Lately there is a lot of talk about how “last-click is dead” and people are discussing other models, like first-click attribution models. However, there is an important distinction to be made between rules-based attribution models and algorithmic attribution modeling.
When you agree to apportion credit based on frequency, recency, perceived strength of the interaction, etc. you are using rule-based attribution. The default attributon model in most tools, last-click, is a rule-based model. By today’s standards, this rules-based approach seems crude. The althernative is the algorithmic approach (statistical algorithms) to attribution in order to the determine the weights. The algorithmic approach is similar in that it aims to weight things like frequency, recency and the quality of the interaction in coming up with a composite credit for each conversion.
I hear you thinking sure, got it. But that’s not all there is to it – because if you use last-click attitribution, that means that you can only award credit to campaigns that get you clicks. And that might be nice for Search, but where does that not work out at all? Right, in Display and even more so in video, where clicks are far from the right metric to judge effectiveness by. In Display banners and video, the impressions and views drive your prospects to conversion. Furthermore, your clickers might not be your converters (source: Quantcast PDF). I hear you thinking “sure, I get that, but how to measure all that?” – Well, make a first move into a better, or less wrong, attribution model by including impression impact in your model and assigning some credit to it.
So, are you rushing into your Google Analytics at this very moment to include impression interactions into your custom attribution model (see here how). Let us know how this works for your campaigns!
Although it has been noted many times that “last-click attribution is dead”, most marketers still use last-click attribution, often simply because it’s the default setting in the tool they use to measure conversions, like Google Analytics. You might say “But Google Analytics has these other models you can pick!”. Sure, but they are still click attribution models, aren’t they? Totally awesome if the first-click attribution model suits your marketing better than last-click, but what do you do when you’re using media where The Click is not the most relevant KPI to judge the media?
In a study, Quantcast argues that:
“For display advertising, clicks aren’t just suboptimal—they’re anti-optimal. And they’re likely to produce significantly poorer results”. (PDF here)
It’s not a surprise that marketers are so click-focused; the Click and the CTR are great indicators of your campaign health – when you run a Search campaign, that is. And Search just happens to be the channel online marketers spend their lion’s share of budget and time one. In search, optimizing clicks is essentially optimizing for conversions because they’re so closely aligned. With display ads, if you optimize for clicks, you get . . . clicks. But, you don’t want clicks, you want conversions, right?
The display experience is quite different from the search experience for consumers, and the influence of a display ad is more nuanced than that of a search ad. Display is most appropriately seen as a touch point that stimulates consumer interest along the path to eventual conversion and purchase. Rather than clicking and immediately buying, ad viewers waited, visiting website sometime later, and perhaps via another channel or device, to make a purchase. Mind you that this holds for paid search ads as well; it takes 3.5 days on average for a customer to convert after an initial click on a paid search link.
Optimizing for clicks is a proven and effective approach for search advertising. For Display, optimizing for clicks to drive conversions will not get you far; if you’re optimizing your campaign for clicks, there’s a good chance you’re actually anti-optimizing for sales. And although clicks may work for some banner campaigns with limited reach, looking at clicks will certainly keep you from seeing the value in video. Which should not be super surprising, as videos are meant to be viewed; last time I checked Watch Time still was YouTube’s video KPI and not Click Time, or is it?
So, instead of clicks, you should optimize toward your ultimate objective—the campaign conversion rate—rather than toward the click-through rate and focus on developing the necessary systems and skills to understand the true impact of your ad investments throughout your customer’s path to purchase. Are you a Google Analytics user? Then a great example of a ‘necessary system’ is this ever-in-beta feature called “GDN Impression in Multi-Channel Funnels Reporting”. The Analytics blog post nicely aligns with the story above:
Every customer journey is different — a customer may see your display or video ads, receive an email, and then click through to your site from a search ad or organic search listing. Often, viewing display ads can attract your clients’ interest in your product and brand even if no click occurs. Traditionally, measurement technology separated out impressions or “view throughs” from clicks, but this separation missed out on valuable data on the impact of display advertising.
Thanks to our integration with the Google Display Network (GDN), Google Analytics can now break down the separation between clicks and impressions and give a more complete view of the customer journey. When a user views display ads on the GDN, or video ads on YouTube, and later visits your website and converts, these interactions with your brand can now be captured in Google Analytics Multi-Channel Funnels reporting.
Quantcast in the article mentioned above lists the following advice for getting the most out of display media:
How to Optimize for Conversions
If, like most advertisers, your objective is conversions rather than clicks, you can optimize your display campaign to capture them. Here are four steps to get you started:
Posted in Attribution in Online Advertising, Attribution Model, Attribution Modeling, Conversion Rate Optimization, Last Click Attribution, Marketing Attribution Tagged with: atttribution model google analytics, display attribution, gdn impression reporting, multi channel funnels, quantcast, video for performance, view through conversions
The last-click attribution model gives 100 percent of the credit for the sale to the last channel a customer clicked on. That model of reality would be fine if there was only one channel and one touchpoint, but with more than 50 percent of customer interactions involving multiple channels, the last-click attribution model does not do justice to key channels that drive conversions. Besides that, does it actually matter to you if your customers see a search or display ad first or last before they buy? Because you care about revenue, less about the order in which ads are served. The Internet is an open and chaotic place, so even if you could know the ideal order in which ads are served, you can do very little to force-serve ads in that way. So, let’s spend your time in a more actionable way; another way to approach attribution is to voice your hypothesis of what a new device or medium would do to your revenue.
So, what do hypothesis-driven marketers do? They run what scientists call a controlled experiment. You add a medium or device to the mix and do your best to measure lift in revenue from adding that over your baseline marketing mix. You create an incremental revenue model, which looks at customer conversion based on the treatment they received. You can randomly select people to be put into two different groups. A percent of people will be put into an A group, and a different percent of people will be put into a B group. So this might be 90% of your users, and then a randomly selected 10% will be put into a B group, which is what we’ll call the control group. Sounds a lot like the biology lab experiments back in school, right?
Incrementality testing is done by separating a small subsection of customers who don’t receive a certain marketing message and then measuring whether there is a difference in the customers who received the messages (the exposed group) and those who didn’t (the control, or holdback, group). You compare the two conversion rates between the groups that was retargeted and the group that wasn’t and see the actual lift. Imagine that your control group had a 6% conversion rate, and your exposed group 8%, which is a 33% conversion lift.The way that you set up and optimize a campaign to deliver optimal incrementality is different than if you were optimizing a campaign just for the lowest CPC or the lowest CPA. No matter what the CPC is, CPA, or anything else, this is the net impact of your medium or device new to your mix.
Here, AdRoll explains the method:
You you have experience with incremental attribution and running controlled experiments? We look forward to your comments below!
Further reading: uplift modeling article on Wikipedia
Posted in Attribution in Online Advertising, Attribution Model, Attribution Modeling, Controlled Experiments, Conversion Tracking, Last Click Attribution, Marketing Attribution Tagged with: click order attribution, controlled experiment, conversion lift, incremental attribution, incrementality testing
Looking at the blogs, articles, and conferences, attribution is a hot topic for today’s marketers. A quick search for the topic on your favorite search engine delivers lots of content and technologies devoted to the subject. But is attribution actually a problem for you? Is it something that currently keeps you from your goals and should you be investing your analytics dollars in understanding?
The key to understanding if you might have an attribution problem is understanding the way that customers buy your products and services. Are the customer journeys simple, short and does your conversion event happen in that single visit? Or are they typically quite complex? If you look at the paths to purchase and you see that the journeys are quite short en “one touch” paths to purchase, i.e. customers either bought on their first visit or they didn’t, then you might not have an attribution problem at all. When the majority of conversions happen on a single visit then it doesn’t matter which model you use, their output will be the same!
However, be careful in looking at your current data set only, because it might be that you have data problems because you’re not tracking journeys across different devices and touch points, meaning that you are not getting a complete picture. In case you are currently using paid search and looking to deploy video campaigns, then this medium will probably not deliver the same results on the basis of the attribution model that worked well in the past for your paid search campaigns. From the data you might quickly conclude that ‘video does not work’, where in face your attribution model ‘is not longer appropriate’. In other words, you might develop an attribution model.
This is not to say that you need to cherry-pick attribution models based on the medium – it means that you select the attribution model that best fits the path to purchase and maximizes business results. Your attribution model is not ‘fixed’, and it should not be, as the path to purchase and the channel mix changes over time to drive more business.
When was the last time when you optimized your attribution model? We look forward to your comment!
Posted in Attribution in Online Advertising, Attribution Model, Conversion Tracking, Last Click Attribution, Marketing Attribution Tagged with: attribution model, attribution modeling, attribution problem, last click attribution
Are you using many channels to drive sales, but do you allow one channel to claim all credit to the conversions? You might just be making marketing’s fundamental attribution error.
In social psychology, the fundamental attribution error, also known as the correspondence bias or attribution effect, is the tendency for people to place an undue emphasis on internal characteristics – personality – to explain someone else’s behavior in a given situation rather than considering the situation’s external factors.
In marketing, the fundamental attribution error, also known as the desktop computer bias, is the tendency for online marketeers to place an undue emphasis on the last-click to explain a given sale conversion being recorded, rather than considering the conversion’s external factors, such as banner, social and mobile campaigns.
This suggests smartphones are more a browse or research platform rather than a buy platform as most sites in this research are by now mobile optimized sites.
90% of people use multiple screens sequentially to accomplish a task over time, and 90% of marketeers with a desktop computer bias kill their mobile reach by bidding down ruthlesslessly, as they just layer the old-world model of desktop performance on mobile. Do you bid 50% lower for mobile because the mobile conversion rate is 50% lower? Are your ads shining bright on the bottom of mobile search pages?
You would agree that smartphone experiences should be personalised to show this different form of usage, but how to measure all this? See what you can do with the experience of retailer ModCloth (PDF):
“Instead of driving sales, we’re focused on generating registrations via mobile ads. Once a visitor is registered, we can leverage channels like email to convert her to a paying customer,” says Post.
A lot of other interactions are also happening on ModCloth’s mobile-friendly site: adding items to the shopping cart, social sharing, browsing numerous pages… All of these are valuable Mobile Conversions that ModCloth is tracking to better understand the real effectiveness of mobile.”
Are you making marketing’s fundamental attribution error? What do you do and measure with mobile? Leave a note in the comments!
Posted in Attribution in Online Advertising, Attribution Model, Conversion Tracking, Last Click Attribution, Marketing Attribution Tagged with: attribution model, attribution online advertising, fundamental attribution error, marketing attribution