Google Performance Max Optimization in 2025: Strategies and Conversion Settings
The central theme is the need to use the right conversion goals (especially considering customer value, LTV) to increase PMax efficiency.
Setting the right campaign goals and tracking issues on iOS
Hello! Today we have an interesting analysis: we are reviewing the minutes of internal meetings of the marketing team. We will look at Google Ads strategies and, of course, Performance Max in particular. This is not theory, but real discussions.
The main idea is to move from simple metrics to real value for the business. Stop chasing cheap registrations that don't amount to anything, and learn how to set up Pmax, which is a complex tool, to work towards these deeper goals, taking external factors into account. There's Apple with their restrictions.
It's clear that simple registrations are easy. But that's just the beginning of the funnel. The key conclusion they came to was that it is essential to send signals to Google Ads about actions that correlate with money. That is, either purchases, or a successful trial, or even some metrics that predict LTV — lifetime value. It took a long time to set up TROAS — value-based bidding. It took almost a year to find it. But it seems like the results were good in the end. And now this same approach is being implemented in DemandGen.
And for Pmax, it turns out to be even more important. Search, YouTube, CMS, Discover, Gmail — everything. And if you don't give it a clear goal, specifically a value-based one, it can easily drain the budget on clicks that convert into cheap actions, but not into money. That's why conveying real value through offline conversions, Enhanced Conversions, and Server-Side Tracking is a must-have.
And, by the way, this also helps with the cannibalization problem, which was also debated. When Pmax sees real value, it competes differently in the auction.
Now about the problem with macOS Sonoma. Gclids started to get erased in Safari. The team checked, ran tests, and confirmed the theory. And this became another impetus to search for more stable solutions that are not tied solely to cookies or parameters in the URL. The main focus was on implementing Enhanced Conversions. The logic is simple. Even if technical identifiers are lost due to the browser or privacy settings, EC allows Google to link the conversion to a click on the user's hashed data. That is, the email or phone number that the user left on the website. In other words, it's a way to fill in the gaps where conventional methods fail.
In discussions, this was clearly seen as a strategic direction. Because, well, the trend toward privacy isn't going anywhere, it's only going to get stronger. Plus, they noted that difficulties with remarketing on iOS remain and were waiting for updates from Google, especially for App Campaigns.
There was a lot of talk about new campaign goals. Namely, attracting new customers, NCA. What is it and why? This is an example of how Google is trying to provide tools for working with value rather than volume. NCA allows PMax to search specifically for new users. Two modes were discussed. The first is Value Mode. The system tries to bring in more new users, assigning them greater value, but it doesn't ignore old users either. The second is New Only Mode. Here, the targeting is purely on those whom Google considers new, based on customer lists and its own signals. Who might find this useful? We see the value in this for businesses where the main goal is to grow their customer base. Retention is handled through other channels, such as email marketing.
The opposite task is retention. We talked about Retention Goals, but this is still in beta, although you can request it. Assign greater value to conversions from existing customers. This is to encourage repeat purchases or other actions through paid channels. It can be useful for reactivating dormant customers or for businesses with frequent purchases.
Optimal structure, must-have creatives, placement control
Now here's a very practical point that has been controversial. Campaign structure. When should you run multiple separate PMax campaigns, and when is it enough to have different groups of objects within one? The conclusion that emerged was roughly as follows. Separate campaigns are necessary if you have different budgets or different goals, for example, one for leads and another for sales. Or different targets. CPA or growth for different countries or products. Or if you are testing completely different creative strategies, such as a promotion and a regular campaign. And in all other cases? In all other cases, when you have different audiences, different creative themes, but a common budget and a common goal, it is better to use different groups of objects within one campaign. This gives the algorithm more flexibility and more data for budget allocation. And if you split it into campaigns, that flexibility is lost.
Creatives. You need everything. Text, images, and definitely video. This is not just critical. It is the foundation of PMax. Imagine Google's entire advertising inventory as a jar. Search is the big rocks. KMS and DemGen are the sand between them. And PMax is like water that fills all the gaps. YouTube, Discover, Gmail, maps, everything. But for this water to fill everything, it needs all the components. Text, images in different formats, and especially video. You simply don't show up in a huge segment of inventory on YouTube or in Discover feeds. You limit yourself to the capabilities of PMax.
It is essential to use voice in videos. Yes. Not just background music, but clear voiceovers. The brand name, a call to action. The argument is simple. Many people listen to YouTube in the background without even looking at the screen. At work or somewhere else. The voice conveys the message. Plus, if there is voiceover, the system generates subtitles. And this is important for those who watch without sound. And now, with AI tools, it has become much easier and cheaper to do voiceovers.
And all formats are needed — horizontal, vertical. Definitely. For YouTube, for shorts, for feeds — you need everything. While DemGen sometimes recommends separating videos and images into groups, PMax's best practice is to mix everything into one group — text, images, videos. This gives the algorithm maximum flexibility.
There is still no direct control over placements, where to show. And it is unlikely that there will be. This contradicts the very idea of PMax. But we considered Channel Reporting to be a positive shift. What does it give? It shows how much of the budget and conversions went to search, how much to YouTube, KMS, Discovery, and so on. This already gives at least some understanding of what is going on inside.
Google says it is working on greater control, but this will most likely be in the form of signals for the system, such as “more focus on YouTube,” rather than direct disabling. And the standard features remain—security settings, branding, exclusion of unwanted topics, and exclusion of placements at the account level.
Fluctuations in effectiveness, what to do if PMax stops working, is there cannibalization of search campaigns
Another question was raised. What should you do if a PMax campaign was working well, but then suddenly broke down and stopped producing results? Sudden drops like this are rare. There is always a reason. First, check the technology. Conversion tracking hasn't fallen apart, everything is being transmitted. Second, look outside. What's happening in the market? Competitors, demand, seasonal decline, Google Trends to the rescue. Third, Channel Reporting. Which channel within PMax has declined? Search, video, KMS. This narrows down the search for the problem. Fourth, creatives. Especially videos. Have they burned out? Maybe the audience is just tired of them. It's time to update them. Fifth is Auction Insights. Auction statistics. What's going on with the competition? And if CPA or ROAS are no longer being achieved, then before you start tweaking them, and doing so gradually, you need to make sure that all the creatives and conversion signals are in order. It's useless to push for goals that the system can't achieve.
Business
conversions
Creatives
set
Right
structure
LTV
forecast
We also discussed Search themes. PMax search themes. It was emphasized that these are signals, not hard keywords. By the way, the limit there has been increased to 50 themes. The system can also show related queries if it considers them relevant to the theme and audience.
The main thing is to use high-quality signals. Customer lists from CRM, audiences by interests, by intentions, custom segments by keywords or URLs. Should they be combined or separated? That was the question. If the message is the same for everyone, you can combine them into one signal. It's easier.
But if you want to show different creatives to different segments, say, one to newcomers and another to regular customers, then it's better to spread them across different groups of objects. Each group has its own signal and its own creatives.
And now for the final battle. PMax vs. search. Does PMax really steal traffic from old search campaigns? Branded ones, for example. Google's official position, which they also checked, is that the system selects the campaign with the higher AdRank for display. That is, the expected effectiveness. If PMax won the brand display over search, then according to Google's calculations, it is more likely to lead to a conversion in this case. But. First. Set up the right deep conversions for PMax. Second. Use brand exclusions, excluding brand queries in PMax, if it has been strategically decided to give the brand only to search. And third. To understand exactly what PMax adds, use an uplift experiment.
Key points
First and foremost, it's the transition from quantity to value. Correct conversions are the foundation.
Second is the completeness and quality of input data for PMax. All creatives, especially videos with voiceovers and subtitles, high-quality audience signals, and themes—without these, there will be no magic.
Third is a strategic approach to structure. Understand when you need separate campaigns and when you need groups of objects. And consciously steer PMax's interaction with search.
Brand exclusions, uplift experiments. And don't be afraid to experiment and be methodical in solving problems. Test new features, such as NCA or retention goals.
And if something goes wrong, don't panic, but take it step by step – tracking, market, creatives, channels, competitors. Analyze.
And here's a final thought to consider: if we can calculate and optimize campaigns for predicted LTV, then maybe the next step is to optimize the LTV prediction model itself? How can we predict customer value more accurately in the early stages to give PMax an even stronger signal? This turns out to be a task at the intersection of marketing and data science.
All the best! If you need help with analytics or PMax configuration, please contact us!
