Attribution is a hot topic at the moment; agencies, advertisers and media owners all seem to be talking about it. Google and DoubleClick Search in particular are making this a focus of theirs and now offering workshops on attribution models and how to use them.
I work in Paid Search so naturally, I operate on a last click model. But I know this isn’t always the ideal attribution model for other channels, or even PPC for that matter!
For example, if we used the last click mantra to analyse how we all get in to the office, it could easily be determined that we walk (as we physically walk through the doors). This is great for the environment but not a true reflection of the data, some may have walked from their car, some from the train station and some from the bus stop. So yes walking is definitely an aspect of our commute but we wouldn’t necessarily go as far as to say we walk to work. Essentially, using the last-click model and applying it to our daily commute to work wouldn’t really help us answer the question of ‘how do we commute to work?’
Attribution can be a hard topic to approach for clients, partly due to the complexity of attribution but also because it is challenging the status quo – and does anybody really like change?! Since I’ve been at equimedia we have always challenged attribution, not necessarily for paid search but certainly for other channels.
But recently, due to refined technology and more stringent marketing budgets we’ve seen attribution models be questioned, particularly for paid search. The recent advancements in DoubleClick Search and AdWords have given us all the opportunity to explore different models and apply them to our data.
Another reason Attribution is a big focus (particularly for Google) is the rise of mobile search. As mobile search is growing and becoming more commonplace we are now seeing more disconnected and fragmented journeys than ever. The real issue is linking desktop and mobile users, Google have done a decent job of linking users across devices through Google apps and Chrome but it still remains a big problem. In some verticals we are still seeing lower levels of efficiency on mobile vs desktop, but we are well aware that more research happens on mobile and desktop still drives the majority of conversions. This is the reason Google are spending lots of time telling us all about the importance of mobile!
AdWords have recently released some interesting attribution reports in the interface which are largely designed to assign more value to mobile but they are definitely worth taking a look at! You can find them under to the ‘Tools’ tab in AdWords.
The cross device activity is relatively new in the interface. These reports can show you assisted conversions by device and common device paths, which are really useful and again can be used to aid your optimisations.
As well as this we can also see lots of new GA style reports such as top paths and time lags. You can learn more about these reports here.
DoubleClick Search have also invested time in attribution recently. They now allow you to apply up to 5 different attribution models to any account.
In DoubleClick Search we can now apply the following default models to our paid search data –
These are fairly typical pre-determined models that most of us would be familiar with. As I mentioned earlier, you can apply these to your paid search data now in DoubleClick Search and within 24hours you will be able to analyse paid search performance using a different model.
But what we are really interested in is the DoubleClick Search data-driven model.We attended an attribution workshop at Google which explained their data driven model and put it in to practice there. Naturally, coming back from that session we were all very eager to start applying our data driven models to our accounts! There is some eligibility criteria required for the data driven model to work, e.g. you’ll need close to 10,000 actions within the past 90 days for example, you can read more here.
Or what is DoubleClick Search’s data driven model and how does it work?
Firstly, for the DDA in DoubleClick Search there are no rules, the system checks data at every point and analyses converting and non-converting paths. After analysing the data the model then begins to assign value to each and every step.
The DDA model can analyse a 6 step process, it will start by assigning a contribution factor to the last 4 steps. The model will then analyse the individual steps and looks at the propensity of it, the model will then tweak the contribution factor of each step.
To put this in to context (and because it’s more fun) they used an example of a couple who met on Tinder getting married (this is the conversion in this example).
The 4 steps were defined as – Swiping right on Tinder, going for a coffee, start living together and proposal.
Now, in my world and the paid search world as it is we would assign 100% of this conversion value to the last step – the proposal. But in reality we know this isn’t the case, there are not many couples that get married from just a proposal.
If we were to ask Tinder or use a first interaction model we would assign 100% of the conversion to Tinder alone, but again in reality we know this is not true.
The most likely conclusion is that all 4 points each contributed in some way to the relationship and eventual marriage. But exactly how much? This is why DoubleClick Search need a substantial amount of data for analysis (at least 400 conversions in a 30-day period). Using the conversion data the DDA model can start to identify trends, throw in the non-converting paths also and you start to have a very decent data set from which to analyse. This DDA model becomes more powerful as DC also looks at Natural Search and Display conversions (based on our floodlight activity) – this allows us to now assign more (or less) value to our PPC activity and even report on this at individual campaign level.
Our early experience of the DDA model has been good, we have been able to tell one client in particular that their Generic activity is contributing to 4% more conversions via a DDA model when compared to the default last click model. As well as this the DDA model can go some way to proving Brand uplift through investment of Generics.
At equimedia we have expanded many of our PPC accounts with traffic driving campaigns which are designed to increase our cookie pools and remarketing lists. Many of our mature accounts cover all the necessary market to drive conversions, and therefore we have changed strategy regarding expansions and are striving to drive users to site with a longer term goal of remarketing them at a later date and driving the conversion. Obviously this means on a last click model these campaigns perform poorly and have a weak ROI, but using a DDA model we can look to attribute more value to them. I think this is a truer reflection of their performance and helps to attribute some value on our tactical campaigns and content.
I find the data that a different attribution model gives you is really quite interesting! However, interpreting the data is the key at the moment and whilst DS continue to refine the tool it should just be used as an education piece to aid results. I certainly wouldn’t expect any radical changes from it at this stage – there’s no immediate scope for switching attribution models! DoubleClick search also gives you the opportunity to optimise and flex bids based on different attribution models; this functionality is welcomed but at the moment we wouldn’t push our attribution modelling that far until we know more. We are going to spend more time studying and understanding the different models here and interpreting the data, it’ll be interesting to see if different models suit different verticals.
If you would like to know more about the various attribution models or are keen to see how you could benefit from attribution analysis give us a call!