation. For Shopping-eligible PMax, feed exclusions and campaign exclusions are reconciled in one view - no more contradicting rules.
• 8
  Quarterly recovery scorecard. A simple client-facing report shows dollars recovered, queries excluded, and what the recovered spend bought (incremental conversions at healthy CPA).

7. The honest caveats

Same rule as the first post in this series: no automation is a free lunch. Three things to hold in mind before you flip the switch.

Over-pruning kills exploration. The whole point of PMax is to discover demand you would not have planned for. If your whitelist is too narrow, you turn a Discovery engine into an expensive Search campaign with worse targeting. The first 30 days should be looser than the steady state - let the algorithm find the gray-zone queries before you decide which to cap.

Exclusions do not fix bad creative or a thin feed. If your asset group is showing for off-target queries, sometimes the asset group itself is the problem (vague headlines, generic images). Exclude the queries, but also audit what the algorithm thought you wanted.

Negative limits are real. PMax campaigns cap at 100 account-level negative keywords (and a few thousand at MCC level on request). A whitelist-first loop respects those limits by clustering exclusions at the theme level - but if your loop generates thousands of literal-string negatives, you will hit the wall fast.