
At the Search Insider Summit yesterday, Avinash Kaushik said much of thinking around attribution is "complete and utter garbage."
"Most marketers suck at attribution. They get all worked up and run around crying like little girls"
Attribution is one of the thorniest problems we've dealt with on the client side in the last year. Avinash had some clarifying views, which I'll share, and then I'll tie in attribution with some of the things we're seeing on the qualitative side of ad engagement:
First of All, Do You Have an Attribution Problem?
Avinash recommends segmenting your traffic by number of site visits required before conversion to see if you even have an attribution problem. If 90% of your conversions comes from one or two visits, you probably don't have a urgent problem. However, if a significant percentage of sales results from multiple visits, further investigation is required. Next, segment by referrer source and again see how many visits are required to convert. This allows you to isolate if you have a problem and where it might be.
Okay, I Have a Problem. What Do I Do About It?
Now you have figure out which of the many models is the appropriate one to use. Kaushik quickly dispatched the first and last click models - "Crap..and crap" - as well as the "Berkeley Kumbaya" model, where everything is fairly distributed in the interests of "universal harmony" and finally, the MUC model - percentages based on "Made Up Crap". The one that gained some level of approval was a modified decay model, where last clicks are weighted more heavily and earlier touches received less attribution value.
The Cognitive Engagement Attribution Model
As Avinash was talking, I jotted a note down on my notepad - "Attribution Model based on engagement with ads". In a post last week, I talked about the
importance of aligned intent. The fact is, there is a switch in our level of cognitive engagement somewhere along the line that can have many contributing factors. Further, even after this cognitive switch leads to higher engagement, not all those ad exposures are equally influential. If we understand more about the nature of influence of these ads, perhaps we can start to pick apart the attribution model.
Let me explain in a couple of scenarios:
Scenario One: Aligned Intent
If you already have an objective in mind, attribution is an easier nut to crack. Let's say that you work in a clinic and you suddenly find you're running short on face masks. Your intent is defined, you search by the product name, check a few sites out and order from the site with the cheapest prices and fastest delivery time. As Avinash said, you probably don't have much of an attribution problem. If the buyer was price conscious, you might have a couple of site visits and the referrer would likely be a search engine in each case.
Scenario Two: Aligned Intent, Longer Sale Process
Okay, let's get a little more complex. In this case, you know you need to replace your aging photocopier but some research is required. In the
BuyerSphere Project I referred to this as a modified repeat purchase, things you buy occasionally but because of changing specs, they come with an education curve each time. Your intent is already defined, but the path is longer. You'll be eliminating risk, so you will likely have more online touch points. In our research, we've seen that there is generally early search activity, then a transference to vertical engines, review sites or directories to help refine options (Business.com's
buying guides provide a classic example) and probably several vendor site visits. Search will often be used navigationally to stitch all this activity together. In a more complex scenario like this (typical in B2B and longer sales cycle purchases - i.e. cars, travel, real estate) how do you slice up the attribution pie?
Perhaps we can gain some further insight by looking at how each of these visits mentally influences the prospect.
Early in the process, the prospect has intent (to find a new photocopier) but the decision criteria framework hasn't been established yet, a key requirement for risk mitigation. This is what leads to the early search activity, as the prospect plots out the "decision landscape". They are identifying the factors required in the eventual purchase decision. This sets up the entire subsequent chain of online activity to follow. Not only that, but this is typically where the first consideration set is compiled. A vendor website touch is common at this point.
As the criteria is set, consideration and evaluation begins to take place. The objective here is to reduce the number of options to a manageable number, typically less than 5. This is where vertical directories, reviews and repeat visits to the vendor site are essential in deciding which alternatives make the list and which ones don't. At this point, the prospect's interests become more focused and specific - i.e. exact pricing information and detailed feature comparisons. They're rationalizing the purchase decision by weighting difference criteria.
Finally, a decision may be made in your product's favour. The conversion event finally happens. Often the path that will lead to this will be a navigational search resulting in a site visit.
In this typical scenario, we have a series of site visits originating from different referrers. Here is where the attribution problem becomes challenging. Using the decay model, the assumption would be that the last event, the decision to favor the brand, is the most important and would be given the most value. Moving up the chain, each event would become less important, until, at the top - mapping the decision landscape - this online activity would receive almost no attribution value.
This attribution model is based on a causal chain, with the values based on percentage of likelihood on going through each point and reaching the end of the chain. Say each touchpoint represented a letter in the chain below
A - B - C - D - E - F
The goal is to take 20 candidates for consideration at A point and reduce it to one candidate at F point. Now, using a simple formula and some assumptions, one could construct the following attribution model:
At A - you're one of 20 candidates, so you have a 1 in 20 or 5% chance of a successful conversion
At B - the list has been narrowed to 12 candidates, so you have a 1 in 12 or 8.3% chance of successful conversion
At C - you're 1 in 6 candidates, so you have a 16.6% chance of successful conversions
At D - the list has been narrowed to 4 candidates, so you have a 25% chance of successful conversion
At E - there are only 2 candidates, so you have a 50% chance of conversion
At F - you're the successful candidate.
This is the thinking behind the decay model, and, incidentally, pipeline weighting in most CRM and sales automation tools. The further you get down the funnel, the greater your odds, so the higher the attribution. In a rational world, this makes sense.
But, as I explained in The BuyerSphere Project, the funnel analogy doesn't hold up very well in the real world. Not all prospective candidates have equal chances and not all stages are linear. Let me use an analogy that Avinash used in his keynote:
Avinash said that first click attribution is like giving his first girlfriend credit for his current marriage. I want to stick to the same analogy, but I have a differing view. If you asked any married person what one of the most important moments in their life was, I'd bet the moment they first met their wife would be high on the list. Us married people consider that moment to be one of those pivotal life changing events. If you followed the causal chain that lead to your wedding day, you'd put a high degree of value on that moment, because if it hadn't happened, none of the following would have happened either. Similarly, the moment you first introduced your intended to your parents, or when you had the discussion about whether you both wanted kids would also be pivotal moments. Humans following convoluted decision paths where not all stages or factors are weighed equally. This is true in finding your life partner or in making complex purchase decisions. If we were to apply a decay model to the causal chain that led to Avinash's marriage, his decision to drive to the church the morning of his wedding would get 20 times the value of the moment he first met his wife, 12 times the value of that first introduction to his family and possibly 3 times the value of the day he proposed to her. Somehow, that just doesn't seem right when we think of it, factoring in emotions. And emotions are an essential part of any decision.
Complex decisions are journeys down paths, and along that path we'll meet points of resistance and other potential paths branching off. I'm not sure you can say the decisions we make near the end of the path are more important than the decisions we make at the beginning. I think it's a complex and nuanced scenario.
Final Scenario: Complex Purchase, No Intent
Let me stick to the Avinash wedding analogy. In the previous example, I assumed that intent was already planted. Even before he met his wife, Avinash intended to get married someday. But what if that was never Avinash's intention? What if somewhere along the line, something had to flick some switch in Avinash's brain and take him from a life plan of avowed bachelorhood to that of a man looking for a wife. This is a powerful and fundamental change, because whatever caused the switch to flip had a dramatic impact on every event that followed. The planting of intent changed the mental framework of Avinash Kaushik from a man who filtered out all talk of marriage to one who was openly receptive and actively seeking it. As I said in a previous post, that planting of intent significantly alters the nature of our engagement with advertising.
This brings me to my friend Gian Fulgoni's point, made
in a comment last week:
There’s also another even more important point that we need to consider: brand building. That needs to occur even when the consumer isn’t foraging for information in support of an impending buying decision. Otherwise the value of an individual brand name isn’t going to be as meaningful to the consumer when he / she is in the shopping / buying mode. CPG manufacturers know this well...They understand that (coupons and flyers) need to be supplemented with “branding” advertising that they run themselves because they need to make sure that their brand value has been firmly established in the mind of consumers before they compare prices across brands at the shopping / buying stage.
Awareness advertising is still required to "flip" our mental switches and make intent advertising possible. So, what does this mean for attribution models? Again, I would argue that this flipping of the mental switch, the aligning of intent, is a critical step in the purchase path. Indeed, it creates the purchase path, and so deserves a much greater value than it would typically get in most attribution models. This switch flipping doesn't happen on search, it precedes it, often through the exposure of a display ad or some other awareness format. Using the Avinash Wedding Attribution Model, doesn't Avinash's decision that at some point in his life, he wants to get married deserve at least as much credit as his decision to drive to the church on the day of his wedding?
So, What Does This Mean?
If you accept that the weighting of influential moments in the mental decision path may be a viable approach to attribution modeling, it throws an immensely complex problem at you. How do you possibly understand these weightings? Again, like so many things that hinge on human behavior, I believe the answer lies first in qualitative exploration, then quantitative verification. For any given product, or even category of products, you start to see commonalities in purchase behaviors. Most of your prospects will have to make the same decisions and encounter the same forks in the purchase path. By gaining more insight into what actually influences those purchase decisions, you can then begin to formulate an attribution model that's based on more than guesswork or "Made Up Crap".