Mar 132014

Video: What's the Deal With Attribution Anyway?

When 70 - 75% of touches on average come through just one channel, how much impact can tweaking your attribution model have on allocating marketing dollars?

Attribution modeling, even with our own proprietary system, has some real issues associated with it such as coming to meaningful and actionable changes and getting institutional buy-in to change the current model.

Watch as we explore this topic.

George Michie: Hi. I'm George Michie. Today, I want to talk a little bit about attribution. Attribution is one of the really hot topics in marketing. Everybody is interested in figuring out what marketing channels are driving what behaviors. Everybody is afraid that their view of what's working for them or what isn't working for them is skewed by the fact that people are interacting with multiple marketing channels, and you may only be giving credit to the last one for conversion events on your site. So people are worried that they're spending money on the wrong stuff and giving too much credit to some channels and not enough to another.

We've been in the attribution business since 2009. We built our own proprietary system for doing this. The way we've talked to clients about attribution has changed, based on our experience.

One of the things that we have found is having conversations with folks about different complex statistical modeling approaches to get a better answer to attribution is a lot of fun, but it throws out the hardest problem first. What we're finding with a lot of folks, in the e-commerce space particularly, is there's much less interaction between advertising channels than you might think. For our average advertiser something like 70% to 75% of the touches come through only 1 channel, which means when you talk about attribution, you're really just talking about that other 20% to 30% that isn't coming through on just 1 channel. So changing the attribution for that smaller chunk doesn't change your view of what's working by as much as a lot of people expect it to.

Moreover, there's a problem, we find, with getting institutional will to actually change how you view attribution. Your numbers and your goals for the year are tied to your current way of looking at the numbers. If you change the way you look at the numbers in a way that materially affects the result, all of a sudden the year-over-year metrics start looking different. They either look great, because you're getting credit for more orders than you used to, or it looks awful and the boss screams at you. So there are all kinds of trouble getting folks to have the will to actually make a change.

There are also limits to what you can change, and this is the last point that I want to make today. If you found that organic search traffic is either driving more business or less business than you currently think it is, aren't you still going to optimize your site to get as much as you can? You know, it doesn't actually tell you to do anything differently. Email is the same way. Email is so cheap, that whether it's over credited or under credited, you're still going to send emails, probably at the same rate you're currently doing. Display advertising, yes, display can be under credited if you're only looking at the last touch. But you can already do public service announcement control tests with display advertising to find out what the incremental lift is.

So what levers do you actually have to act differently, based on attribution? You may find that these levers are fewer than you think.

RKG has been in the attribution business for a while. We like to help our clients understand this and remove the impediment that attribution can often be to acting. People are afraid to make changes because they are afraid they're not seeing the whole picture. We can help them understand what that picture might look like under different scenarios.

So, I hope you enjoyed this and will watch more of our video series. Thanks.


4 Responses to "Video: What's the Deal With Attribution Anyway?"
Dan Merton says:
Hi George, You mentioned using PSAs to test attribution. Could you expand on that, or can you point to past post that explains that? Thanks, Dan
Hi Dan, thanks for your comment. DSPs can be used to test creative by showing different versions of the ad against a control. In similar fashion, using a public service announcement as the ad creative (totally unrelated to your brand) you can get a very clear picture of the fraction of view through display conversions that represent organic return visits, unrelated to the ad. The notion is if you served an ad to every browser everywhere in the world one day, you would not ascribe every order on your website to have been caused by that ad that day. Display should get credit for the lift over the baseline provided by the PSA, but no more than that.
Casey Carey says:
George - I agree with your points, and frankly from an ecommerce standpoint, I hold RKG's attribution expertise in very high regard. As you said, you've been doing it a while. A couple observations I would like to share from our attribution journey across 75+ brands. We see multi-touch and multi-channel paths averaging 65% and 45% respectively. In fact, I was looking at some data today for a major consumer electronics ecommerce company and it was 73% and 44% respectively. This could be from a couple different factors: * Look-Back Period: As we are using a statistical model, we open up the window and let the data decide on the appropriate look-back period. * Impact of Impressions: Where possible, we are capturing impression data in addition to clicks including display, video, paid social, email, direct mail, and sometimes affiliates. This provides a more complete and often complex view of the journey. * Cross-Browser and Cross-Device Mapping: More and more customer journeys are occurring across devices (and browsers). Let's face it, most attribution today is device-based (cookies) rather than user-based. We have started mapping and attributing marketing and conversions at the user level. The result is longer, more complex paths with higher conversion rates. I have worked with a handful of companies where their full-funnel results where not significantly different than last-click. After further investigation, I found a classic case of "You are what you measure." They had highly optimized all their marketing strategies and tactics around last-click and on a per channel basis. For these companies, it becomes more of a strategic question about the growth opportunities available in top and mid-funnel marketing, as well as cross-channel optimization rather than comparing methodologies. I completely agree with your comments about the change management aspect of a new attribution approach. It is hard and there is a lot of inertia and organizational history built on the current models. But honestly, that is a poor excuse for not doing something better. Where else are you going to go to improve your marketing more than a couple percentage points? Finally, you highlight an important point in that you have to be willing and able to do something different based on the results. To touch on a couple of your examples: * In the case of paid search lifting organic search, of course you are going to do as much SEO as you can. However, if you understand that specific paid search investments are driving organic, you may make more investment in those terms, as their fully attributed performance is better than you thought. * Yes, you will continue to send emails, but you may gain insights into the interplay and timing between display, paid search, site visits and email that allows you to improve the performance of all through creative, flighting, and other integrated tactics. * And yes, hold out panels are valid in determining the lift of display, the problem is you have no idea which tactics and channels are under-performing on a cross-channel basis. So you end up with a business case to spend more on display, but have no idea where which channels will have lower spend. I've had this conversation with a CFO and it didn't go well. Personally, I think the attribution journey is just starting and things are evolving rapidly. Maybe someday we'll look back and think, "I can't believe we did it that way." Keep up the great work and I look forward to future posts. Casey
Casey, thanks for your commentary and for sharing your experiences at Adometry. You make a number of excellent points, first and foremost that what is true for Direct Response focused retailers may not be true for Auto brands, CPG, and vacation travel. More considered purchases, more brand-building focus leads to more touch points along the path. As I mentioned in an earlier post, which I think you commented on, high-rent statistics can give us more confidence than we should have. In our research with respect to look-back periods for example, we found that the farther you go back the more channel interactions come into play, however looking at the paid search keywords also indicates the problem: the greater the time separation the less connected the early keywords are to the products purchased. Strong likelihood that those early searches were part of a different shopping mission; one which this client lost. We believe there are opportunities to take action, particularly with respect to affiliates, but experience has taught us that few folks pull the trigger, and often when they do pull the trigger, they undo that work when the next manager comes in. Perhaps I'm jaded by that experience! We find there to be far more opportunity to move the needle by doing a better job of core blocking and tackling (executing core programs well, understanding economic fundamentals of the business like lifetime value of customers coming from different channels, etc) than by boiling the ocean of attribution.

Leave A Comment