In the digital age, estimating the impact of your online campaigns has become a lot easier. You open up your favorite web analytics tool and look at the channels report. Organic search, paid search and social media, they all get attributed a piece of the pie; whether it’s generated leads or ecommerce revenue that matters to you. One channel that typically does not perform very well is display advertising. I mean, who buys a new bicycle immediately after clicking on an online ad, right? The word ‘immediately’ is the most important word in my last sentence. Because it’s true, it’s not because the user didn’t buy anything after clicking the ad, that the ad is useless.
Enter: attribution modeling. To what channel do we attribute the purchase of a bike? The ad the user clicked on two weeks ago? The review article that linked to the bicycle store’s website? Our facebook post that linked to our blog message about the bicycle being nominated for ‘safest bicycle for cities’. The paid search ad the user clicked on? Or the visit where the user typed the URL or our website himself, right before buying his new ride? It’s called ‘modeling’ because there’s no right or wrong way to determine the value of each channel. It is entirely up to your expertise. How you decide to attribute the generated value should depend on your product and on the goals of your campaigns and channels. e.g. If you are introducing or positioning a new product in the market, that display ad has a more important role than when you’re showing ads in a remarketing effort.
Google: Doubling down on attribution modeling
However, in the model comparison tool, you can compare the ‘last non-direct click’ model with other attribution models such as:
- Last interaction model: a slight variant to the last non-direct click model in that it considers ‘direct’ as a full-fledged channel.
- Last AdWords click model: another variant to the last non-direct click model. It attributes all the value to the last click that came from an AdWords search or display ad.
- First interaction model: attributes all of the value to the touchpoint that the user first interacted with to reach your website within the selected lookback window (default, 30 days).
- Linear model: attributes an equal value to all the touchpoint a user used to reach your website within the selected lookback window.
- Time decay model: attributes more value to recent touchpoints and less value to earlier touchpoints.
- Position based model: attributes 40% to the first and 40% of value to the last touchpoint. The remaining 20% is attributed to every touchpoint in between.
It doesn’t end there. If you have access to Google Analytics 360, you are able to use the data-driven model. This model is completely different from the other models. For those with some background in game theory, the model uses the concept of the Shapley Value to calculate how every touchpoint contributes to the aggregated value: it looks at all the possible paths that users take to convert or to not convert to understand how each touchpoint impacts your users’ probability of conversion. This model will only work if your website receives more than 15000 visits and 600 conversions in 30 days. And if those out-of-the-box attribution models don’t work for you, you can create your own custom model. For example, a modification that makes sense is taking out ‘direct’ as a channel.
Finally, if your company is a power user of Google’s marketing product, you can integrate Adwords, DoubleClick and Google Analytics. When everything is integrated, not only clicks are attributed value, but literally every digital touchpoint (not just website visits) is taken into account. Think ad impressions, but also guest blogs, social stories, etc. In that case, please be careful when choosing an attribution model, because you don’t want to attribute too much value to ad impressions.
Why it matters
If you are responsible for large advertising budgets, you really want to put your budget where the money is. If you have been spending most of your budget on campaigns that generate the most leads or revenue and you have been looking at it from a last-click perspective, you might have been misallocating your budget. By comparing attribution models you will gain more insight in the role of each touchpoint. e.g. If you were expecting to see a lot of revenue from your Facebook advertising campaigns and you don’t see that in the standard Google Analytics reports, try using one of the other attribution models. If you see that your Facebooks ads get more value attributed when using something like a first-interaction model, you have proof that Facebook ads should be considered a top- or middle-of-funnel marketing touchpoint. In this case, using budget on Facebook advertising might be just as important as using budget on channels that produce the most value from a last-click perspective.
Here is another use case. Let’s say you are sitting around the table with different agencies, one who is responsible for search advertising, a media agency for real-time bidding ads through DoubleClick Campaign Manager (DCM), another one for social ads and finally, the agency that provides you with marketing insights. In most cases, their numbers won’t add up. It is probably because they use different ways of attributing value. The search advertising agency will pull their numbers from AdWords, which attributes conversions to the last Adwords click. The media agency will use the last DCM event model, which attributes value to their DCM campaign as either a view-through or click-through session. Lastly, the social advertising agency will pull numbers from their Facebook for Business account, whose technology doesn’t even communicate with Google ecosystem. The result is that you are left in confusion and the promise of big data, that it will help you taking the right decisions, becomes just a vague whisper. If that’s the case, it is up to agency in charge of marketing insights to unify those numbers in a cohesive attribution model that helps you decide on what channels to invest your marketing budgets.
Yet there are still hurdles to be conquered. Right now, attribution modeling through Google Analytics does not work across devices. If a user visits your website on their mobile phone through a facebook ad and then visits your website on their desktop computer through Google search, there’s no straightforward way to unify those touchpoints in one conversion path, unless you have a user id of every visitor on your website. In that respect, attribution modeling still lacks robustness.
However, by integrating DoubleClick Campaign Manager and through Attribution 360, Google promises that its attribution models are both cross-channel and cross-device, based upon the trusted devices that Google users add to their account. What’s even more spectacular; you will be able to establish a feedback loop between Attribution 360 and advertising platforms such as AdWords and DoubleClick, so that search and display marketers only need to focus on setting up the campaign, while machine learning will take care of managing it optimally.
Lastly, attribution modeling works perfectly as long as you stay in Google Analytics. As soon as you’re pulling data out, through the Analytics API, through Google BigQuery, or in Google Data Studio, you lose the capabilities to use and compare the attribution models that Google has developed. Manually pulling reports and copying it to an Excel file or Google Spreadsheet is currently the easiest way to go. If your company uses a data warehouse to combine data from different systems, it is up to your data scientists to build an attribution model from scratch. Although this might seem intricate, it can be a reasonable thing to do.
The concept of attribution is nothing new. Some decades ago, when marketers were discussing if it were either radio commercials or television commercials driving people to the store, they were having a discussion of attribution. However, with the explosive growth of online advertising and online channels, a last (non-direct) click attribution literally makes no sense. With Attribution, Google is simply trying to formalize a complex reality in easy-to-use tools. Not only is Google trying to prove its advertising products are working, they are also challenging competitor Adobe Analytics (formerly known as SiteCatalyst) which has had different allocation methods integrated in all reports for years now. With the accelerating release pace of new features in Google’s marketing products, we think it is safe to say that Google will stay online marketing leader for a long time to come.