Few things in marketing data intelligence are more confusing and contested than attribution models. There’s really no one-size-fits all solution for every business. Numerous things--your objectives, long-term vs. short term goals, available resources and budgets, what your company values and what each member of your team believes or feels--can determine which attribution model is best suited for your business. And there are several to choose from.
Before you go picking an attribution model out of a hat, or simply relying on your analytics platform’s default model (Google’s is last non-direct click, BTW, more on that below), let’s discuss the most common models out there.
NOTE: All of these attributions models are available in Google Analytics, and are the most common ones used in the industry. If you are using a different analytics platform or working with a professional, they may offer other models in addition.
This attribution model gives all credit to the first channel that drove a sale. Let’s say a customer clicks on one of your Google AdWords ads, visits the site, then leaves without make a purchase. That same customer comes back to your site by clicking on a banner ad, but still does not purchase. They click on one of your Facebook posts, but again, they do not make a purchase. Next, they search Google for your brand name and click on the organic search listing. And finally, they type in your URL directly and make their purchase. If using a first-touch model, the Google AdWords ad would be given full credit for the sale.
Not fair, you might say. What about the display ad, the Facebook post, Organic and Direct, you might cry. And you should. First-touch is a flawed model that does not take into account the full scope of a consumer’s journey.
Why, then, would anyone want to use this imperfect model? Well, if you’re only interested in seeing which channels generate new prospective customers, or which are driving the most awareness for your brand or products, it might be the best fit.
This model gives all credit to the last channel that drove a sale. In the example above, credit would go to Direct.
Again, this model doesn’t tell the full story. But it can be useful if you’re only interested in learning which channels are your closers, or have great awareness but a limited budget, and need to decide which channel(s) gets the spend.
Last Non-Direct-Touch Model
In this model, all direct traffic (people that type your URL directly into their browser, have your site bookmarked or click on a link in “dark” medium, such as a text or instant message) is ignored, and credit is given to the last click. In our example, Google search would get the credit, not direct.
This is the default model used in Google Analytics. And it too is imperfect. Not only does it overstate the value of the last click and skim over the full story, but it completely ignores the value of direct traffic. So any work you may have done to build brand awareness via any number of campaigns, channels or efforts is thrown out the window.
Linear or Even-Weight Model
The linear or even-weighted attribution model gives equal credit to every touch point along the customer’s journey. In our example, since there are five channels used by the customer, each gets 20% of the credit.
As Analytics Master Avinash so cleverly put it, this is kind of like a participation award--just because Display showed up to the race, it gets a ribbon. Even if it didn’t really do anything. Even if it made the consumer second-guess their purchase decision. Even if it tripped and took Email and Referrals out with it.
But the linear model does look at the broader picture, and takes every step into account. So it’s better than the ones we’ve already discussed, but it’s still not the best solution.
Position Based or U-Shaped Model
Here, the credit is spread out over all touch points, with more going to the first and last clicks (40% to each). Going back to our example, 40% would go to AdWords, 40% to Direct, with the remaining 20% being spread evenly across Display, Social and Organic.
This is starting to feel a little more legitimate. It is taking into account all channels, but highlights the first touch--the customer’s initial contact with your brand where you (we must assume) made a good impression--and the last touch, the point where you sealed the deal and they made a purchase.
The time-decay model gives the most credit to the channel closest to the sale, and each other channel get slightly less credit as you get further from the sale. Think of it like this: The more time that has passed since the click, the more the value of that channel decays.
This is probably your best, out-of-the-box attribution model (at least of the ones available in Google Analytics). It still gives some credit to each touch point, helping to show the broader picture, but rewards the ones that are closer to the purchase. In Google Analytics, you can set your own half-life for the decay, so that a click occurring, say ten days before the conversion, gets half the credit of a click that occurs on the day of the conversion.
The custom model is whatever you make it. You can divvy up credit percentages across the various channels, based on what your company values.
This model can often show the truest picture of performance based on the your specific needs, desires and goals. But it’s not for the faint of heart. You don’t want to assign credit percentages willy nilly. Start by comparing the other models using the Model Comparison Tool in Google Analytics, and get alignment from stakeholders across your organization.
Picking the right attribution model for your business doesn’t happen overnight. Even if you have no intentions of creating a custom model, you should still compare the different models and compare different metrics, too (look at cost per conversion, average order value, conversion rate, signals of intent, etc., not just conversions or sales). It does take some time, but the results can be well worth the effort when you are truly able to understand the impact of each channel and make smart business decisions based on performance.