So What? Marketing Analytics and Insights Live
airs every Thursday at 1 pm EST.
In this week’s episode of So What? we focus on how you can create attribution models in Google Analytics 4. We walk through what data you have available in GA4, which models you can recreate, and how to build them in Google Data Studio. Catch the replay here:
In this episode you’ll learn:
- What kinds of attribution are available in Google Analytics 4
- How to build an attribution model in Google Data Studio with Google Analytics 4 data
- What to do if you need more complex attribution than first and last touch
- Video SEO – TBD
- How do you benchmark a website’s performance? – TBD
Have a question or topic you’d like to see us cover? Reach out here: https://www.trustinsights.ai/resources/so-what-the-marketing-analytics-and-insights-show/
Katie Robbert 0:20
Well, hey, Happy Thursday, I can’t believe it’s already Thursday already. So welcome to so what the marketing analytics and live show today we are covering attribution modeling in Google Analytics for. And so just to sort of set the stage and attribution model, if you’re not aware is basically it’s a calculation to help you understand what gets credit for all of the different things that you’ve done. And we’ve covered the different kinds of models in other episodes. But at the end of the day, it’s really, you know, what’s working, what’s not working. And this is in context of your digital marketing. What we do know about Google Analytics four, is that it’s very different from Google Analytics three. And the big difference. And that’s what we’re going to cover today is that, that attribution model projects that you can set up and Google Analytics three does not exist in Google Analytics four. And so what Chris has been working on is figuring out how to rebuild an attribution model using the data from Google Analytics four, and he’s built it in Data Studio. And so we’re actually going to cover that today. So, Chris, John, any other comments before we sort of dive in and see what it looks like?
Christopher Penn 1:36
No, let’s go. So we’ll look at attribution models in Google Analytics for it for different places, so we can show what’s different. But to your point, Katie, I’m gonna go ahead here and switch us to the screen. This is what you’re used to seeing, right? You got all Google Analytics, three, you’ve got things like assisted conversions and stuff in here, your model comparison tool. And of course, that attribution beta model, which is there, it’s nice, it’s easy to use, relatively, I want to move over to Google Analytics for a it’s not there. Which is a little bit painful. So the question is, where did it go like, Google could not have possibly built a piece of software that didn’t have attribution in it? Right, that would be weird.
Katie Robbert 2:27
Well, and what we do know is that their Google Analytics four is still technically sort of a beta project. And so what we are advising our clients and sort of anyone else who will listen to anything we have to say, is to not abandon Google Analytics three, but to set up Google Analytics four in parallel, because Google is slowly rolling out features. So for a while, we actually thought the channel groupings would not exist in Google Analytics four. And we actually just saw it was either yesterday or the day before that they do exist. Now, they look a little different. And you can’t edit them the same way that you could in three, but they’re back. And so they might bring back attribution modeling, but we just don’t know if and when that will happen. And we don’t want to have that gap of not having that information, especially if you’re making decisions with it around how to budget and how to resource your teams. So we want to make sure that as best we can, we can recreate some version of attribution modeling, using Google Analytics for data. That’s right.
Christopher Penn 3:32
So there is attribution modeling built in. But instead of it being its own discrete section, it’s kind of hidden away, it’s tucked away inside Google Analytics for so let’s take a look at this. The first thing you need to have prerequisite is you need to have gone through and set up events. So you would have done this in Google Tag Manager. And you declare them here in Google Analytics, you create your events, and then you obviously mark those events that you think are important as conversions. Once those conversions exist, then you can start to put together at least two very, very basic attribution. So I’m going to go into conversions here. And these conversions, these are the ones we said, Yep, that’s a conversion. I’m gonna go with any thank you page, because that’s an easy one for us to do, right. So here, we have conversions by event source. What is your what what happened, and I’m going to go to the little widget up here. And here is where attribution kinda got tucked away. So this gives you four different methods for computing attribution in Google Analytics for for any of your goals, and that’s kind of cool. That’s kind of handy. It means that you can look at models on a per conversion event basis, the challenges. These conversions are very different in definition than what we’re used to. So you have the first one there, cross channel, last, click Second is cross channel last engagement. Third is Google Ads preferred last quick. And fourth is Google Ads preferred last engagement. Okay. Your facial expression says, What’s going on here? Right?
Katie Robbert 5:14
Well, that to me says that, you know, and you know, I don’t know the deeper definitions, but they all look like basically a version of a last click attribution model, which is the version that we try to advise people against unless you’re using a very simplistic ecommerce website.
Christopher Penn 5:32
Not only that these are Firebase conversions, because remember, Google Analytics for really is just google Firebase. So what what are those definitions cross channel last click says that ga for aka Firebase looks at all the clicks from all sources and attributes conversion to the last click that was is not direct. Right? So it is a non direct last click. Cross channel last engagement is Firebase under the hood looks at the last click or ad impression. If there is no last click, excluding direct, so it’s either ad impression, or last click is that second one. The last two. Google Ads preferred last click simply means we’re going to give credit to Google ads for the last click No matter what no matter what happens. Yeah, exactly. And again, for mobile ads, Google Ads preferred last engagement says we’re still going to give credit to Google ads for the last interaction. So those are the four models built in here that you can see now, you could be forgiven for saying this is sort of not really helpful, right?
Katie Robbert 6:47
It’s not because essentially, what you’ve been given is a last click model and a position based model, a position based model is where you determine what gets the most credit. And in this instance, Google has said we do we get the credit. So give it to us.
John Wall 7:03
Exactly. So the account managers basically built this. Exactly.
Christopher Penn 7:11
Okay, so what can we do to get to get at this maybe in a way, it’s a little bit more sensible, I’m going to go and actually bump this into the analysis hub, which is the exploratory BI tool. And now you can see here we’ve got, you know, event source, cross channel, last click, and then number of conversions, I can add more columns here than just event source cross channel. Last click. So I’m going to do let’s do a user’s us, let’s look at the different sources we have, I have a session source and a user source. So these are the different kinds of attributes that I could potentially plug in here. Remembering that their user session and event are different scopes. So the user source is that person was the last source that we know of from them for the last click, the session is just that particular session. And the event is literally that conversion action alone. So you could add in user session or event to try and do some comparisons, let’s go ahead and add in user source here. drag us the source in. And now I’m starting to be able to see, I can start comparing the user source, the source by which the user was first acquired. And then the last click write the source which conversions attribute based on the very last click so I can see 41 users almost timely was both the origin. And the last thing they did some of these however, like Trust Insights, so our website was the first one. And the last was the was Google itself. Some of these almost time is the first touch and then Gmail, their regular gmail account was the last some like, was organic, and then email is the last.
Katie Robbert 9:03
So shouldn’t those two columns be reversed?
Christopher Penn 9:06
Sure, I can just drag that out.
Katie Robbert 9:07
Okay, cuz I’m like, wait. And it’s one of those things like, as you’re putting it together, you want to make sure if you’re giving this to someone else to look at it, it’s intuitive. So Chris, you were reading it right to left, but we want it to be left to right.
Christopher Penn 9:24
Yes. Oh, unless you’re Japanese, Chinese or another language?
Katie Robbert 9:29
Well, I can’t speak those languages, unfortunately. So I would be looked at. Right,
Christopher Penn 9:34
exactly. So now we’re starting to get at least in the analysis of this broad analysis, but the challenge is, this is still kind of hard to read, right? It’s still not the most intuitive thing. But those definitions are really important. The user source and Google Analytics four is essentially the first touch the event source of the session source is effectively the last To touch. And so you could put the two together to get a sense of Okay, I can at least see where people are coming in first. And then what was the last thing to push them over the edge? What analysis have been used what you saw here? This is not the place we want to send people. This is not what you want to show the CEO. Right? This is a this will not go well for you. So what do we do?
Katie Robbert 10:26
Data Studio. Right. That’s, that’s the name of the show today.
Christopher Penn 10:34
Exactly. So God hadn’t connected Google Analytics for here to our Data Studio workbook, I’m going to put in a new chart, Ellis is going to call it a heat map. And I want in this case, let’s do our first touch, right? Our user source. And maybe our user medium factor might even be better have medium, which is because remember, medium is effectively the default channel grouping in Google Analytics for and I want to see conversions.
Unknown Speaker 11:12
Maybe views too, but conversions are kind of more important. So let’s go ahead and make conversions first on the table.
Christopher Penn 11:20
Make this a little bit wider here. and resize our columns fit the data evenly make it a little bit bigger. And so now I have a nice first touch table, right? This is the first touch for the conversions. And let’s sort it by conversions. Because again, we like conversions. They they make us happy.
Katie Robbert 11:47
If people are wondering why John and i are watching this kind of slack jawed is the first time we are also seeing this come together in this version. We’re both like, Wow.
John Wall 11:56
It’s like watching Bob Ross. He’s like happy conversions. These are happy.
Katie Robbert 12:01
Chris is the Bob Ross of data science.
Unknown Speaker 12:04
So that means we need to start changing colors here like Yo, what’s a happy color?
Katie Robbert 12:12
Blue and teal are pretty happy.
Christopher Penn 12:13
Yeah, that’s they’re pretty happy. Okay, so let’s label this too. Because as with any good thing, we do want to have some kind of labeling. So first touch attribution model. Bold, crank up the size a little bit. There, the first touch attribution model, that’s, that’s good. What we want to do now is clone that table. I’m going to rename
do less touch table. Now, instead of user medium, know what we were saying earlier. You have session medium is sort of the last touch. Let’s go ahead and get rid of the usernames put in session medium session source.
Katie Robbert 13:16
So I don’t know if I’m getting ahead of you, Chris. But it looks, are you restricted to being only able to sort of recreate the first and last touch models? Like are you unable to recreate things like the linear and the time decay models?
Christopher Penn 13:31
As of right now? That’s correct. We can talk about that in a little bit. So what you can do as a workaround for that, but it’s not pretty. Okay. So now, we’ve got ourselves a last touch attribution model with session. This is the last known thing. And you can see there are a few differences along the way. For one thing, the numbers are slightly different. Let’s kick it out of design mode into interview mode here. The numbers a little bit different in the first search modeled, the this email newsletter gets 100 conversions and the last search model 107. Right. So the there is a difference there. You see, there’s differences on view counts as well. And you see like for number five in the first touch model, Trust Insights, the website itself gets a credit for 15 commercials that same slot for last such model a different website gets credit for it. Alright, so there are differences between these two. And so, the question that you want to ask yourself as a marketer, the so what is which model? Do you have more control over? Which model Do you have numbers that are, you know, maybe more concerning to look at. So, for example, when you look at this last touch model and and compare the views to the conversions, we can see that there are some places where the numbers are pretty substantially off, right? So we see you know, nine conversions here and 760 views of This year, and it’s 717 and six conversions here. So last touch model for that email channel. last touch seems to be the the model that is more, I would say, favorite for an email newsletter Make sense? I mean, it’s less touchy. We tell people that our newsletter, hey, you should buy our data science one on one course for 299 from TrustInsights.ai dot AI well.
Katie Robbert 15:24
But Chris, but that’s based on our current attribution model, which we do outside of Google Analytics. That’s
Christopher Penn 15:31
right. And this is where this is where the challenges are, you have these two models. And today, which is April 22 2021, this is all there is in ga for unless you have the ability to build your own attribution models from your raw data, which is a slightly heavier lift.
Katie Robbert 15:58
So Chris, one of the questions that we have, is this what you’re doing here doable in the free version of Data Studio? My understanding is that Data Studio is free to everybody. And so is Google Analytics. So Google Analytics 360, is the paid version. And Google Analytics three universal classic, however you consider it is free to everyone. And I believe so is Google Analytics four, it’s sort of it’s now the add on to Google Analytics three.
Christopher Penn 16:29
That’s correct. And as far as I know, as of right now, there is no premium version of ga for and so everything that we’ve just done is all with all the free products.
Katie Robbert 16:41
Make sense? And that’s super helpful, too. Because as we’re going to see, you know, at least I would imagine, for the foreseeable future budgets are still going to be pretty restrictive. And this kind of thing is not where people are going to be spending their money. If this is something that they have to pay for, they probably skip over it for now.
Christopher Penn 17:00
Exactly. Right. Exactly. Right.
Katie Robbert 17:04
So John, you know, as you’re talking with prospects, and as you’re, you know, just sort of chatting with the community. I’m guessing that this isn’t even a question that’s being brought up yet. Like, can I do attribution modeling in Google Analytics for Are you still seeing that people are sort of still wrapping their head around attribution modeling and Google Analytics? Three?
John Wall 17:25
Yeah, it’s, most people realize that if they want to do some attribution, that getting it all on Google Analytics is not going to be enough, you know, they’re, they’re gonna have to figure out some way to set up some additional goals to get Salesforce or Marketo, or Eloqua, or whatever, to get that tied in there to get things working. And then I think a lot of people are just not even aware that ga four is coming down the pipe, and that it’s going to be considerably different. You know, so yeah, very few people are talking about, you know, I don’t hear it, I’ve not had anybody asked me about attribution and ga for like, that’s completely beyond where they’re going. And where it’s at now to is, again, people are usually coming to us, because they want their attribution model to be a little more advanced and take more into account. And so going to them and saying, Hey, you know, you’re going to get a first and last touch option. That’s just, that’s a no go right from the start.
Katie Robbert 18:13
Make sense? Alright, so Chris, so you said that there’s other ways that we can be looking at our data in an attribution model, format, what does that look like?
Christopher Penn 18:27
You’re not gonna like this answer.
Katie Robbert 18:29
I mean, it’s the day ending and why.
Christopher Penn 18:33
The way that we’re going to get good multi touch attribution models, or any of the models is to use the raw data itself. So one of the features that is available in Google Analytics for and it does cost money, not much, but it does cost money, is the ability to send your Google Analytics data to a Google Big Query database, Big Query is their huge Cloud SQL database where you know, it lives in the sky, and all your data goes there to die. Not really
Katie Robbert 19:07
well, it can go there to die if you send it there and never pull it back out again.
Christopher Penn 19:11
Exactly. And what you get inside of BigQuery, is you get the ability to see what’s happening at the individual hit level inside Google Analytics. So you get for example, the event names, the event timestamps and what, what page somebody was on, and then all of the attributes that go with each individual event. So I’m scrolling over here you can see the operating system, the language, refer what country the person is from the region, there’s our traffic source or traffic medium. Our stream ID if you want to know if you’re gathering for both mobile and non mobile devices. And obviously if you’ve configured any of the other fields that are available, you also get critically timestamps, and the user, the anonymized user ID. Now, to build an attribution model, what you can do is query the Google BigQuery database, and pull out this group by these individual IDs that exist in here. And the actions that somebody took, like, what page were they on? Or what source? Or what medium do they come from. And with a sufficiently robust enough piece of code, reverse engineer and attribution model, and at that point you would use, you could use, you know, the traditional ones, like linear or time decay, if you want to write code to do that, or the more sophisticated ones like Shapley values or Markov chains. But the key is, because Google Analytics four gives you the raw data, you have the ability to build these models yourself. And I suspect, I don’t know for sure, because I don’t know. I don’t get inside information from Google, none of us do. I suspect that Google has made this available so that people can build an app ecosystem around Google Analytics for right around around the Firebase ecosystem. There are, for example, plenty of apps in the app store that allow you to do more with your Firebase Analytics. And all they’re doing really is tapping into these databases. Here’s the catch, you’ve got to be able to code to do that. That’s, there is no, currently there is no way to build those models, without dipping into the coding.
Katie Robbert 21:38
So it’s not something that you could say, Okay, I found my data in BigQuery, I’m going to bring it into an Excel spreadsheet or even Tableau and, you know, group things together.
Christopher Penn 21:52
No, because you could absolutely look at the raw data itself in a spreadsheet, and it’s effectively just what is the spreadsheet. But the method by which you build attribution models requires computations that you can’t easily do in Excel, you can do, you could do some of them, you know, for example, you could do first touch pretty easily, you sort by your group ID, and you see, what was the first source, you know, by by timestamp, that’s pretty straightforward. Same for last touch, sort by group ID. And then, you know, what’s the last event source that you see there? Okay, that’s your last touch. But for things like linear credit models, time decay models, and multi touch attribution, you gotta be using something a little bit more horsepower, as much as I love Excel.
Katie Robbert 22:37
I think that’s totally fair. I think that one of the other things. So you mentioned multi touch. So people are always wondering about assisted conversions as well. And so it looks like that’s not available at this time through ga for either, is there a way to add that in? Like, could you, I guess, either in here or in Data Studio, add another column for another piece of data that would tell you an assisted conversion, for example?
Christopher Penn 23:15
Katie Robbert 23:17
And I’m not sure what that would be? Because I feel like you have you have user source and event source, which are really the two things that you want to look at. I don’t know if there’s anything that falls in between that that says, Ah, I don’t know, they left and came back, you know, maybe it’s some other kind of session information.
Unknown Speaker 23:38
Mmm, hmm. Let’s try this, this time to experiment. Let’s go back to page 12. And I work over here.
Christopher Penn 23:50
And let’s see about creating a new table is my need to choose our ta for.
Katie Robbert 24:07
And as you the viewer is doing these kinds of things, it definitely makes sense for you to be labeling. We’ve covered Google Data Studio reports before and one of the things that John and i talked about was making sure that you’re labeling the pages. And so as you can see, Chris labeled the page in the book, our book of reports as the GA for attribution. So you’re not looking at and going gold was the three was this for? I don’t really remember. So making sure that you’re doing that kind of work upfront so that people can very easily tell what it is that they’re looking at.
Christopher Penn 24:43
I think there is a way to do
Unknown Speaker 24:52
let’s take a look. Is that name
Christopher Penn 24:58
Okay. Yes, I think is a way to do like you’re saying the least a linear journey. But it’s not pretty
Katie Robbert 25:12
well, this is the first time you’re attempting it. So I think that if we can at least get it somewhere, then there’s definitely room for refining it a bit. You know, it might also just be worth noting that one of the questions that we, as consultants get all the time is, can I look at the individual user level data in Google Analytics, and in Google Analytics three? Not really. I mean, you have the de identified unique ID, and it looks like that carried over into Google Analytics four, you still cannot tell who the individual person is. And that’s by design. Google Analytics is not the place to be housing and looking for personally identifying information. Now, if you have other systems like a CRM, you can certainly hook up Google Analytics into that and start to do some of that matching. But just straight Google Analytics alone, you will never know who exactly the person is. And again, that’s by design.
Christopher Penn 26:19
So what we had to do here is, we had to connect directly to the Big Query database itself, we couldn’t we were not able to connect with, we can’t get the data that we want out of Google Analytics for its integration into Data Studio. So what I did was I connected to the BigQuery table itself, right? This lovely thing. And the attributes that are available in there, even though it’s functionally the same thing as the Google Analytics for table, the attributes that are available from digging into BigQuery directly, are different. What you can see is I can get at that user ID, I can get at the time sequencing, the things happen, and I can get at the source and or the medium, and then be able to process that. It’s gonna take a little bit more work. This is not complete. But I think this is how you do certainly linear attribution. Because you’re able to see all the sources that occur on the way to a conversion. And this would be the starting point for building the more sophisticated attribution models, you could theoretically I guess, bake a calculated field for half life’s for a time decay model. You can’t do multi touch this way. It will not be correct. But that speaks to the I think, just how still under development, Google Analytics four is. And the fact that the package software, namely Google Data Studio, has been formatted, to take one data source, and look at it two different ways. So you get two different sets of answers. And I can guarantee almost no one using Data Studio knows to look at it both from a Firebase perspective, which is what this is, versus looking at it from a Google Analytics perspective, which was our previous charts here. So there’s two different ways to look at this the same data and potentially get different answers. My advice now would be if you if you need the other attribution models, just get your DeLorean and go back in time stick with Google Analytics three, until the attribution features in four are more mature. Because, as we saw, even though, we were able to assemble first and last touch based on the data, these are not the ideal models.
Katie Robbert 28:55
No, and not for the type of business that we operate. And so, you know, again, the reason for that is because we use more than one channel, we use email, we use organic social, we’ve now included paid social and paid ads, and eat and other methods and direct. So what this is not telling us is where all of those things play together, this is only just telling us the very first thing somebody saw, or the very last thing somebody saw. And as you have a complex mix of channels, these are not the models that you would want to that you want to be using to determine what’s working. There are the models you want to use if you don’t care what the other teams are doing. And you just want your stuff to look really good. Which quite honestly, that works for a lot of companies because they are over investing in one specific channel, even if people are still coming to them through, you know, SEO or organic social, they don’t really care about things they just want to know. You know, where does email fall? So it really depends. But for us, these models don’t make sense. Exactly,
Christopher Penn 29:59
yeah, and Anything which is more than just a single touch single transaction, these models are starting point. But they are by no means the finished product. But as as we were talking about, you can see there’s differences even in just, you know, looking side by side, the sources and the mediums and the conversions of the views show up differently for each of these. So that’s sort of the state of Google Analytics for attribution right now, it’s not great. I fully expect there to be a cottage industry popping up as more and more companies adopt ga for of companies that can offer software that can do some of the attribution modeling, I would suggest that every marketer, take a hard look also at the attribution models that are available in your marketing automation software, as long as the data is connected properly. And it’s added has more than last touch. I mean, if it’s last touch the know better than this, but if they do have true multi touch attribution, you may want to be looking at that as a source of truth as well.
Katie Robbert 31:02
So Chris, as a marketer, what should we be thinking about in terms of Google Analytics for and reporting? Because you’re obviously you’re talking about attribution modeling, which is a more advanced application of the reporting. But even basic reporting, these are the Google Analytics four doesn’t allow for any kind of reporting within the system itself. So should marketers be worried?
Christopher Penn 31:28
This some, this some I mean, so you have the basics, right? Then you have, like, Where did my users come from, which, again, this is user source source medium, right? You have traffic acquisition, which is session source medium. So again, a lot of these things are kind of baked in a little bit. And these are okay, starting points. Again, for English. The thing to remember is that Google Analytics, four is not a reporting tool, right? It is a BI tool, it is a true analytics piece of software. And that’s its do, that’s what it’s designed to do. And that’s why it’s so different, because it really is all about analyzing your data. That’s why the analysis have really is where Google wants you spending your time doing analysis in here. And then you’re picking apart your data. And it’s pretty clear by the design, if you want reporting, you should be reporting in Google Data Studio, right Google days is where Google says, Hey, this is the reporting mechanism. Just like they say, you know, when you look in ga fours, configuration, where’s goal setup, it’s not there, there’s you can mark things as conversions. But they really want you doing the implementation goals in Google Tag Manager. That’s, you know, that’s is very clearly how they want all of us working. So from a marketing perspective, right now, if you’ve got stuff that’s working in ga three, keep working with it, it’s not going to go away soon. And start the planning process for migrating to Google Analytics for it, the first things that you should be doing no more creating goals that are destination goals, or anything else in ga three, every time you put together a new goal, make it an event based goal, because it is much easier to when you’re setting it up and Tag Manager to then add the corresponding ga for goal right, it’s so much easier to do that do it once but properly with in Tag Manager then to set up destination goals a ga three and then go, I got to go and set up a tag and trigger and stuff for the same goal in ga for to at all tag. So no more destination goals. And start moving your reporting into Data Studio, whether you’re on ga three or four, doesn’t matter. Move as much of your reporting into Data Studio as you can get it away from get people out of the habit of using Google Analytics for reporting. It should just be an analysis tool.
Katie Robbert 33:54
That completely makes sense. John, any final thoughts?
John Wall 33:59
Yeah, I just love that idea of getting, you know, users over in Data Studio because you know, everybody’s had their horror stories of when they’ve let the whole world into the analytics and they broken stuff and thrown filters on you You weren’t expecting and all that stuff. So yeah, I’m all for Data Studio.
Katie Robbert 34:14
I agree with that. And if you want help with any of this, we’re around feel free to reach out to us. All of our contact information will be available in the parting video, which we love to play at the end. It’s got our fun techno music. And I think unless people have questions, feel free to drop them into the chat. That wraps it up for this week’s So what?
Christopher Penn 34:37
All right, we’ll sort of we’ll talk to you all next time. Thanks for watching today. Be sure to subscribe to our show wherever you’re watching it. For more resources. And to learn more, check out the Trust Insights podcast at Trust insights.ai slash ti podcast and a weekly email newsletter at Trust insights.ai slash newsletter got questions about what you saw on today’s episode. Join our free analytics for markers slack group at Trust insights.ai slash analytics for marketers. See you next time.
Transcribed by https://otter.ai
Need help with your marketing data and analytics?
You might also enjoy:
Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, Data in the Headlights. Subscribe now for free; new issues every Wednesday!
Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new 10-minute or less episodes every week.