Will Google's AI Max Start a BIDDING WAR? | TEMU's aggressive EU takeover.

Episode Transcript

00:00:00
Mike: Welcome to another episode of Growing Ecommerce. We are your hosts. I’m Mike Ryan.

00:00:07
Chris: My name is Chris. Let’s go.

00:00:10
Mike: Let’s go. Chris, I don’t think I told you that your voice is really, really special.

00:00:16
Chris: Why thank you. I like it.

00:00:18
Mike: I really like it. Can you say it again? Tell me something with the most beautiful voice there is.

00:00:24
Chris: Put me on the spot here! I really like it. Fee-fi-fo-fum. Let’s go. That’s what the giant says in Jack and the Beanstalk. By the way, great movie.

00:00:46
Mike: I’m going to have to check that out. Talking about giants, let’s talk about one giant in particular.

00:00:54
Chris: Yes, man. What a segue. That was beautiful. Go ahead.

00:01:00
Mike: We’re going to talk about Google a bit here.

00:01:03
Chris: Oh, favorite. How unusual for this podcast.

00:01:09
Mike: I’m not going to derail us already. I’ll mention it to you afterwards, Chris. A couple of episodes ago, we were talking about some changes to reporting available on Google, like PMax search term data. There is a new match type when you’re looking at your search term data, and there’s another new match type which is AI Max.

00:01:43
Chris: Search term match is existing in reporting, and I’ve been having a look at that lately to try to see what I can make of it.

00:01:51
Mike: What did you find? Well, thank you for asking, Chris. I thought you would never ask.

00:01:57
Chris: What did you find, Mike?

00:01:58
Mike: Well, first let me talk about PMax. PMax is a keywordless technology. You do not target keywords, and it’s going to automatically match you to shopping ads and text ads on Google Search Network as well as the search partner network. Now, there’s no network segmentation, so we can’t know how much of that is on the Google search network compared to the search partner network. That’s a bit frustrating.

00:02:37
Mike: What I typically see so far is not surprising; we see a median of 90% of costs in PMax campaigns is feed-based. An awful lot of that, of course, are shopping ads. So when you look at your search term report, there’s a ton of shopping data in there. It looks an awful lot like your search term report for your standard shopping campaign. Standard shopping is and always has been a keywordless technology as well. This is something that Google has been working on for years: this keywordless matching of products and search terms. Basically, it looks fine, but there’s one thing that I noticed.

00:03:36
Mike: There are very generic search terms matched against. As an example, I was looking at the large skincare and beauty brands in the US, and they show up against stuff like “skincare”—one word and two words. This accounts for a lot of their impressions. It doesn’t account for a lot of their conversions, which isn’t surprising, but it creates clicks. The intent is seemingly not there. We’re supposed to trust Google to be familiar with the user’s intent, but when I tried searching “skincare tools,” I don’t get served any ads at all. They’re even showing up for the word “skincare” or just “skin.” I typed the word “skin” and I get a Wikipedia article, but someone else is getting served shopping ads for sure. It’s really interesting.

00:04:42
Chris: Honestly, it’s not that surprising. I would like to talk about an old approach we as a company successfully implemented in the standard shopping niche. Before we jump into that, it’s not a big surprise that the search terms are similar to standard shopping because, as we stated in the State of PMax report—which was globally recognized as a hell of a hit piece, by the way…

00:05:12
Mike: Well, that maybe.

00:05:13
Chris: Yes, there is this 90% shopping share in PMax. So I think it is quite normal that those search terms are very similar to standard shopping campaigns. What’s also not very surprising is that Google is always giving these very generic, broad search terms a shot. Why? Because it’s a hell of a cost driver. Google looks at your campaign, you put the ROAS target on it, and they reach a ROAS on average. Google doesn’t care that “skin” or “skincare” drives a lot of clicks as long as the average ROAS is met.

00:05:58
Chris: The big question, though, is: could you force Google to be more efficient and selective with these search terms? This is the segue I want to create now. We had a technology called “search query sculpting” back in the day. Mike, elaborate for our listeners on what search query sculpting is in principle, how we implemented it, and maybe then we can discuss if it’s a good approach for PMax.

00:06:35
Mike: Sure. I remember that, and you were driving this. That’s a hell of a feature, by the way. It was built on ideas from our friend Martin Röttgerding at Bloofusion. Years back, he had this genius idea that you could basically combine negative keyword lists and campaign priorities in shopping to funnel the traffic. Since there’s no positive keyword targeting in Google Shopping, you use negative keywords to exclude.

00:07:11
Mike: In the end, you could create campaigns where, through carefully building these negative keyword lists, generic search queries would land in one campaign, more mid-funnel ones in another, and then extremely specific ones in a third. You could funnel brand or different kinds of builds. We developed technology at the time to automate that. It became obsolete, unfortunately, when smart shopping came around and then later PMax, because there was no search term data in the first place and no negative keywords. But now it is available again.

00:07:56
Chris: That’s why PMax is, from my perspective, a way superior product to smart shopping. The idea was great, and I think it’s still great to tell Google that yes, there are these broad search terms where I want to be present, but only to a certain degree. Especially when you have limited budgets, you want to focus on commercially intended, product-specific searches because you have a way higher conversion probability. This concept was a great idea. What’s your take on it regarding PMax? Since the search term report is available, you can basically steer this through the Google Ads API.

00:08:58
Mike: In principle, the idea remains good. Google has basically asked us not to worry about that and says they’ve got it covered. That’s why they didn’t provide this data for a long time or provide negative keywords. They said their audience signals are so strong they will handle it for us. But here’s the problem: it’s very all-or-nothing. You can still add those as negative keywords in PMax now. You can say you don’t want that generic crap. But there might be some cases where it’s not so black and white.

00:09:45
Mike: You might want to appear for that traffic, but you want a different budget for it or you want to bid it differently. You want a different return on ad spend expectation for that. There’s still the strategic goal, and that’s what something like the old-school query sculpting approach offers. Even if Google’s going to outperform on a broad level, there are still strategic and cost-related reasons why you might want to implement a structure like that.

00:10:12
Chris: 1,000%. To our listeners who are operating with Performance Max as a major channel: the cool thing is we’re exploring query sculpting now for PMax. Maybe in one of the next episodes we can shed some light on if this approach works. In principle, I just love the idea of telling Google which search terms I want to be aggressively shown for and where I’m rather opportunistic, especially when you have a limited budget. With an open budget, you can somewhat trust Google, but when you are tight on budget, this query sculpting idea is one I’m very fond of.

00:11:15
Mike: For sure. I just want to mention the single largest limitation here. We’ve got the data and the negative keyword control, but the other core requirement is prioritization. There’s no campaign prioritization yet in PMax.

00:11:32
Chris: But think about it. It gets complicated.

00:11:36
Mike: It gets complicated, but we are here for the complicated stuff. Let’s have a look at it.

00:11:44
Chris: I think it’s a great report, and let’s talk about it over the next couple of episodes. What can you do with this report?

00:11:51
Mike: Before we move on from search term reporting, I want to talk about AI Max, another new search match type. AI is everywhere.

00:11:59
Chris: It sure is.

00:12:01
Mike: Just a short tangent: that technology, if we look at the history behind the scenes, has been in various pilots and tests for a few years. It was initially called Smart Match, then Search Max, and most recently AI Max. Now they’re spinning up this story that AI Max is a technology that will help you appear on AI overviews when those things get monetized through ads. There’s probably some truth to that, but that’s not the origin of this technology or its main purpose. It applies to all of Google Search. This is an example where you really want to have your eyes open and see through the product marketing and the branding Google slaps on it. It’s very convenient and opportunistic for them, but calling it AI Max isn’t really what it’s about—though it sounds fancy as hell.

00:13:37
Chris: The strategy Google applies here is a very similar playbook to every new channel. The big question for me as a retailer is incrementality. What incrementality does it provide? This is the big question with everything Google implements. Is it just a transformation from one channel to another, or is it incrementality? Does it drive incremental revenue, or is it just moving money from the left to the right pocket?

00:14:23
Mike: I have looked at things anecdotally. The danger is that anecdotes might not be representative, but I’ve looked at just a handful of accounts with this technology activated and there were problems in all of them. For context, Google was looking at migrating search campaigns to PMax, but they abandoned that strategy. The other logical way of solving this was bringing some of the PMax technology into search.

00:15:25
Mike: Basically, there’s your standard match types and keyword targeting, and then you can optionally add on some keywordless layers and automated stuff similar to Dynamic Search Ads. What I see consistently when it’s been activated is that these accounts take off on an impressions basis and sometimes a conversion basis. You can see a pretty strong scaling effect. Google’s figure they like to throw around is a 14% conversion uplift, which is a hell of a promise to make.

00:16:18
Mike: If you do not have mature, built-out search campaigns, this could be more beneficial. If you have a mature search account structure right now, the odds are not likely that this is going to deliver good incremental traffic. It might find something, but it might not be what you want. I saw one case with a large multi-brand retailer in Germany where their impressions went way up and conversions also went up. When you look at where that’s coming from, they suddenly started bidding on competitor terms like crazy. The algorithm went insane and just started bidding hardcore against a competitor. This advertiser had chosen not to do that before for good reasons. They weren’t targeting it, but they didn’t exclude it either. Then they turned on AI Max and it went insane. It was 69% of impressions in the whole campaign on this one competitor, and nearly 90% of all impressions were on competitor terms. That’s not acceptable.

00:18:25
Chris: One last question: was it justified in terms of incremental revenue and ROAS?

00:18:31
Mike: It was bringing in conversions, but the question is: is it smart to do that? Is that on strategy for you? In another case, an advertiser had a big spike in impressions but not in conversions or cost. That’s because the algorithm decided it wanted to go crazy on the Search Partner Network. The problem with the Search Partner Network is that it’s too big. YouTube and eBay are search partners, but there’s a lot of “everything else” where there’s very little intent. It’s not what you expect from a campaign that was already in good shape.

00:19:54
Chris: It’s a classic playbook of Google. The core idea is a good one, but with AI Max, you have to really steer and manage that algorithm. It’s more important than ever.

00:20:10
Mike: If you have a less mature search build, this could find some good stuff for you. If you have a mature one, you might find it’s looking in odd places and you’ll need to put guardrails in place, like excluding competitor accounts or opting out of the Search Partner Network. It’s a very new technology. Maybe in a year from now, I’ll have a revised opinion.

00:20:55
Chris: We’ll keep an eye on that and talk about it more at DMEXCO. The more data flows in, the more information we can share. Let’s talk about conversion lag and conversion latency. It sounds nerdy, but it has a very real impact on metrics relevant for both operational marketing managers and decision makers.

00:21:44
Mike: Conversion lag, also known as delayed conversions, is naturally occurring consumer behavior. Google measures things by default from the interaction date, usually the click time. Someone interacts with your ad and some time passes before they actually complete a purchase. That can take a while depending on your cookie windows. This typically correlates with average order value; more expensive items often have a longer consideration phase. It can also have a seasonal characteristic. In the weeks leading up to Black Friday, we see that lag increase because people are clicking on ads but waiting for sales to begin. In the peak holiday season, people are under pressure, so they click an ad and buy the same day.

00:23:28
Mike: Where it’s important is that there will be a discrepancy between this and your backend data. Your backend data reports purchase data when it happens, but Google attributes it back to the date of the click. There is a solution: a set of metrics called “conversions by conversion time.” This means the data is reported by the purchase date instead of the interaction date. That’s automatically going to line up with your backend data better and helps you get better like-for-like comparisons on recent time periods.

00:24:37
Chris: Very interesting. This leads me to another topic for next time. You were talking about the backend in general. A lot of retailers take decisions based on backend data and try to optimize Google campaigns based on that, but the backend data often doesn’t represent the Google data. We should discuss why this data needs to be harmonized and what ways are available to do that.

00:25:33
Mike: I like that. Let’s do it next time. Let’s talk about Temu real quick. Temu is required to share lots of data with the EU because of regulation. They shared their monthly active users for the last six-month period.

00:26:44
Chris: Let me guess: they’re growing.

00:26:47
Mike: They’re growing, Chris. What would you consider a good growth rate for a normal retailer?

00:26:55
Chris: In today’s economic environment, double-digit growth is something you should celebrate.

00:27:17
Mike: For Temu, different standards apply. Their “worst” market is Denmark at about -6%. The Netherlands was the only other market that had negative growth.

00:27:39
Chris: I suspect Eastern European markets with dominant homegrown marketplaces are going to be tough for Temu. Shout out to Allegro; they’re a hell of a player.

00:28:10
Mike: Overall, Temu in Europe grew 23%, which is definitely faster than the market at large. That means they are concentrating market share. Germany was a bit slower at 18%, which is still crazy, especially since Germany is quite flat right now.

00:28:44
Chris: Germany is the biggest market, but it’s growing the slowest. Why?

00:28:50
Mike: I think Temu has a particularly poor reputation among consumers and in the media in Germany. I could be wrong, but that’s the sense I have.

00:29:11
Chris: Even so, people still buy there. The German consumer is probably the most critical about these Chinese mega-cap companies. The bottom line is they’re growing rapidly and faster than the market, so European retailers have to buckle up.

00:29:54
Mike: On that cheerful note, that was another episode of Growing Ecommerce. I enjoyed it.

00:30:05
Chris: Thank you for listening and for tolerating us. This podcast is brought to you by Smarter Ecommerce, also known as smec. You can learn more at smarter-ecommerce.com. We really appreciate your support through ratings, reviews, or word of mouth. See you next time!