Stan Consulting · Marketing Atlas · Case File · Performance Max Failure

The PMax Campaign That Ate Its Own Branded Search.

Stan Tscherenkow Principal · Stan Consulting Filed 2026-05-07 Read time · 13 minutes

case_type: composite
proof_level: composite_pattern
cluster: performance-max-failure
published: 2026-05-07

Read the Case File → Read the Position

01 Section 01 · The setup The setup.

A Shopify Plus DTC apparel brand. Eight million annualized. One-hundred-forty-five thousand monthly Google Ads spend across one Performance Max campaign and six Search campaigns. The marketing director four months into the role. Brand exclusions had never been configured at the account.

That is the composite. The names change. The shape does not.

The brand had been on Shopify Plus for two years. The Performance Max campaign was added eight months before the audit, on a recommendation from the prior agency, framed as the "consolidated full-funnel layer" that would simplify the account. The Search campaigns — six of them, mapped against the brand's product lines and audience cohorts — were left running in parallel. Nobody on the operating side had asked what would happen when the two surfaces collided on the same query.

The marketing director was new. The prior agency had been replaced four months in. The new agency was reading the account the way the prior one had set it up. Performance Max was reporting strong numbers. The Search campaigns were softening. Nobody had connected the two reads to each other.

Four months in, the marketing director caught a single anomaly in a weekly export. PMax conversions were up. Branded-Search conversions were down. The dates lined up to the day. That is the moment this case file begins.

Stage: Shopify Plus · DTC apparel brand
Annualized revenue: $8.0M
Google Ads spend: $145K monthly
Active campaigns: 1 Performance Max · 6 Search
Brand exclusions: None configured
Account age: 26 months
Marketing director tenure: 4 months
Anomaly window: The 4 months PMax had been live

02 Section 02 · The visible problem The visible problem.

Two numbers and a feeling. That is what the marketing director brought to the audit.

Number one. Performance Max conversions had risen forty-three percent over the four months it had been live. The agency report led with this every week.

Number two. Branded-Search conversions had fallen thirty-eight percent over the same window. Branded-Search impression share had dropped from ninety-one percent to sixty-two percent. The agency had attributed the decline to "seasonal softness" and a "competitive auction shift" without naming the competitor.

The feeling was that total revenue was flat. The reported PMax wins were not showing up in the bank account. Quarterly revenue had moved less than three percent. The reported lift in ad performance was not translating to the top line.

The board was scheduled to see the quarterly. The CFO had asked the marketing director to reconcile the agency report with the revenue line. She could not. The two reads were telling different stories about the same business.

The audit was scoped at this point. Seventy-two-hour written verdict. The brief was one sentence: tell us why the agency numbers and the revenue numbers do not agree.

03 Section 03 · The wrong explanation The wrong explanation.

The agency had given three explanations. Each one was almost-right and pointed away from the layer that actually mattered.

Wrong reason 01

"Branded Search is naturally softening; PMax is picking up the slack." This is the explanation the agency led every weekly call with. It is the most dangerous of the three because it sounds like a substitution argument and the substitution argument is exactly the failure mode the data was hiding. PMax was not picking up new conversions. It was claiming conversions that branded Search would have served at lower cost. The "substitution" was a rebranding of the cannibalisation.

Wrong reason 02

"A competitor entered the auction on our brand." A reflex explanation that gets reached for whenever branded-search impression share drops without an obvious cause. The auction insights report had not been opened in fourteen weeks. When it was opened during the audit, no new competitor had entered. The bidder eating impression share on the brand query was the brand's own Performance Max campaign.

Wrong reason 03

"PMax is a black box; we cannot see what queries it is running on." This is the structural alibi for not configuring the things that are actually configurable. PMax does obscure asset-group-level placement. It does not obscure brand exclusions. The brand-exclusion list is a setting at the account level. The agency had not configured it because doing so would have moved branded-query revenue out of the campaign that the agency was reporting against. The "black box" framing was protecting the report shape, not describing a platform limitation.

All three explanations let the agency keep the report shape. The structural defect was upstream of the report. None of the explanations went there.

04 Section 04 · The structural cause The structural cause.

No brand exclusion at the account level. That sentence is the verdict. Everything below names the consequence.

The structural defect is small enough to write in a single line. The Performance Max campaign was eligible to serve on branded queries because nothing told it not to. Brand exclusions are a single account-level configuration. They were not in place. PMax read the data, found that branded queries were the cheapest source of conversions in the account, and indexed there. Smart Bidding was doing exactly what it had been told to do.

Five things were true at the same time. None of them were independent. All of them compounded.

One. No account-level brand exclusion. PMax was eligible to serve on every branded query the brand had ever ranked on. Customer Match signals attached to the campaign included the operator's own existing-customer list, which trained the algorithm toward the highest-converting users in the account — the people already searching the brand by name.

Two. The Search-campaign branded ad group had a target ROAS that was higher than the PMax campaign's effective ROAS on branded queries. The auction did what auctions do. The lower-ROAS bidder, the one with broader audience signal and looser query controls, won the impression. Branded Search lost impression share to its own account.

Three. Reported attribution credited PMax for the conversion because PMax served the click. The Search campaign that would have served the same click had it won the auction received nothing. The agency report read PMax as the cause of the rising number; in fact PMax was the cause of the falling number on the next line.

Four. Smart Bidding inside PMax learned that branded queries were a high-conversion segment and weighted toward them. The longer this ran, the more PMax indexed against branded traffic and the less it indexed against new-customer prospecting, which was the only thing PMax was supposed to be doing in the first place.

Five. Total revenue did not move. Branded-search converters convert. They were going to buy. The question is not whether they bought; it is which campaign got credit and at what bid level. The bid level on PMax for that traffic was higher than the bid level on the dedicated branded-Search campaign because PMax had no way of knowing the query was already a brand-loyal one. The account paid a premium for traffic that was already converting.

Five things, one shape. The campaign was performing the job nobody had asked it to perform because nobody had told it which job to refuse.

05 Section 05 · The decomposition The decomposition.

The decomposition reads in three layers. The platform-level rule that was missing. The campaign-level interaction that broke as a result. The reporting-level distortion that hid the break from the operator. This is the order they have to be read in. Skip a layer, miss the chain.

L1 Platform configuration Account-level defect

The brand-exclusion list was empty. This is the single configuration that would have prevented the cannibalisation. Brand exclusions exist at the account level and apply across PMax campaigns. The list takes about three minutes to configure. It had never been touched. Every other defect in this case file is downstream of that one missing setting.

Customer Match signals were attached to the PMax asset group as audience signals. The list contained existing customers. Training PMax against existing customers without excluding their branded queries is a near-guaranteed cannibalisation pattern.

Brand-exclusion list · empty · no campaigns excluded brand queries
Customer Match audience signals · existing-customer list attached to PMax
Conversion-goal hygiene · primary purchase event in place, no double-counting
No URL-expansion controls · PMax was free to serve on the entire brand site map

L2 Campaign interaction Auction-level defect

With the Layer-1 defect named, the auction behaviour was deterministic. PMax bid into branded auctions on signal it had been trained on. The Search campaigns running on the same queries lost impression share because their target ROAS was set tighter for the branded query type. The lower-target bidder won. Search lost. PMax claimed the credit. The campaigns were not "competing" in any operator sense. They were two surfaces of the same account routed to the same auction by a missing exclusion.

Asset-group structure inside PMax compounded the issue. The campaign had one asset group covering every product line. Asset groups are the unit at which PMax learns. A single asset group across the entire catalog pushed the algorithm toward whatever query type was converting fastest, which was branded.

Branded-Search impression share · dropped 91% → 62% over four months
PMax served on branded queries · estimated at 38% of total PMax served traffic
Asset groups · one group across all product lines, no thematic separation
Search-campaign target ROAS · tighter than effective PMax bid on the same queries

L3 Reporting distortion Read-layer defect

The agency report grouped Performance Max wins as new revenue. The Search-campaign decline was attributed to "seasonal softness." At no point did the report line up the two trends side by side. A single chart with PMax conversions and branded-Search conversions on the same time axis would have shown the substitution within the first month. That chart did not exist in the deliverable shape.

The reconciliation between agency-reported revenue and Shopify revenue had also not been done at the campaign level. Agency reporting summed PMax + Search reported revenue and called it total ad-driven revenue. Shopify total revenue had moved less than three percent. The gap was the cannibalisation, double-counted across two campaigns.

Weekly report · PMax conversions and branded Search not charted together
Auction insights report · not opened in 14 weeks
Search-terms equivalent for PMax · not exported, not reviewed
Agency-reported revenue vs Shopify revenue · never reconciled at campaign level

06 Section 06 · The fix or better move The fix, in install order.

The audit's written verdict named the install order. Order matters. Configuring brand exclusions before pausing campaigns is the difference between fixing the source and rotating the symptom across two new campaigns.

The audit drove into the Conversion Second Opinion engagement format and from there into a thirty-day install. The Search-Plus-PMax co-existence shape below is what was installed.

Day one · Configure brand exclusions at the account level

Brand-exclusion list populated with the brand's name, common misspellings, and trademarked product names. Applied at the account level so every PMax campaign in the account inherits the exclusion. This is the single change that ends the cannibalisation. It takes minutes. It should have been the first thing the prior agency did when PMax was launched.
Week one · Restructure the asset groups

One asset group split into four. One per product line, plus one for prospecting against new-customer audience signals only. The signal stack handed to each asset group is now coherent. PMax has a job per group, not a generalized "find anyone who converts" instruction.
Week one · Re-aim the audience signals

Existing-customer Customer Match list removed from PMax. Existing customers are not the audience PMax should be hunting. Replaced with prospecting signals derived from high-AOV new-customer cohorts and Similar Audiences against the existing-customer list. The point of the change is to stop training PMax to find the people who already bought.
Weeks two and three · Rebuild branded Search as the owner of branded queries

Branded-Search campaign rebuilt with exact match, broad-match modifier where appropriate, and a tightened target ROAS reflecting the actual conversion economics of the brand-loyal segment. Branded Search now owns the brand query set without competing internally for it.
Week four · Reporting reconciliation

Weekly chart added: PMax conversions, branded-Search conversions, total Shopify revenue, on the same time axis. Auction insights report opened weekly. PMax search-terms equivalent exported and reviewed weekly. The reporting layer is now built so the next cannibalisation pattern would be visible inside the first reporting cycle.
Month two onward · The board read

The board read shifts from "PMax conversions are up" to "PMax is acquiring new-customer revenue at a measurable cost; branded Search is defending the brand baseline." The two campaigns each have a job. The board now reads the account against those jobs, not against the agency's preferred framing.

07 Section 07 · The lesson The lesson.

Performance Max is a complementary surface, not a Search replacement. The agency that pitched it as the "consolidated full-funnel layer" was not technically wrong; it was incentive-misaligned. PMax that runs without brand exclusions claims the cheapest conversions in the account, and the cheapest conversions are almost always branded queries that would have converted at a lower bid through the Search campaign. The reported PMax win is the operator's branded-baseline being repackaged at a higher CPC.

The four-month delay between launch and detection is the part that compounds. Branded queries do not stop converting; they get more expensive to serve. The operator sees rising PMax conversions and softening Search conversions and reads them as separate trends. They are the same trend. The bigger the brand baseline, the more there is to cannibalise, and the longer the substitution can run before the revenue line catches up to the report.

The lesson is that PMax requires a configured signal stack at the account level before it is allowed to read traffic. Brand exclusions are the first item in the signal stack. Asset-group structure is the second. Audience-signal hygiene is the third. The default Google Ads setup does not include any of them. The default is the failure mode.

Five Cents · Stan's note

What I keep seeing in this pattern is that the agency is not lying. The agency is showing the numbers it has been compensated to show. PMax conversions went up. That is true. The fact that those conversions came out of the branded-Search column is also true, and is also missing from the deliverable. Both things can be true at once because nobody on the operating side asked for the chart that would put them on the same axis.

This is the part that bothers me about Performance Max as a product. Google built a campaign type that is excellent at finding the cheapest converting traffic in your account. For most operators, the cheapest converting traffic is your existing customers searching for your brand by name. The platform did not invent this incentive. It just made it easier to act on by default. The brand-exclusion setting exists because Google knows. The setting is not on by default because turning it on by default would make the launch numbers look worse.

What the case file is for: if you have PMax running, open the brand-exclusion list right now. If it is empty, you have this case file. The next move is the read, and the read is what the Conversion Second Opinion delivers in seventy-two hours.

Stan Tscherenkow · Marketing Atlas · 2026-05-07

09 Section 09 · Related Atlas entries Related Atlas entries.

Each link below points at a related Atlas page that handles a piece of the case file in more depth. Reference pages give the definition. Position pages give the firm's defended doctrine. The hub gives the map.

If this is the pattern in your account

Open the brand-exclusion list. Then call.

If the case file maps to your account — PMax running, branded-Search softening, total revenue not moving the way the agency report says it should — the engagement that runs this diagnostic is the Conversion Second Opinion. A written verdict against the PMax signal-stack methodology, scoped at $999, delivered in seventy-two hours. If the verdict says install, the Sprint engagement runs the install. If the verdict says hold, you keep the read.

Get the Conversion Second Opinion · $999 Or read the methodology →

The PMax Campaign That Ate Its Own Branded Search.

Day one · Configure brand exclusions at the account level

Week one · Restructure the asset groups

Week one · Re-aim the audience signals

Weeks two and three · Rebuild branded Search as the owner of branded queries

Week four · Reporting reconciliation

Month two onward · The board read

Five Cents

Performance Max

Brand Exclusions

Asset Groups

Performance Max Is Not a Replacement for Search

The Shopify PMax Signal Stack

Marketing Atlas

Open the brand-exclusion list. Then call.