We killed Performance Max. CPA dropped 24% in the same 30 days.
A multi-location powersports dealer group was pouring budget into Performance Max on the assumption that the algorithm knows best. The algorithm was learning from 3,300 corrupted lead form submissions. Here's what we did about it.
The situation
The client runs a multi-location powersports operation across Southern California — motorcycles, ATVs, side-by-sides, and personal watercraft from every major manufacturer. Their Google Ads account had been running Performance Max as a large share of total spend, on the assumption — the one Google reps push in every quarterly review — that automation would find efficiencies humans couldn't.
What we found under the hood was different. The Smart Bidding models feeding Performance Max were being trained on lead data that was structurally broken. Approximately 3,300 form submissions in the recent history were what we call ghost leads — form fills with no phone number, no email, or filler data that never converted downstream in the dealership CRM. Every one of those bad conversions was telling Google's algorithm: this pattern of user is valuable. So Google kept buying more of that user.
Meanwhile, we had no clean way to see which campaigns, keywords, or creatives were actually producing sold units. The CRM had the truth. The ad platforms didn't. Attribution was a wall of aggregate numbers with no closed-loop signal.
What we did
Shut off Performance Max as an experiment, not a permanent decision.
In mid-May, we paused Performance Max campaigns entirely and moved budget into standard Search with tightly targeted keyword and location structure. This was framed as a 30-day test — kill it, measure it, decide.
Cleaned the conversion data feeding Smart Bidding.
We identified and audited the ~3,300 ghost lead submissions, traced their origin (a mix of form-fill bots and an internal test flow that was firing production conversion pixels), and rebuilt the conversion event definitions so only qualified leads with valid contact info counted.
Built a Google Merchant Center vehicle feed.
Set up a proper vehicle inventory feed for Shopping and Vehicle Ads — the surface Performance Max was supposed to be dominating but wasn't. Now vehicle-level ads run on structured inventory data, not automated guesswork.
Ran deep CRM funnel analysis.
Pulled months of dealership CRM data and mapped ad-platform-reported conversions against actual salesperson touches and closed units. This exposed which campaigns were producing real pipeline versus which were producing noise.
Cleaned up Google Business Profile citations across every location.
Auditing local listings caught mismatched NAP data across dozens of directories — a slow leak on local pack visibility that had been invisible in the ad account.
Competitive teardown of the largest regional player.
Ran a full audit of the largest regional competitor's paid and organic footprint to identify where the client had structural advantages and where they were being outbid on high-intent inventory queries.
The results
- Cost per acquisition dropped 24% across the account in the first 30 days.
- Cost per click dropped 28% on high-intent inventory keywords.
- Conversion volume held flat — same lead count on materially less spend.
- Freed budget was redeployed into structured Search and Shopping campaigns tied back to specific inventory units and, via the cleaned attribution stack, to closed sales in the CRM.
What this means for a business like yours
If you're a dealer, contractor, or any local business running Performance Max because a Google rep said you should, ask them one question: which of my conversions was manufactured by ghost data? They won't have an answer. Performance Max is only as smart as the conversion signal you feed it, and most accounts are feeding it garbage without knowing.
The play isn't PMax is bad. The play is: prove your conversion signal is clean before you let any algorithm optimize on it. Most of the time, the audit itself is the ROI.