‍

Let’s talk asset groups.

‍

What Are Performance Max Campaigns?

Yeah, we’re just gonna hop right in there. If you need to learn more about Performance Max (PMax)  campaigns, head here: Google Support on PMax

‍

A Brief Definition of Asset Groups

Asset groups are a new segmentation option within Smart campaigns, which allow unique assets to be set with unique audience signals and product listings in order to better feed the machine learning incorporated by Google in serving the correct ads to the correct person. 

‍

The Fatal Flaw in Asset Groups

My take on asset groups in Performance Max campaigns is they are brilliant creations that marry the coming onslaught of machine learning, with user created assets targeted to unique signals. 

‍

In an ideal world, they would allow us to marry specific products, with specific assets, to specific audiences so we can better entice a certain audience with certain creative. This has been afterall, the power of social advertising such as Facebook, Instagram, and TikTok. You can identify an audience of “cyclists” and therefore target cyclist focused products and images/videos to them. 

‍

I said “ideal world”, because unfortunately we live in the real world and you can’t quite do all of this (product > assets > audiences). To be clear, you are able to target unique product groups to unique asset groups with listing groups, however my issue comes with our inability to target specific assets to specific audiences for the same products. 

‍

Admittedly, Google’s recommendation has been to avoid the same products in multiple asset groups… likely to avoid the concern I have brought up. But this is an oversimplification of the fact that different audiences purchase the same products, but can be enticed with different creative.

Herein lies the fatal flaw of Google’s current asset groups, and in writing this, I am hoping to do more than complain, but to actually enlist their help in solving this problem. 

‍

The flaw rests in the generalized nature of the audience signals. 

‍

According to Google, (check out this great Greg Finn article with more detail), signals are just that… they guide the machine learning process, which ultimately determines what and who to target. The problem is, your ad asset groups can be targeted to any audience the machine learning deems viable to target based on performance, and not the actual audience signals you have created your assets for. It forces a generalized approach to ad assets… which is counter to a historically agreed upon value in targeted creative. 

‍

Allow me to illustrate. 

‍

Let’s say you are Lumos (I have no connection, just think it’s a cool product) who invented, and now sells, bicycle helmets with lights on their helmets.

You investigate Performance Max campaigns, and determine to sell them to your three groupings of lit-helmet purchasers:

• Families with small children for visibility.
• Commuters who do not want to perish on their way into work in the Big Apple.
• Road cyclists who want to be clearly visible to all those doing their best to run us off the road. At least then, we’ll know they were trying to kill us because they could, in fact, see us (I’m not bitter). 

You decide to build a PMax campaign segmented by these three asset groups, and you task your creative team with creating unique video and images around each of these groups. 

‍

You also have a fourth asset group for more “general” cyclists and you sit back, proud of your well-designed, well-guided, asset groups. 

‍

You push things live, and suddenly begin to realize that Google isn’t really, technically, supporting your asset creation. You have road biker imagery showing to families, and small children in Lumos helmet videos showing to commuters. Why does a childless person in Chicago care about keeping the kids they don’t have, safe? 

‍

See the conundrum? The way I see it, the current asset groups lend themselves to blended audience signals, rather than utilizing assets as a crucial part of that process, and therefore we are tempted to actually reduce the little remaining power of marketing we advertisers have - assets - by pulling back from ultra-specific audience/asset match and moving back towards more of a generalized branding creative and videos. 

‍

This is problematic, as I think it is killing the potential of using targeted assets to their full potential. 

‍

A Suggested Solution

How can this be solved? 

A few ideas, 

‍

1. Allow audiences to be specifically targeted by more advanced advertisers. Google could keep the audiences as more generalized signals by default in a new PMax campaign, but allow a checkbox that allows for more granular control of audiences to an asset grouping if a more advanced advertiser has access to excellent, specific creative, that aligns well with an audience they are able to target. 
2. Incorporate the asset details into the machine learning to train the algorithm for that specific asset group. It would make sense to me for Google to use the ad asset uploads as part of their algorithm (they may already do this) to better target specifically for the asset created, and not just simply “reported performance” which any mature marketer knows has ample opportunity for issues. 
3. Allow unique bidding targets at the asset group level for advanced advertisers to better send signals for audience value to the algorithms per audience signals rather than keeping everything at the campaign level. 

‍

While number two above is important, I think number one above is the core change that could be made to better the system for more advanced advertisers, while leaving the default alone for the majority of advertisers who may not care. 

‍

What do you think about all of this? How have you been seeing Performance Max act? Please hit me up on Twitter or LinkedIn (I accept all non-sales connection requests) and share your thoughts on all of this.

‍

EDIT: A Response by Google (ongoing)

Thanks to the amazing Ginny Marvin (Google Ads Liaison) for a quick reply on Twitter/LinkedIn. It's not the most surprising thing I've ever read, but it was helpful to get some initial feedback. Will keep things updated as the conversation continues: 

Ginny Marvin: Hi Kirk, Appreciate the thoughts here and wanted to follow up on a few points about how asset groups and audience signals work:While audience signals aren’t hard targets, the aim is still to show the right combination of assets to the right audience to drive conversions/conversion value. The benefit of Performance Max is that it may also find conversions outside of the audiences you've already identified in audience signals, but this should not preclude advertisers from creating specific & excellent creatives tailored for your target audiences.As in the bike helmet example, if you want to customize your messaging/assets by audience, you can use multiple asset groups. Creatives aren’t assembled arbitrarily to “have road biker imagery showing to families.” But if at auction time, a combination of road biking assets is predicted to convert a “family” user, then it could serve. Conversely, if this doesn't convert then machine learning would learn from this signal that we shouldn’t show road biking ads to a family user. Performance Max can also find new audience segments you may not have expected to convert. When available, audience insights on the Insights page will show you which audiences are “signals” that you proactively added and which are “optimized” and found by automation. Additionally, Conversion Value Rules allow you to indicate higher value audiences. While these are at the campaign level, the principle is the same: advertisers express audience value and Performance Max optimizes the bid and assembles the asset combination best predicted to convert for each audience member in real time.Thanks again and hope this is helpful.

‍

Kirk Williams: thanks Ginny, I appreciate the detailed reply. It doesn't surprise me knowing the way PMax is geared towards controlling all elements and learning over time, but it was helpful to see the official response 😄 One thought in response to your response, is that my "solution number one" can help guide the learning more quickly, essential for smaller advertisers or more targeted audiences who may not have the cash or time to allow the machine to learn by trial and error what could simply be told to it immediately with a specific audience/asset setup.That being said, I am curious if you can give insight into how creative is being utilized in the auction? Does the algorithm take the video/images/headlines/landing pages/etc into account as signals? Or is it just identifying the conversion behavior of unique aspects of the creative without taking the actual creative into account (does that question make sense? I.e., is the algo capable of reading/viewing the creative to target unique audiences)? Thanks!