If you’ve been monitoring the Google Shopping Ads, Smart Shopping Campaign, ongoing data transparency discussions (okay, so all 2 of you) you might have run across this sound article by Patrick Gilbert on LinkedIn in response to my article, both linked to here:
The Concerning Lack of Data Transparency in Smart Shopping – Part 5: The ZATO Guide to Google Smart Shopping Campaigns by Kirk Williams (hey, that’s me! Insert Homer Simpson “WOOHOO” here).
One friend DMed me and asked, “did you and Patrick plan this?” The answer is, no! Which makes it even more awesome since unplanned disagreements are often the discussions that push us to really think deeply about a topic and our viewpoint on it.
Patrick’s pushback article to mine was filled with good thinking, and I especially appreciated the tone of the article. Disagreement with respect is the best kind, no?
While I appreciated some of Patrick’s arguments and his overall tone, I still disagree with his overall premise… so here is my attempt to respond to the response… in the same respect he afforded me.
I should note that this turned into an exceptionally long article, so I’ve filled it with images of (some of) my office Lego Star Wars collection to keep you going…
Let me begin by pointing out things I agreed with in Patrick’s article, as I think that’s important.
- At a high level, I hear us saying the same thing: “additional data included in SSC can be helpful for good PPCers to make informed decisions in their Google accounts.” We disagree on what data is our “right”, but we’ll get into that later.
- We have both seen visible success in SSC in our accounts (though editor’s note: I seem to have seen SSC struggle more than Patrick).
- We both acknowledge (albeit for me specifically in other contexts or writing or speaking) machine learning has improved and the systems are getting smarter and better able to handle data well.
- We both agree the reason Google engineers are withholding this data is an attempt to help advertisers overall, and not some nefarious purpose. I have experienced the same thing in discussing this with Googlers as Patrick seems to have experienced. There is a concern that PPCers will in essence, “mess up the algorithms” by constantly tweaking things too early, or without enough data, or from boredom (more on this later).
A Purposeful Pondering of Patrick’s Primary Points
Agreements aside, let’s move on to the disagreements, you know… the fun stuff! I have been trying to think of the best way to respond that does Patrick’s article justice and also gets through the jumbled thoughts in my brain so I am going to address what I viewed as the Primary Points of Patrick’s article. These may not necessarily be what were visually obvious, but are what I believe to be the common theme throughout, and more importantly, what are also the primary points I have heard voiced from Googlers and others on SSC (or automation in general), and thus deserve a specific response. This is again where I see the conversation expanding beyond simply SSC, and into automation in general within Google’s channels.
Primary Point #1 from Patrick: “Google doesn’t owe us anything.”
In many ways, Patrick rightfully identified this as a core aspect of my original article: I believe Google owes a certain amount of data to the ones paying for the ads.
Here is what I have found, there are three views of looking at buying advertising and thus the data that comes with it (I’m not a lawyer BTW, so it would be an additional interesting take on things from someone who knows law and PPC):
- NO RIGHTS View: The advertiser has no “”right”” to any data they pay for in advertising.
- ALL RIGHTS View: The advertiser has every “”right”” to all data they pay for in advertising.
- NECESSARY & BENEFICIAL View: The advertiser has the “”right”” to essential, necessary data needed to make informed business decisions well for the benefit of the account.
I disagree with the NO RIGHTS view since I believe payment for advertising should include *some* level of data. Even billboard companies when selling billboards (I’ve purchased billboard advertising before) will include data at some level. They will trumpet the number of eyeballs viewing their billboard every day (btw, companies like BlipBillboard have gone so far as to monetize that data by “peak times” during rush hour and lower periods of time like evenings and weekends which impacts their auction based pricing. Data!). Data is impressions and clicks. Some level of data is, I strongly believe a requirement in an advertising ecosystem.
I’m ambivalent (torn) on the ALL RIGHTS View simply because I am not a fan of the race to collect data for the sake of data, but I also don’t want to set a precedent for platforms to determine what data is necessary. I certainly want my clients to have the right to see every piece of data, but I believe this is something in the legal system that also falls into the “what is reasonable” camp. If Google was responsible for reporting on every ounce of data they collect for every advertiser and making that publicly available, I’m not even sure we would be able to process that mammoth ecosystem (certainly not smaller businesses), and it would inevitably lead to analysis paralysis. That is, we couldn’t make decisions because we couldn’t even see through the mountains of data we would be receiving on every client. In that regard, one who knows my other writing may be surprised to hear me admit this, but I’m not convinced the ALL RIGHTS view is the best one when it comes to advertising data.
What do I think is best?
I fall into the third camp here (and my guess is as do most advertisers when they actually think through the ramifications). That is, I just want to make sure I have access to the data that I am able to make good decisions by for the good of my clients and business.
But this is the rub, and why we’re having this conversation:
We’re all going to disagree on what data is necessary to do our jobs well.
Logically Patrick and I are both in number three. He has admitted to the need for conversion data to be reported which means he believes some level of data is a right, so really we’re arguing about what level of data we need to see to do our job well (what is necessary and beneficial). Since that’s the case, that’s another conversation entirely since some level of data is a right, and it’s possible others will have a need for data Patrick doesn’t particularly need, but may someday if a different type of client comes along. In that case, I’d plead with Patrick (and others making this argument for Google), that we may not all agree on what is necessary and beneficial, but let’s not allow Google to determine what that is without some level of advertiser feedback.
At the end of the day, I believe seeing Channel insights, audience insights, placement/video insights, and search query insights is essential to doing my job with SSC… or at least doing it well. Let me give specific cases of this below:
Channels, Channels, Missing Channels
Patrick compared the lack of data in Smart Shopping Campaigns to TV or the Wall Street Journal to demonstrate there are data points that traditional advertisers have lived without for a long time, but I believe this comparison isn’t compatible with what is actually happening with SSC.
Here’s the problem with his analogy, because SSC doesn’t include Channel data insights his analogy isn’t an accurate description of what’s actually taking place. It’s not apples to apples to compare SSC to a TV network, or newspaper.
Rather, it would be more like if:
A TV network also owned 1000 billboards across the country, as well as 25 newspapers and sold you ad space to “their entire ad ecosystem” including all those platforms. You paid $40,000 and the Media network determined where they showed ads and then didn’t tell you which publications your ad was appearing on (what if your full-page ad was in a racist newspaper in the deep south? You don’t know), or any insight into which billboards your ad was appearing on (are you paying more for the better billboard spots? You don’t know) and on and on.
SSC can’t be compared to a single traditional channel, because it’s multi-channel (you show on Search, Display, Gmail, Video)… and you can’t even tell how much of your budget is going to Youtube as opposed to Search. In that regard, there is actually less data/control by the SSC advertiser than a traditional advertiser, because at least the traditional advertiser can allocate budgets between billboards and TV spots!
Let me leave you with two anecdotes to explain this, and then this will feed nicely into the next point or response here.
While Patrick was fixated (even titling his article) on my concern for Search Terms data, I detailed in my article (and believe strongly) that Search Term data is only one of the data transparency concerns here. I used it as an example because it’s an easy one. Here are two more ways (placements/audiences) in which Smart Shopping Campaigns have prevented two different Brands from running them because of lack of data:
Placement: Video transparency: we have clients (and have heard this from multiple others), who require more transparency into where Smart Shopping is appearing for brand safety concerns. This isn’t just data for the sake of complaining, this is a lack of data that is preventing SSC from even being used. A fatal flaw!
Audiences: Returning or New Visits? Advertising for retailers or smaller brands is a different animal from advertising on household brands who have unique ROAS targets, budgets, ad content, and strategies for visits based on where they are at in the funnel. In SSC, with everything lumped together in one campaign and no transparency with what is happening between remarketing and prospecting visits, we have had clients hesitant to use it for that purpose. We know we can own certain prospecting terms (where tracked ROAS isn’t the goal) with Standard Shopping because of visibility into that and our ability to control that situation and that is a better fit with their overall business strategy. While Google is working on a Beta solution for something like this, it proves the original point that no data on audiences (added to an inability to separate out ad text, budgets, and ROAS targets between new and returning visits) is a fatal flaw in SSC.
Data is important, we may disagree on exactly what data is essential for management but hopefully I have demonstrated above (especially with my two brand examples), that we have certainly run into cases where the lack of specific data prevented us from using SSC… which suggests that data is essential, specifically the data currently missing from Smart Shopping Campaigns in Google.
Primary Point #2 from Patrick: “If we had the data, we would simply ruin Google’s smart process.”
The second point that I want to discuss is related to the first point in that is an argument for not needing insight into the data. Why? Because if we could see the data, we would be tempted to use the data and that would mess with what the machines are trying to do (so goes the argument).
I have a ton of sympathy with this argument, in fact, in favor of this argument is my friend Martin Roettgerding’s excellent session-turned-blogpost on this very issue within ad testing (TL;DR: humans can make decisions too quickly on ad tests where the results would have swung back the other way if they had waited longer). The machines are even making decisions on data we don’t have access to (for instance, data we don’t even have access to things like competitor bidding and vertical historical behavior), so the argument from Patrick (and Google) pits the weak, bumbling human against the infallible machine in all its glory (ok, I got a little snarky there 😛 ).
There is a lot of truth to this, and I trust machines more than I used to. We’ve seen smart bidding for instance work in great ways, but here are three issues I have with this line of thinking as presented by Patrick, that is the human having no role as a feedback mechanism but only getting in the way of the machine:
The machine isn’t infallible, it’s created by humans, for humans.
Saying there is no place for humans to offer feedback, especially the advertisers working on the account isn’t appreciating the reality that the machines are already human created. To reiterate my point in the original article, a machine is human created and human guided and ongoing human implemented boundaries are absolutely essential for a successful machine learning system.
As part of his argument, Patrick noted:
“In the post, he [Kirk] mentions that Google often says, “Machines do great as long as they have guidance.” This is true. But the necessary guidance primarily relies on quality conversion data and customer signal data. I don’t see how our own search term analysis would help develop any sort of better feedback mechanism.”
To directly respond to the last sentence above, I don’t see how one *can’t* use search term analysis as a better feedback mechanism!!
We have a large brand we manage for Shopping Ads, and the constant Search Term mining and negative keywords added to remove completely irrelevant keywords that continue to come through (even with specific feed optimizations and query filtering campaign strategies), constantly ************IS************* the feedback mechanism!! I’ve joked before quite seriously, that anyone who has run Search Terms reports on Shopping campaigns (even advanced strategies with tight query control like we run at ZATO) loses trust in just how accurate the algorithms can be.
That’s not to say all machines are bad, or all algorithms are bad. That’s to respond to the notion that we humans don’t play a vital role in ongoing feedback for the machines.
Distrust and transparency prevents wrongdoing.
The second point I would bring out, is one I’ve rarely heard discussed but is important. Let’s step back from the conversion-focused and which-data-is-helpful arguments for a second and ponder a darker reality.
In a world of *constant* corporate fraud, I would argue that closed ecosystems encourage wrongdoing. If someone is using Data Driven Attribution (DDA – a form of attribution where Google determines what visits get credit for a sale based on proprietary factors) and SSC… then they are handing Google a ball of cash, and trusting Google to spend it wisely and then report to them telling them whether they spent it wisely, with no potential for third party verification of the claims. The DDA SSC mix is a mix astonishingly devoid of third party verification.
I don’t think this point should be skipped over, transparency can make things inefficient for Google engineers but it doesn’t mean it’s not the right move.
Why do we need basic transparency such as which channels and search terms and placements/videos our ads showed on… even if we don’t do anything with that information? Because it helps guard against unethical behavior and ensures this Google thing keeps its checks and balances.
A completely closed system spending billions of dollars, will. get. taken. advantage. of. somehow. in this world.
Primary Point #3 from Patrick: “Smart Shopping works so well, that who cares about data?”
Finally (yeah, I didn’t think this article would get so long either), let’s consider the third primary point in Patrick’s article. He states specifically here:
“Ultimately, the only thing that matters is that Smart Shopping has been the greatest driver of success for our e-commerce clients over the last 18 months.”
On first blush, this point seems to track strongly (and I’m sure resonates with the “don’t bring an opinion to a data fight” crowd) with our objective as PPCers… to make our clients money!
However let’s first look briefly at the claim, and then step back to end this post by looking at the bigger picture first.
First, while I’ve definitely seen SSC work well in driving tracked profitability and conversions, this has not been a wholesale success across our accounts.
I’m always cautious in case studies (I view a claim such as that given above, and in the single account metrics of success attached in Patrick’s article) because context is so important. The Shopping Ads practitioners I’ve talked to who use advanced query structures (such as ZATO) have not typically seen as clear comparison success when testing SSC against their more advanced strategies.
SSC Boo! For instance, in one (Brand) account, we kept the Top Selling products (with plenty of conversions and data) in SSC when we took on the account and then built out a more advanced Standard structure. We’ve seen the Standard actually perform better in ROAS over time, even though the Top Products aren’t even in the Standard Shopping campaigns!!
SSC Yay! On the flip side, we’ve also seen a retailer with low margin, high converting products in another account (where brand search queries didn’t work great for us because of competition and retailer pricing issues) perform better with SSC.
From my perspective, SSC doesn’t win any and every time simply in tracked profit and revenue… though it certainly sees a lot of success.
Second, never, ever, ever forget the fact that Smart Shopping is targeting more channels than Search so a direct 1-to-1 test needs to keep this in mind, your Dynamic Remarketing campaigns (or lack thereof) are part of the comparison. But that’s part of the problem, you can’t even measure solely a Search to Search change in traffic since we can’t see Channel data in SSC).
My point? Actual success depends on many factors, and a good advertiser will determine what elements are in what account, what tactics the retailer or brand is trying to accomplish, and will implement advanced SSC and Standard as necessary and manage things well. The argument that we’re not all good PPCers is a poor one, IMO. Then maybe as I have argued previously, Google should be doing less to bypass the agency and more to help identify and match the actual good PPCers with accounts… for their benefit too!
To Consider A High-Level, Philosophical, Conversion Attribution View
Last but not least, I’d like to leave you all here:
I believe the original quoted statement is getting to the core of how PPCers need to begin shifting in thinking, and I recently wrote on exactly that here:
Digital advertising is not the dot com bubble, improper attribution is
Tagline: The real problem in paid search right now is our belief that we can track everything.
(As if you need more Kirk-writing to read, but if you’re a glutton for punishment then by all means… head over there! The TL;DR version is that we have forgotten how to market, and have instead become obsessed with chasing tracked conversions at any cost.)
If I have a primary concern with Patrick’s voiced sentiment, it is this final point. I’m not even yet arguing with whether SSC does always *work*, I’m suggesting even when it does that we should be cautious about how that fits into our marketing strategy rather than vocalize the need for all to switch to SSC because the machines do it better. But those machines are always limited to the data that is reported to them, and what about brands where some of the data that is valued isn’t something that can be worked specifically into a tracked ROAS target in the campaign settings?
To illustrate, there is a difference in how a retailer will and should advertise and how a DTC brand will advertise. We do a lot of advertising for DTC or DNVB brands, and we have noticed our core issues with SSC in these (rather than our retailers), that is when it comes to building and growing an actual brand… the weaknesses with the SSC system stand out remarkably more. We see an inability to easily maintain brand safety with placement data, an inability to easily target specific query groupings that a brand wants to break into on Search, an inability to easily allocate and play with the other Channel teams (it’s not uncommon for us to work alongside separate Video marketing teams, Search teams, Display teams and Email teams all for a single Brand… and you gotta love how SSC plays in all of those spaces sapping budget without any insight into which of those spaces it’s appearing in!).
In other words, I noted previously and agree in part, that we have seen SSC convert successfully… but to be frank, I think it’s too early to declare “business” success, and not simply Channel (or with SSC, multi-channel) success. That is why I’m so cautious right now. We’re so early in the SSC game that we don’t truly have the value of time to anlayze a brand’s transition wholesale to SSC and whether that really is ultimately good for a brand that wants to still be around in 20, 50 years. I’m concerned about losing all of that data, and switching wholesale to a black box solution for businesses playing the long-game and not simply obsessed with internal channel ROAS growth (especially when it involves an undisclosed chunk of remarketing/returning visits rather than prospecting).
I hope that point makes sense, as I think it’s an important one!
As noted originally, I very much appreciate Patrick’s article and response, and it’s helped me sit down and chew through just what I believe is worth fighting for, and worth considering, and worth giving on. I hope this exchange causes our entire industry to think together on this, and more importantly, causes Google to consider changing course on some of the data transparency directions on which they are moving.
I look forward to the conversation continuing!
This was the next segment in our series on Google Smart Shopping Campaigns (SSC). You can read the first 5 posts in this series here: