Don’t Settle for Last-Touch Attribution in Marketing
Last month, I talked about factors marketers should consider for attribution rules. Here, I would like to get a little deeper and discuss last-touch attribution, as just talking about contributing factors won’t get you anywhere. As in all data-related subjects, the devil hides in the details. How to collect the data, what to consider, how to manipulate clean and dirty data, and in what order one must execute different steps.
I wonder sometimes why last-touch attribution is such a popular industry, with all of the flaws embedded in that methodology. Without even getting into geeky programming details, let’s think about the limitation of last-touch attribution, in a logical sense.
First off, by giving all of the credit for a conversion to “one” medium of the last touch, you would be ignoring all of the previous efforts done by your own company. If you are the lucky channel manager of the last touch, you wouldn’t mind that at all. But is that fair? C-level executives should not accept such flaws in the name of efficiency or programming convenience.
Why You Shouldn't Settle for Last-Touch Attribution
Let’s use my own experience as a buyer to illustrate a typical customer journey in a multichannel marketing environment. Like any man who shaves daily, I’ve always felt that most quality brand blades were way overpriced. And I found it quite inconvenient that I had to visit a physical store to buy them, when I knew that I would need new blades at a regular interval. All of that changed when a few blade delivery services popped up in that lucrative men’s grooming market a few years ago.
I was one of the early adopters who signed on with one of the programs. But after cutting my face a few times with defective blades, I just canceled the delivery service, and went back to my favorite brand of my adult life, knowing that it would cost more. I considered that to be an affordable luxury.
Then one day, I saw an ad on Facebook, that my favorite brand now offers home delivery service, at a significantly lower price point in comparison to store purchases. Call that my first touch before conversion (to the newly offered service). But I didn’t sign up for it at that time, even though I clicked-through to the landing page of its website. I was probably on my mobile phone, and I also wanted to examine options regarding types of blades and delivery intervals further when I had more time.
That means, I visited the site multiple times before I committed to the subscription model. I remember using Google to get to the site at least once; and later, I hit on a bookmark with its URL a few more times. Let’s say that Touch No. 2 would be labeled as “Organic Search,” and touches No. 3 and No. 4 would be considered “Direct-to-Site.”
If you employ last-touch attribution, then Facebook and organic search would get zero credit for the transaction here. That type of false intel may lead to a budget cut for the wrong channel, for sure. But as a consumer, I “know” that it was the Facebook where I first learned about the new service from the brand.
Imagine if you, as a marketer, had a toggle switch between Last Touch and First Touch rules. When thousands, if not millions, of touch data points are aggregated in an attribution report, even a simple concept, such as “the most important acquisition channel,” will have a different look depending on the attribution rules. In one, Social Media may look like the most effective channel. In another, Organic Search may take the cake. The important lesson is that one should never settle for last-touch attribution, just because that is how it’s been done within the organization (or by the analytics vendors).
There Are a Few More Attribution Methods
The Last and First Touch rules are the easy ones — if you have access to all touch data on an individual level (because you’d have to line touchpoints up for each buyer, in sequence). As I briefly introduced last month, there are a few more methods to consider. Let’s dig a little deeper this time:
- Last Touch: Although there could have been many touchpoints before conversion, this method would just give all of the credit to the last one. As flawed as it may be, there are some merits. For one, last touch would be the one that is directly attributable (i.e., connected) to the transaction (and the session that led to it) without any assumptions or rules. I suspect that the simplicity of it all is the main reason for its popularity.
- First Touch: This would be useful for the study of acquisition sources. Timeline is an important consideration here, as effectiveness of channel or offer may decay at different rates, depending on product and channels in question. A consumer may have researched for a washing machine four months ago. And saw a newspaper insert about it three weeks ago. And then got an email about it a week ago. How far back can we go with this? A catalog that was mailed six months ago? Maybe, as we are dealing with a big-ticket item here. And are we sure that we have any digital touch data that go back that far? Let’s not forget that the word “Big Data” was coined to describe click-level data to begin with.
- Double Credit: If a person was exposed to and engaged in multiple channels before the purchase, why not credit all involved channels? Overkill? Maybe. But we use this type of reporting method when dealing with store-level reports. There is no law that one customer can visit only one store. If one visits multiple stores, why not count that person multiple times for store-level reports? So, with the same reasoning, if a transaction is attributable to multiple channels, then count the transaction multiple times for the channel report. Each channel manager would be able to examine the effectiveness of her channel in an isolation mode (well, sort of).
- Equal Credit: This would be the opposite of Double Credit. If there are multiple channels that are attributable to a transaction, create a discount factor for each channel. If one is exposed to four channels (identified via various tags and tracking mechanisms), each would get ¼ of the transaction credit. When such discounted numbers are aggregated (instead of transactions, as a whole number), there will be no double-counting in the end (i.e., the total would add up to a known number of transactions).
- Proportional Credit: Some channel managers may think that even Equal Split is not a fair methodology. What if there were eight emails, two organic searches, three paid searches and a link on a Facebook page that was clicked once? Shouldn’t we give more weight to the email channel for multiple exposures? One simple way to compromise (I chose this word carefully) in a situation like this would be to create a factor based on the number of total touches for each channel, divided by the total number of touches before conversion.
- Weighted Value: An organization may have time-tested — or politically prevailing — attribution percentages for each employed channel. I would not even argue why one would boldly put down 50 percent for direct marketing, or 35 percent for organic search. Like I said last month, it is best for analysts to stay away from politics. Or should we?
- Modeled Weighted Value: Modeling is, of course, a mathematical way to derive factors and scores, considering multiple variables at once. It would assign a weighted factor to each channel based on empirical data, so one might argue that it is the most unbiased and neutral method. The only downside of the modeling is that it would require statistically trained analysts, and that spells extra cost for many organizations. In any case, if an organization is committed to it, there are multiple modeling methods (such as the Shapley Value Method, based on cooperative game theory — to name one) to assign proper weight to each channel.
I must point out that no one method would paint the whole picture. Choosing a “right” attribution method in an organization with vastly different interests among teams is more about “finding the least wrong answer” for all involved parties. And that may be more like Tony Soprano mediating turf disputes among his Capos than sheer mathematics spitting out answers. That means the logically sound answer may not void all of the arguments. When it comes to protecting one’s job, there won’t be enough “logical” answers as to why one must give credit for the sale to someone else.
While all of this has much to do with executive decisions, people who sit between an ample amount of data and decision-makers must consider all possible options. So, having multiple methods of attribution will help the situation. For one, it is definitely better than just following the Last Touch.
Start With Proper Data Collection
In any case, none of these attribution methods will mean anything, if we don’t have any decent data to play with. Touch data starts with those little pixels on web pages in the digital world. Pages must be carefully tagged, and if you want to find out “what worked,” then, well, you must put in tracking requests properly for all channels.
A simple example. In a UTM tag, we see Medium coded with values such as Paid Social. A good start. Then we go to Source, we would see entries like Facebook, Instagram, Twitter, Pinterest, etc. So far, so good. But the goal is to figure out how much one must spend on “paid” social media. Without differentiating (1) Company’s own social media page, (2) Paid ads on social media sites, and (3) Referrals by users on social media (on their Facebook Wall, for example), we won’t be able to figure out the value of “Paid Social.” That means, all of the differentiation must be done at the time of data collection.
And while at it, please keep the data consistent, too. I’ve seen at least 10 different ways to say Facebook, start with “fb.”
Further, let’s not stop at traditional digital tags, either. There are too many attribution projects that completely block out offline efforts, like direct mail. If we need to understand where the marketing dollars must go, why settle with one type of tracking mechanism? Any old marketer would know that there is a master mail file behind every direct mailing campaign. With all those pieces of PII in it, we can convert them into yet another type of touch data — easily.
Yes, collecting such touch data for general media won’t be easy; but that doesn’t mean that we keep the wall up between online and offline worlds indefinitely. Let’s start with all of the known contact lists, online or offline.
Attribution Should Be Done in Multiple Steps
Attribution is difficult enough when we try to assign credit to “1” transaction, when there could be multiple touchpoints before the conversion. Now let’s go one step further, and try to call a buyer a “Social Media” responder, when we “know” that she must have been exposed to the brand at least 20 times through multiple media channels including Facebook, Instagram, paid search through Google, organic search through some default search engine on a phone, a series of banner ads on various websites, campaign emails and even a postcard. Now imagine she purchased multiple times from the brand — each time as a result of a different series of inbound and outbound exposures. What is she really? Just a buyer from Facebook?
We often get requests to produce customer value — present and future — by each channel. To do that, we should be able to assign a person to a channel. But must we? Why not apply the attribution options for transaction to buyers, as I listed in this article?
That means we must think about attribution in steps. In terms of programming, it may not exactly be like that, but for us to determine the optimal way to assign channels to an individual, we need to think about it in steps.
Now, if you are just settling for last-touch attribution, you may save some headaches that come with all of these attribution methods. But I hope that I intrigued you enough that you won’t settle so easily.
Stephen H. Yu is a world-class database marketer. He has a proven track record in comprehensive strategic planning and tactical execution, effectively bridging the gap between the marketing and technology world with a balanced view obtained from more than 30 years of experience in best practices of database marketing. Currently, Yu is principal and chief product officer at BuyerGenomics. Previously, Yu was the head of analytics and insights at eClerx, and VP, Data Strategy & Analytics at Infogroup. Prior to that, he was the founding CTO of I-Behavior Inc., which pioneered the use of SKU-level behavioral data. “As a long-time data player with plenty of battle experiences, I would like to share my thoughts and knowledge that I obtained from being a bridge person between the marketing world and the technology world. In the end, data and analytics are just tools for decision-makers; let’s think about what we should be (or shouldn’t be) doing with them first. And the tools must be wielded properly to meet the goals, so let me share some useful tricks in database design, data refinement process and analytics.” Reach him at email@example.com.