#37 How attribution works in Google Analytics 3 and 4

The Measure Pod
The Measure Pod
#37 How attribution works in Google Analytics 3 and 4

This week Dan and Dara chat through how attribution works in Universal Analytics and Google Analytics 4, and the differences between them. They discuss the nuances and pain-points they have come up against, and what to look out for whenever using any of the built-in models.

The blog post from Dan they mentioned on how the Universal Analytics Campaign Timeout works is here.

Here’s a good resource on how to build various attribution models using the GA4 BigQuery export data from Krisjan Oldekamp at Stacktonic – https://bit.ly/3sdfuVa.

In other news, Dan watches snooker and Dara goes full Downton!

Check out on LinkedIn:

Music from Confidential, check out more of their lofi beats on Spotify at https://spoti.fi/3JnEdg6 and on Instagram at https://bit.ly/3u3skWp.

Please leave a rating and review in the places one leaves ratings and reviews. If you want to join Dan and Dara on the podcast and talk about something in the analytics industry you have an opinion about (or just want to suggest a topic for them to chit-chat about), email podcast@measurelab.co.uk or find them on LinkedIn and drop them a message.


[00:00:00] Dara: Hello, and thanks for joining us in The Measure Pod, a podcast for people in the analytics world. I’m Dara, I’m MD Measurelab, and I’m joined as always by Dan, who is an analytics consultant also here at Measurelab. Hey Dan, how’s it going?

[00:00:30] Daniel: Yeah good thank you, how are you?

[00:00:31] Dara: I’m also good. And I believe we have no news this week, so we’re going to launch straight into our juicy topic, which is drum roll please.

[00:00:39] Daniel: Attribution nuances across Universal Analytics and GA4, and it’s a bit of a broad, wide-ranging subject. But recently we did a couple of articles around attribution and lots of conversations have been happening at least internally at Measurelab. And actually all started with an article we wrote about explaining the campaign timeout in Universal Analytics and through that process actually, because that’s still obviously very prominent in terms of a tool and it’s still a question people have all the time, or maybe even a feature people aren’t even aware of. It’s actually wider than that, this is a one feature of the attribution conversation and as we move to GA4 from something like Universal Analytics, it’s not like for like, it is different and how they approach attribution is completely different. So I thought it’d be a good idea to just discuss it and maybe help ourselves as well as other people that are listening with some of those nuances.

[00:01:29] Dara: Well it is a favourite topic of yours, attribution, which might surprise some of our listeners.

[00:01:33] Daniel: That it is yeah. So I don’t know if it makes me an authority, but it definitely makes me opinionated.

[00:01:38] Dara: And the campaign timeout mode specifically. That’s something you’ve got some views on that, I know it’s always been something that you’ve had some strong thoughts on.

[00:01:46] Daniel: Yeah, for sure. And you probably have this as well Dara but running Google Analytics training, it’s one of those things that you have to introduce quite early on when you talk about Google Analytics, because it has such an impact on a lot of the metrics, especially a lot of the campaign data that you’re reporting on a Universal Analytics. So you have to introduce this subject quite early on when you do any kind of training in Google Analytics. But it’s also one of those things that it almost feels like it comes out of nowhere like it kind of takes people by surprise because it’s such a big feature that has such an impact, but most people or a lot of people aren’t aware of its existence, let alone the impact it may have. Even people that profess to be advanced at Google Analytics, sometimes this is quite often a feature that they’re not aware of.

[00:02:28] Daniel: So it’s one of those things that if you know where you’re looking and what you’re looking for, you can find it quite easily. But if you don’t then it’s just one of those things that you’d never come across on your own.

[00:02:37] Dara: So go on then what is the big impact of the campaign timeout setting?

[00:02:42] Daniel: Well, the campaign timeout, obviously I wouldn’t do as good a job as the article we wrote because that goes into a bit more detail with examples, but ultimately it comes down to the treatment of the direct channel within Google Analytics. So let’s assume quite simplistically at the moment that direct is the channel, which is navigational so if someone comes to visit your website via typing in the URL in the browser bar, I suppose the same can be said for clicking on a bookmarked link as well. If someone visits via one of those methods, that’s classed as direct.

[00:03:11] Daniel: So it’s not organic search, it’s not paid search it’s not social, it’s direct. And the thing that Google Analytics does is it tries to take value away from direct wherever possible. And the method it’s got to do that is using the campaign timeout, which is defaulted to six months where you set up Google Analytics, but you can change that but quite often people don’t but what it means is if a direct session has been seen in the data, Google will go back in that user journey up to six months to find a non-direct session to then overwrite the direct session with that previous attribution. So what that basically means is if a direct session has been made, so if someone visits your website by clicking a bookmark or just typing in the URL, it’s going to go back in your user history or at least in your Google Analytics history and find a previous campaign, let’s say the last time you clicked to the website was via an email link. And it’s going to copy the email campaign information, and it’s going to paste it on top of the session today.

[00:04:05] Daniel: So in a sense, you’ve actually had one click from email a couple of days ago, but you’ll have multiple sessions in Google Analytics, even if you navigated directly to the website.

[00:04:14] Dara: And the thinking, the rationale for that is that the assumption, whether you agree with it or not, is that you will be returning directly to the site because of that previous campaign visit. So the only possible reason you would ever have for coming back direct to the site is because you had previously been through a paid campaign. Sorry, I shouldn’t say paid campaign it could be a referring site or a social media link, but you know, the only reason you would ever visit directly is because you had previously been through a campaign.

[00:04:44] Daniel: Well, that’s the theory at least. So, you know, it can’t possibly be offline advertising or word of mouth or everything else. No, exactly there is no such thing. God forbid you went into an actual store, brick and mortar store and had some brand awareness that way. So yeah, in a very specific scenario, it makes perfect sense. It’s not necessarily a good or a bad feature, it’s not something that is inherently good or evil, you know, for the context of Google Analytics it’s just a feature that always exists. The way I suppose you can in a sense understand or justify it is because Google Analytics is a marketing tool there’s no secret there, it’s part of the Google Marketing Platform, it’s built by the biggest marketing company in the world.

[00:05:20] Daniel: Direct isn’t really a marketing channel, direct is in a sense the absence of a marketing channel. And what we mean is that you can’t invest time, money, effort, people, agencies into improving direct. You can do that for every other channel, you can spend more time optimising email, SEO, display, social, you know, all those channels can be optimised and they can improve, direct can’t. So in a sense, giving value to direct is not as useful as trying to reward it to the previous campaign that you can measure, because you can’t do any more of direct even if you wanted to. So I think that’s the justification, at least that’s my understanding of the justification behind the window, in practice, it just doesn’t work as well as maybe we’d like it to just because the nuances with direct as a channel.

[00:06:05] Daniel: So direct in itself, isn’t just navigating directly to the website. For example, if you’re not using UTMs, quite often that click on that link, regardless of the source might be classed as direct. So direct is actually, if you’re not doing decent UTM tagging of your links, or even if you’re not doing it at all, quite a lot of your traffic will be classed as direct. And if it’s classed as direct, it goes through this crazy six months campaign timeout process. So in a sense, let’s say you’re tagging all your email links, but you’re not tagging any of your social links. Then anytime someone clicks on a social link, you could be up waiting and reattributed back to the email campaigns, you’re bumping up another marketing channel just because you’re not using UTMs. It’s a very odd process, very odd thing to think about and try to explain it in an audio medium at least anyway.

[00:06:47] Dara: Probably an important clarification on the campaign timeout is that resets every time you visit the site. So in theory, it could be a lot longer than six months, because every time you come back to the site, it’s going to extend that cookie lifetime. And of course, I guess another clarification is that this is all based on cookies in the first place, which are becoming less and less reliable. The chances of somebody having a cookie for six months is probably pretty slim, but yeah, it’s worth noting that that six month window could continually get extended if somebody keeps coming back direct to the site.

[00:07:18] Daniel: Yeah, for sure. I think in safari now it’s got a seven-day cap right in quite a lot of instances. That’s assuming you’re not doing server-side tagging or anything like that of course. But yeah, there’s very limited situations where someone will have the same cookie for six months, but nevertheless, it’s still impactful even in the very short term, because any direct navigational visit is being classed as email, social, display, anything else.

[00:07:39] Daniel: But actually this ties into attribution modelling as a whole, because this is talking about session attribution, understanding the source of the clicker, the source of the session. This campaign timeout messes around with the session attribution. But when we talk about attribution modelling in terms of conversion attribution, it also has a part to play there because quite often you see documented or written down is that Google Analytics works in a last non-direct click attribution. That’s their default model is last non-direct click. So in a sense, it’s saying that if the last click was direct you go to the previous one, and then if that’s directly go to the previous one, if that started to keep going back in time until you find something that’s not direct for probably very similar reasons as we discussed for the campaign timeout.

[00:08:17] Daniel: However that’s not exactly true. The model is using is last click but if you layer on the campaign timeout on top of the last click attribution, that is one and the same thing. So if you’re overwriting the session to become not direct, wherever possible, that’s the same as doing last non direct attribution. So it’s semantics, we’re talking about the same thing, however, Google Analytics’ default attribution model is last click wins not last non-direct click. But the last click in combination with the campaign timeout is in a sense doing last non-direct click, but this is the point is that it’s important to understand the distinction because you can change that campaign timeout window. So you can make Google Analytics work, or at least the standard reports work in a pure last-click basis. But luckily we don’t need to do too much there because we do have the multi-channel funnels data and the sets of reports to be able to explore that.

[00:09:04] Dara: Yeah and I mean, multi-channel funnel reports are, I was always a big fan, I say always, I still am. I’m starting to think about Universal Analytics disappearing now, but I think the multi-channel funnel reports can be really insightful. But that is a different, it’s a different set of reports entirely, completely separate from the traffic data, the acquisition data that you get in the standard reports. And apart from the one change that you mentioned, so you can force the standard reports to be true last-click rather than last non-direct by setting the campaign timeout to be zero. But aside from that, you can’t change the default attribution model within the standard reports in GA (Google Analytics), which has been a frustration, not for everybody, but I know for certain users that has been a frustration over the years, that you haven’t been able to actually layer different attribution models over the kind of core reporting data within GA (Google Analytics) because the multi-channel funnels reports are a separate set, you don’t get to see that data alongside the traffic data for the different channels.

[00:10:02] Daniel: No, but as you said, the multi-channel funnels reports do give you a view of this. So you can go in almost, you can recreate those standard reports by going into two places, it’s a bit of a faff, but you can still do it. The multi-channel funnels reports, the reason why they’re separate is so that we can do this kind of user journey, attribution analysis. But one big thing to bear in mind if you are looking at the multi-channel funnels reports, there’s an API for this as well, if you wanted to pull the data out, but the six-month campaign timeout does not apply there. So direct traffic is classed as direct, and there’s a real big distinction because even things like just doing last-click attribution or last non-direct click attribution in the multi-channel funnels reports is not going to match the standard reports even if it’s using a very similar model, just because of the slight differences in how it will work everything out.

[00:10:48] Daniel: Again, the big difference between the multi-channel funnels and the core reports within Universal Analytics is the fact that the six-month campaign time is not present in the MCF reports. But what it does enable us to do is to do our own attribution modelling or using the standard models. If you want to, there’s an attribution project beta I think that came and went the way you could use data-driven attribution as well, for the free version of GA (Google Analytics) as well. But at its core, it’s a set of reports that enables you to do some attribution reporting and in the attribution model comparison report, as it suggests, you can compare multiple attribution models side by side. And this is a really good place where you can take that last-click and the first-click views and the linear, and maybe even the data-driven attributions or the time decay models, and you can see them all side by side to get your conversion and revenue numbers by channel, and if you want to go pull the traffic numbers, you can go do that from another report and work out your conversion rates or your ROI if you want to pull in spend as well.

[00:11:38] Daniel: So it does give you the option to be able to do all of that stuff. But like I said before, the key thing let’s say, if you take a linear attribution model, it will treat direct equally as it would treat every other channel because everything is what it is. You know, direct is direct, SEO is SEO, paid search is paid search. But in comes GA4 right, that’s not how it works in GA4, so I suppose, slightly left turn and go into the GA4 world when we’re comparing these things that’s where it all changes.

[00:12:05] Dara: Before you get too carried away with the GA4 stuff, that’s another interesting quirk I guess, nuance of the attribution data within Universal Analytics. As you say, because if you pick a linear model or time decay model or whatever, because direct is treated as a first-class citizen, just like any other channel, if you compare the default last non-direct click model against any other model pretty much. Direct suddenly gets a lot more credit and every other channel goes down as a result. I always found that a bit odd, I mean, it makes sense I understand why, but it does mean that if you pick pretty much any other bottle and compare it against the default in GA (Google Analytics) all of your marketing channels are going to look like they’re getting less credit and direct this big kind of black hole like we talked about earlier, which could be a mixture of offline, it could be true direct, and it could also be incorrectly tagged campaign traffic. That’s suddenly all going to get more credit than any other model that you compare against.

[00:13:02] Daniel: It’s actually really interesting you bring that up. When we talk about this in the Google Analytics training as well, a common initial reaction to this is like, why would you do that? That’s stupid. Because in a sense you can’t even measure accurate traffic by channel in GA (Google Analytics), right? Because it’s manipulating that kind of fundamental number that you kind of expect to do what you think it does. When I say to people, if you take it off, you can, you can take the campaign timeout window right down to zero, as you said Dara. But what’s going to happen at the moment you’re artificially up weighting all of your marketing channel performance and your down weighting direct. If you take that off obviously it resets back to the norm. But the thing is for the last 3, 5, 10 years, the norm for you has been this other way round, so it looks like your marketing performance just takes a dive. If you were to turn that window or that six-month campaign timeout window, if you turned that off today, as of tomorrow it will look like your campaign performance dates would take a dive, however, that’s not true. It hasn’t changed, it’s just the baseline is resetting. The baseline is becoming more evident, but it’s a kind of a very panic inducing, stress/anxiety inducing visual to see in Google Analytics. All of a sudden your conversion rates or even just traffic by marketing channel just take a dive off a cliff but for nothing, for no reason, nothing’s actually changed it’s just the way that we’re measuring it has changed.

[00:14:13] Daniel: Quite often that is a disincentive to actually change it because you know, the understanding, the baselines, the benchmark you may have made have all been in one way. And even if you don’t agree with it now, it’s kind of too late to go back and redo all of that.

[00:14:26] Dara: Exactly a bit like the decision Google would’ve made in the first instance to down weight direct, so there’s no perfect way of doing this. You kind of have to accept, we talk about this a lot, don’t we, but there’s always gaps, flaws, nuances in the data, and it’s never going to be perfect. If you try and seek that perfection, you’re just going to drive yourself around the bend because there’s always going to be a compromise in one way or the other, and this is just another one of those areas isn’t where the general consensus, at least for Google Analytics users is that direct traffic in GA (Google Analytics) terms is, is not a real channel, and therefore it gets down weighted.

[00:14:59] Daniel: I have a really interesting story of actually working with a client a number of years ago and we all collectively agreed that the relaunch of their new website, we turned off the campaign timeout because measuring accurately traffic volumes by channel was really important to them. And they understood the risks or the change in data to expect, and everyone was kind of on board and it was all great. And literally within weeks, I’d say about four weeks or so, we had about five questions from different parts of the business of like, why is our numbers looking so low? What’s happening? What’s going on? And in the end a couple of months later, I think it may have been only turned off for about two or three months, their PPC agency convinced them to put it back on because the PPC numbers look too low. So in a sense, it’s like change this feature to make our numbers look better, and they went for it, they ended up doing it. And I felt a little bit frustrated just because we tried to go in with our eyes wide open and understand what the effect would happen.

[00:15:51] Daniel: But in a sense, it’s like they haven’t made the advertising performance any better. They just made it look better via the, in a sense, the attribution model they picked. So I find it a bit disingenuous actually coming from an agency it’s like, we need our numbers to look better, not be better, but look better and so can you turn this odd feature that you didn’t want on in the first place, can you turn it back on please? So yeah, it was, I’m not going to name any names, but it was an interesting couple of weeks, you know, going into it. We all agreed, we went into it and then we backtracked out because of an agency’s numbers look too low.

[00:16:20] Dara: Yeah and that’s a common theme with attribution, isn’t it? This is a conversation we’ve had so many times where different people are going to have a different view of how they’d like to cut up the same pie. Yeah it’s often a case of making the numbers look better rather than making any actual positive changes.

[00:16:34] Daniel: You can actually make the numbers say whatever you want to say and that’s the risk of attribution modelling as a concept, because you can pick a model that makes your numbers look more favourable, your channel, your campaign, your remit look better in some way or another. And actually we see this already, whether you were aware of it or not. So if you go into each advertising platform, let’s say I go into Google Ads or Facebook Ads or Microsoft Ads, whatever ad platform I go into, or even your ESP, when they do any kind of performance reporting, well you have to use an attribution model if you’re doing any performance reporting.

[00:17:03] Daniel: So they’re going to be using an attribution model, that quite understandably, it makes their numbers look more favourable so that you spend more money and invest in them and keep going, right. And Google Ads does this, everything gets attributed back to the last Google Ads click because why wouldn’t you, if your Google Ads and the same happens in other ad platforms too, including impression data where possible, again, tries to up weight and attribute everything back to the channel that you’re spending money through ultimately. So right now if you’ve got an agency or if you’ve got marketers in house and you’re working with five different ad platforms and Google Analytics, you have five different, well, six different attribution models, all telling you exactly the same thing, but you’re cutting that pie in different ways. It’s not saying one is wrong, one is right, it’s not saying that why is my Google Ads report saying I made 15 grand this month from Google Analytics is saying five grand, neither one is incorrect, it’s just a different lens you apply on the data to tell you a slightly different story. And again, that’s the potential damage that attribution modelling can do, because if you take the best model in each case for each channel and add them all together, you’re all over-reporting right, you’re all claiming the same conversions and the same revenue.

[00:18:05] Daniel: So I suppose the nice thing about Google Analytics, or at least the nicest thing about Google Analytics is that it is a multi-channel reporting tool. It does measure all marketing channels in a relatively equal basis. The totals will still match basically, if you take all of the different numbers out of Google Analytics and add them together, they’re still going to be the total. We’re not going to be over or duplicating any values here.

[00:18:25] Dara: I took you away from your favourite subject you would just about to take us on to some of the differences in GA4 and then I clung onto the past and dug a bit deeper into Universal Analytics, but let’s tackle GA4. So how is GA4 different, I know for example, there’s no campaign timeout in GA4, how else is GA4’s approach to attribution different from Universal Analytics?

[00:18:49] Daniel: Yeah, you’re quite right there is no campaign timeout, or at least there’s low campaign timeout window we can configure any more and I think that’s a big distinction. Everything comes back to attribution modelling again, and rather than having these two concepts of attribution modelling and campaign timeout windows, they’ve kind of rolled them all up into one thing now. So you won’t see any mention of campaign timeouts but every single attribution model in Google Analytics 4 unlike Universal Analytics, excludes direct wherever possible. So the last-click attribution in GA4 is technically last non-direct click. The first click attribution in GA4 is first non-direct click, I suppose. So whatever model you use, whatever attribution model you use, there’s always a small asterix and a fine print that it’s like excluding direct, whereas in Universal Analytics it wouldn’t.

[00:19:34] Daniel: So the example we had about using linear attribution rewarding direct equally the same as any other channel if it was present in the user journey is not the case for GA4. So it might come out to the same kind of result, you can even recreate this kind of modelling in GA3 in Universal Analytics, because you can do custom attribution models in Universal Analytics you can actually build your own, let’s say linear attribution and add a rule to remove the value for direct down to zero. So I think you can recreate that but as a default, as a standard, GA4 excludes direct everywhere, and there is no options to not do that. Whereas in Universal Analytics, it defaulted to include direct everywhere and you had the options to remove it. So again, it’s just another thing to be slightly aware of when it comes to GA4 you are in a sense still down weighting direct, direct still has the same nuances in Universal Analytics. So if you’re not doing your UTM tracking properly, or if you’re not doing it across all marketing channels, direct means an awful lot more than navigational traffic as well. So still the same nuances, still the same caveats, the only difference is that you are going to be down weighting direct a lot more in GA4 than you ever did in Universal.

[00:20:39] Dara: I’m guessing the answer is no here but just to be really clear, there’s no dimension that tells you if it was a direct session or is there a no equivalent to the conversion path report that you get in multi-channel funnels in Universal Analytics?

[00:20:53] Daniel: Yeah so you still get rather than calling them multi-channel funnels, it’s now the advertising workspace and there are two reports in there. It’s basically MCF reports, but re-skinned for the GA4 world and you get your attribution model comparison report and you also get your conversion path report. So you do get both of those in GA4, it’s just that the modelling won’t value direct unless it’s the only channel used.

[00:21:14] Dara: So there is a paths report that is equivalent to multi-channel funnels, conversion paths. So you can see conversion paths with the true steps in the journey. Is it in the aggregated data it doesn’t give a value or?

[00:21:29] Daniel: Yeah it’s in the attribution modelling it’s the model rules ignore direct wherever possible, regardless of the attribution model you pick. So there is no, there is no attribution model where you kind of reward direct equally or fairly. The only time direct gets any value is if it’s the only channel used in the user journey. But there’s another can of worms there that we won’t open too much, but just to kind of tap on the lid as it were, you mentioned about the direct being the attribution for the session, and there’s a slight nuance now in GA4 where a session could have multiple different channels, multiple different campaigns. So in Universal Analytics, if you reset your campaign somehow, let’s say you had different UTMs, you clicked via a PPC ad, you left, you came back via an email, that will be a new session. In GA4 It doesn’t, so there’s now a question mark around that session now has PPC and email attached to it. Which one do you attribute to, which one do you class as the session attribution is all a whole world of interesting different ways of cutting this data. All of this data is available of course via the BigQuery export, because it’s an event schema, so each event has its own attribution, and then as you aggregate up into sessions, you can see how many different channels someone’s used within the course of that one session and then you can do your own, obviously attribution modelling on top of that.

[00:22:38] Dara: Does the export treat direct traffic as direct traffic?

[00:22:42] Daniel: Yes from what I understand direct is treated direct because it’s in a sense, it’s the absence of a campaign. So the source, medium, campaign field will be blank because there is no campaign, just because again, direct isn’t a specific type of navigation, it’s not actually typing in the URL it’s the absence of any UTMs and it just assumes an absence of UTM is direct and that’s how we’re doing it. If there was UTMs, it gets pulled through into those fields, those respective campaign fields.

[00:23:07] Dara: And I guess there’s a point that highlights, which is if there’s any activity that you possibly can tag then do so otherwise it’s just going to go into this unknown, direct bucket which could contain any number of different actives. So if it’s within your control to tag incoming traffic, then do so.

[00:23:25] Daniel: Yeah, exactly it’s just make sure you do UTM tagging as effectively and efficiently as possible and consistently, so that all channels are tagged wherever possible. With GA (Google Analytics) being a multi-channel measurement tool, if you don’t tag one channel it actually affects every other channel. So especially in Universal Analytics, because you have no real control over this six-month campaign timeout where people are unlikely to take it off. So you can’t use Google Analytics just to measure PPC it doesn’t work that way. You have to measure all of it or none of it to have the best quality data otherwise what’s going to happen is everything you are measuring is going to get up weighted by everything you’re not measuring, and it’s not a very fair representation.

[00:24:05] Daniel: We’ve seen this happen a number of times actually, when you start working with a new client and it turns out that they haven’t UTM tagged their email or their social media activity as just an example. And as they start tagging the email and the social activity, you see the value of something like organic search slightly go down because it’s taking away some of that attribution through that window.

[00:24:25] Dara: This is going to sound a lot easier to do than it actually is, but you could, if you wanted to, you could use the BigQuery export, you could create any custom attribution modelling you want, if you have the skillset to do so, including treating direct differently to how, if you, for some reason you decide, actually I don’t want direct to always be down weighted you can create your own custom logic for that through the BigQuery export.

[00:24:48] Daniel: Exactly yeah, I’ll try and find a couple of links to copy and paste attribution models you can use yourself even if you’re not proficient in the BigQuery world or the SQL world. So there’s loads of people out there because the lovely thing about the BigQuery export for something like GA4 is because it’s standardised because everyone that’s got this export it’s the same schema, a lot of people are sharing their code on GitHub, or they’re writing their own blogs with code you can copy and paste and use yourself, including things like attribution. So rather than letting the advertising platform Google do your attribution for you. If you wanted to have, I suppose the least bias perspective and do it yourself, you have that ability. But like you said, as long as you got the skills to be able to do that otherwise it just might not be an option for a lot of people.

[00:25:26] Dara: No and I wonder, so obviously, as we kind of covered a little bit earlier in Universal Analytics, you’ve got a custom model builder, that could be introduced as a new feature to GA4, but I’ll go out on a limb and say, I think that’s probably unlikely because firstly, you’ve got the option to have the raw data exported to BigQuery, which obviously you had with Universal Analytics as well or have with Universal Analytics, but only for 360 customers. But with GA4, it’s going to make it even more accessible. And also because they’re relying on data-driven modelling, I’m not sure why they would allow you to create your own custom attribution models. It seems like they want to make this a little bit black box.

[00:26:03] Daniel: I think that’s it really isn’t it? It’s the fact that they’ve gone all in, all their eggs are in this data-driven attribution basket or their modelling techniques. And I know we’ve mentioned a number of times about the different models that GA4 are introducing, but data-driven attribution is very, very important to them. For one thing, it is using your own data to predict forward and all the other crazy stuff, but it also means that they can layer in biases and nuances if they really wanted to, to upweight the Google advertising ecosystem, which again, you can’t blame them. But again, it doesn’t put your data quality at the centre it puts their profit margins at the centre.

[00:26:36] Daniel: You’re quite right with the data-driven attribution, I don’t think they’re going to even look or entertain the idea of custom attribution modelling because of the export, because of data-driven attribution. They’re moving Search Ads 360, Google Ads, DV360, all of them are being moved over to data-driven attribution models in some form or another, because of very similar things and even GA4 now defaults you to data-driven attribution. So I think in a sense they’ve invested a lot of time, effort and probably a lot of capital in doing this, I’m sure they’re not going to enable us to undo a lot of that.

[00:27:04] Dara: No and also just thinking about this, if the default position in Universal Analytics is last non-direct click attribution, and the majority of websites are still using that. This is the reality, as much as we’ve probably talked and lots of people have talked about different attribution models over the years. The default attribution modelling in Universal Analytics is still by and large the one that’s use for reporting data. If the default in GA4 is data-driven attribution modelling, then surely that’s a step forward, like a big step forward. So for anyone who wants to do their own custom attribution modelling, you can do that. You can do through BigQuery, for anyone else, this is a tricky thing to say, because what I was going to say is the data quality is going to be improved by the fact that they’re using data-driven modelling, but then you’ve got all the other issues working against that in terms of less collected data because of cookie issues etc. So on balance, maybe this is an area that’s improving, but then that’s just trying to offset some of the degradation of the data quality in other areas. So I don’t know, on balance, it’s probably a good thing?

[00:28:07] Daniel: It’s swings and roundabouts, isn’t it? It’s not better or worse it’s just different. And actually, this is one of my biggest recommendations with Google Analytics, with any data product, actually, but specifically talking about Google Analytics 4 a lot at the moment. Is that it doesn’t matter if you have the skills to go and undo this or do this yourself and we’re talking about if you’ve got a team of data scientists and engineers that can pipe the data out, play around with it, do your own reporting, awesome, great got do that. Even if it’s purely to validate their modelling, just to see if they are biased or not, you know, you can play around with a lot of clever stuff if you have the availability or access to those resources. Most people don’t, and it’s not about tough you’re out of luck, you can’t use this stuff anymore sorry. That’s really not the message that I want to portray, the point with all this stuff is just being aware of what’s going on so that at least you have an understanding of what they’re doing with the data before you make really big decisions, potentially around your marketing mix or your investments, or even the kind of the business or the merchandising even that you might have on the website.

[00:29:06] Daniel: So understand the perspective Google is coming at it from, understand what they’re doing to the data, even if it’s not down to the exact models they’re using, but it’s understanding what their approach is and how they’re manipulating data in the broadest sense. So that when you do decide to do your analysis, do your reporting, take action off the back of it, at least you have that in the back of your mind. Don’t be like, oh, it looks like display is got a 300% ROI that’s great, let’s double the investment there, you think oh actually wait a sec we don’t tag any of our email activity with UTMs, maybe we should do that first. Maybe before I start shifting around thousands of pounds of budget per week, per month, maybe I should take a breath and see if there’s another way to look at this, and especially when it comes to applying different attribution models. Google does make it very easy for us to layer on different models, to just see different perspectives basically, it’s different opinions of the same data it’s like asking five different people what they think, that’s what each attribution model is doing.

[00:29:58] Daniel: So yeah, like I say, it’s not about having the skills, it’s more about just being aware of what’s going on so that when the time comes to make some decision, you don’t go head first into something that you might regret later on.

[00:30:08] Dara: I think that’s a good place to draw a line under this for now. I know it’s a topic that certainly you could talk about forever I’d imagine, in fact is this going to bleed into your wind down. Is your wind down this week going to be working on attribution modelling?

[00:30:22] Daniel: I’d love to say yes, but no, actually the last two weeks or so has been completely taken up with the snooker world championship. For anyone that does enjoy snooker, I do, it’s almost quite Zen. It’s quite meditative watching, so I love watching it and then it’s very quiet, it’s very uneventful. And that’s actually one of the reasons why I like watching it, but the world championships is probably one of the biggest tournaments of the calendar wrapped up this last weekend on the bank holiday Monday. Yeah congrats Ronnie O’Sullivan, won his 7th world title, very impressive. Yeah let’s see if he can break the title next year and win it again.

[00:30:52] Dara: Well I also watched the snooker, but I’m not going to piggyback on yours too much. I did watch it, it was great, but I also went and saw Gary Newman perform in Brighton on Sunday night at the Brighton Centre. Gary Newman, in case you don’t know is an electronic musician, well kind of electronic rock crossover. So it was really, really good, he put on a really good show. He’s 64, and he was jumping around the stage and rocking out so it was really, really impressive. And I also went and saw the new Downton Abbey film at the cinema, I’m not sure what that says about me, and I think I enjoyed that as much as Gary Newman.

[00:31:28] Daniel: Oh, are you up to date then? Have you watched every Downton Abbey TV show and everything?

[00:31:32] Dara: Maybe. So what if I have?

[00:31:35] Daniel: Nothing, nothing. There’s no judgment, it’s just very interesting, very curious. I don’t think I would have put you as a Downton Abbey fan.

[00:31:40] Dara: I don’t think I would’ve put myself as a Downton Abbey fan. It’s just, it’s nice, everything works out in the end, it’s all very happy.

[00:31:47] Daniel: No spoilers, we don’t know it’s a happy ever after situation.

[00:31:51] Dara: So everyone ends up happy except for all the terrible things that happen.

[00:31:54] Daniel: Okay cool.

[00:31:55] Dara: All right, where can people find out more about you Dan?

[00:31:57] Daniel: As ever on LinkedIn, probably the best place to get in touch if you want to, and on my website, dananalytics.co.uk, and whenever I post there’s a email subscribe link there to get an email whenever I post stuff to the blog.

[00:32:09] Dara: And for me it’s LinkedIn, you can look me up on LinkedIn and find me on there. All right that’s it from us for this week, you can find out more about us at measurelab.co.uk, or you can find the previous episodes of The Measure Pod at measurelab.co.uk/podcast. As ever, if you want to suggest a topic or come on the show and talk to us about that topic, you can reach out to Dan or myself or both of us on LinkedIn, or you can email us at podcast@measurelab.co.uk. Our theme music is from Confidential, you can find links to their Spotify and Instagram in the show notes. I’ve been Dara, joined by Dan. So it’s a bye from me.

[00:32:47] Daniel: And bye from me.

[00:32:48] Dara: See you next time.

Written by

Daniel is the innovation and training lead at Measurelab - he is an analytics trainer, co-host of The Measure Pod analytics podcast, and overall fanatic. He loves getting stuck into all things GA4, and most recently with exploring app analytics via Firebase by building his own Android apps.

Subscribe to our newsletter: