Urban Threads: Performance Max Audit Failure in 2026

Sarah, the Head of Digital Marketing at “Urban Threads,” a growing DTC fashion brand, had a sinking feeling as she looked at the Q1 2026 Performance Max report. The brand’s new spring collection, launched with a big budget and lots of noise, was tanking. PMax had promised the world: simpler campaign management, massive reach across all of Google’s properties, and better conversion rates. But three months in, their cost-per-acquisition (CPA) was a full 30% higher than what they usually saw for new launches. This wasn’t the efficiency they were sold. It was painfully obvious they needed a deep-dive Performance Max audit, but where do you even start auditing a system that’s mostly automated? The influence of the AI agent feels like a total black box. Urban Threads had jumped on the PMax bandwagon for its promise to find customers across Search, Display, YouTube, Gmail, and Discover from one single campaign. You just feed it your best creative, your product feed, and your conversion goals, and let the Google AI handle the rest. For the first few weeks, it actually looked like it was working, initial conversion volumes were strong. But as the campaign ran on, the CPA started to creep up. Then, in February and March, the return on ad spend (ROAS) just fell off a cliff. Sarah was pretty sure the AI agent performance was the problem. It was either reading their signals wrong or just burning cash inefficiently, but proving it felt impossible. Their first move in the audit was to rip open the asset groups. Every PMax campaign is built on these bundles of headlines, descriptions, images, and audience signals. Urban Threads had set up four, each targeting a different angle for their spring collection: “Casual Chic,” “Evening Elegance,” “Active Lifestyle,” and “Sustainable Staples.” The strategy looked solid on paper. The data said otherwise. The “Casual Chic” group was eating almost 60% of the budget but was only responsible for 35% of the conversions they were seeing in the Google Ads dashboard. That kind of lopsided spending is a huge red flag. It’s a classic PMax story we see all the time: if you don’t watch it, the AI will always chase the easy conversions for volume, not the profitable ones, especially if your conversion signals are even slightly off. So Sarah’s team, along with their agency partner, dove into the asset group performance details right inside Google Ads. They went straight for the “Combinations” report, which shows you exactly which mix of headlines, text, and creative is actually getting results. The findings were a real eye-opener. The AI was obsessed with certain dynamic images and videos that, while pretty, were attracting clicks from window shoppers, not buyers. These broad-appeal assets were getting clicks but no sales. At the same time, some of their best-performing combinations, the ones with specific product shots and detailed descriptions, were barely getting any screen time. “It’s like the AI got excited about the shiny objects, not the ones that actually sold dresses,” Sarah said in a team meeting. This is the reality of managing AI campaigns: the machine only optimizes for the signals it’s given. Urban Threads was using standard Google Analytics 4 (GA4) conversion tracking, but their initial setup didn’t really distinguish between a micro-conversion like “add to cart” and the real prize, a “purchase”. That lack of clear instruction was likely sending the AI on a wild goose chase. They weren’t alone in this. A recent eMarketer report from 2025 showed that almost 40% of advertisers struggle with this exact problem, where fuzzy conversion signals cause the AI to misallocate huge chunks of budget. The team came up with a plan. First, they went into the “Casual Chic” asset group and immediately paused all those underperforming videos and images, replacing them with new creative that looked more like the stuff that was actually working in the “Combinations” report. They also set up more specific conversion value rules in Google Ads, telling the system that a completed purchase was way more valuable than an add-to-cart. This gives the AI a clear pecking order of what you actually want. It’s a simple change, but so many people forget to do it. If you tell Google a “purchase” is worth the same as an “add to cart,” the AI is going to get you a ton of “add to carts” because they’re easier, and your ROAS will go straight into the floor. Next up: audience signals. PMax uses these signals to get started, but Urban Threads had just set them and forgotten them. Big mistake. The audit showed their customer lists were old and the website visitor segments were full of people who never converted. They immediately uploaded fresh customer lists focusing only on recent purchasers and built new custom segments targeting users who had looked at specific product categories, not just anyone who hit the homepage. They also added some detailed demographic exclusions for areas where shipping was too expensive, a small tweak that saves real money. Then there was the negative keywords issue, something a lot of people don’t realize you can even do in PMax. You can’t just add negatives at the campaign level like you’re used to. Urban Threads had totally forgotten to submit a brand-level negative list to their Google account rep for the new collection, which meant they were showing up for all sorts of irrelevant searches. Once they sent over a complete list of competitor terms and unrelated fashion keywords, they saw a huge drop in junk impressions and clicks within a week. It’s a must-do for brand safety and stopping budget waste, but it’s hidden, so tons of marketers miss it.

The budget allocation was another problem area. PMax is supposed to be smart about spreading money across channels, but the audit showed it was dumping a ton of cash into garbage Display Network placements with terrible conversion rates. We’re talking about mobile game apps that had no business being associated with their brand. Getting control isn’t easy. You have to use the “Placement Exclusions” feature by contacting your Google account manager to block those specific apps and sites. It’s a manual process, but it immediately forced the budget back into channels that actually worked. Don’t expect to find this spelled out clearly in the Google Ads docs, by the way. You usually have to ask for it. Three weeks later, the changes were working. CPA dropped 18%. ROAS was up 25%. They weren’t back to their historical averages yet, but it was a massive turnaround. The “Casual Chic” asset group was finally pulling its weight, with conversions that were actually in line with its budget share. More importantly, this wasn’t a one-and-done fix. They put a new process in place: weekly checks on asset groups and conversions, with a monthly deep-dive on budget and audiences. With AI campaigns like this, you understand that campaign optimization is a continuous loop of analysis and tweaking. Sarah thought about the whole ordeal. “The AI is powerful, no doubt. But it’s not a set-it-and-forget-it solution. It’s more like a highly intelligent, but sometimes easily distracted, employee. You still need to provide clear direction, regular feedback, and occasional course corrections.” For Urban Threads, the lesson was clear: these AI agents have huge potential for scaling things, but they absolutely require a human hand on the wheel for strategic analysis and regular audits to be successful. The point of AI in advertising isn’t to replace marketers. It’s to make them more strategic by handling the grunt work.