A wise person said: "To guarantee success, spend 95% of your time defining the problem and 5% of the time solving it."
I believe deeply in that quote. In my life I spend an extraordinary amount of time understanding the problem and attempting to define it clearly. As if by magic, I find that it is then much easier to find the optimal solution (or realize none exists!).
Multi-Channel Attribution is a red hot topic in our industry, and yet it is so poorly understood. I'm convinced that the resulting problems (confusion, FUD, angst, daily prayers, and wasted budget) are due to the lack of a clear framework that can help clearly define the problem.
In this post my hope is share a framework that will help define the problem clearly. Included in the post are recommendations for measurement and data analysis. And as if that was not enough, :), I'll close the post with my thoughts on digital marketing attribution models.
This is going to be a lot of fun. Roll up your sleeves, put a smile on your face, grab a pinch of common sense, a heavy dose of reality and let's go…
Three Types of Multi-Channel Attribution Problems.
A huge amount of confusion and disagreement on this topic exists simply because there is no general consensus about those three words. Multi-Channel Attribution.
So let's try and fix that problem.
There are three types of attribution problems in our non-line world.
Multi-Channel Attribution, Online to Store:
This is the attempt by Marketers and Analysts to try and understand the offline impact (revenue/brand value/butts in seats/phone calls/etc) driven by online marketing and advertising. We'll refer to this quest for doing effective attribution as MCA-O2S.
While I'm using the term Store here, it encompasses sales (or leads or catalog requests) driven to a retail store or company call center, people driven to donate blood via online campaigns, or essentially any offline outcome driven by the online channel.
An example of MCA-O2S is Verizon wanting to know how many in-store offline phone activations are driven by online search advertising, for every online activation that the same search advertising drives.
[In case you were curious... It's 5 new accounts activated offline for every 1 activated online! If you are not calculating the offline impact, and you are not giving your online channel due credit. In this case, it would be 5x less credit! You can see why MCA-O2S is supremely critical for every company on the planet. You can also download the slides from a HP online-to-store experiment and how they learned that every $1 spent on online advertising delivered $5.3 in offline store sales. Or watch a video of how Quaker Oats boosted sales by 9% in store via YouTube videos.]
Here's the Post-It on which I'd sketched MCA-O2S in planning this post. The red dots represent activity we would like to ensure we are measuring to 1) ensure we understand behavior, and 2) deliver insights that will influence our marketing and advertising…
I spend a lot of time with CEOs and CMOs and when they talk about multi-channel attribution, they're invariably talking about MCA-O2S. Yet when most of my digital peers talk about this topic, they're not talking about MCA-O2S. You can imagine why things might get a little confusing.
So when you meet a CEO and they use say "Help me solve the amazing multi-channel attribution problem", you say: "which type of MCA are you interested in?" Clarity will help foster a valuable conversation.
Almost all current, hot and heavy, literature on the topic of attribution modeling does not cover MCA-O2S. That's because when it comes to MCA-O2S your only bffs are a set of 16 strategies I've outlined over two posts (links immediately below) or the fantastic world of controlled experiments (as in the Verizon case above). So less automated algorithms "distributing credit" and more thoughtful deliberative discreet measurement strategies that inform strategic decisions.
MCA-O2S. It's mandatory. Attribution is driven by experiments. And when you win, you win huge!
Multi-Channel Attribution, Across Multiple Screens:
Senior leaders, especially in larger companies, have started to refer to this when they use the magical words multi-channel attribution.
With the massive adoption of mobile phones and tablets we are all increasingly "four screen" people (TV, desktop, tablets, smart phones). That has directly translated into a more complex fragmented influence landscape (drives the "old timers" bananas). That in turn has translated into many senior leaders deeply desiring, as they put it, "multi-channel attribution." What they really mean is MCA-AMS.
What they really really want is to understand how individuals experience a company or government's digital existence across multiple devices, what media (advertising and marketing) they are being exposed to, and what outcomes (conversions!) are happening as a result.
An example of MCA-AMS is the ability to understand that a search I did on my tablet computer while watching a television commercial resulted in a click on a paid search ad to a camera site which logged into my memory which later caused me to read reviews of the camera on my Nexus S while stuck in traffic and that finally caused a sale for Sony when I got home and happened to be on my laptop.
Attribution in this case is the quest to apportion credit across the TV commercial, tablet paid search ad, reviews read on on the mobile phone for a "direct" conversion on the PC. Amazing, right?
Here's my sketch on MCA-AMS and the raw complexity of the customer experience that we are trying to understand… the red dots indicate what we're trying to measure and understand the impact of…
The primary challenge is that as we switch devices it is increasingly difficult to keep track of the same person as they interface with our digital existence (and are exposed to online and offline marketing and advertising). Actually, I should not say increasingly difficult, I should say almost impossible (cookies, uuids, privacy, government, et al).
Perhaps the only exceptions to the "its almost impossible" scenario would be companies that service customers who are mostly logged in (think Amazon, NY Times) across all four screens all the time. Such companies usually also own massive data warehouses where they have an ability to periodically do cannonballs into the data and identify correlations in consumption and purchase patterns. Often, though not always, they can also tease out causations between devices used during outcomes (five-second segmentation in say Google Analytics) and their media plans while focusing on customer analysis (not visitors, not cookies, not uuids, customers).
Even then it is hard, very hard. And for the rest of us this will remain a complex, and I'm sorry to be so real, unsolvable challenge. At least for now.
Some ideas from the two multi-channel blog posts above can help with MCA-AMS. I've leveraged controlled experiments to get very good "kinda sorta understanding" of reality.
I believe that real solutions will come from the evolution of cookies, updating privacy policies, government decisions and evolving user habits. All that first, then our ability to have nonline data.
Because of all of the above you can see why attribution models don't even enter the picture with MCA-AMS. But when you meet executives and they say "help us with our multi-channel attribution problem", most definitely ask the clarifying question: "do you mean MCA-O2S or MCA-AMS?"
MCA-AMS. Complex, hard challenge. Not a huge problem yet for most, but heading in that direction.
Multi-Channel Attribution, Across Digital Channels:
Almost all of the time when people in our ecosystem (unlike CEOs, CMOs) talk about Multi-Channel Attribution, this is the one they are referring to.
MCA-ADC is the effort to understand which digital marketing channels (Social, Display, YouTube, Referral, Email, Search, others) contributed to a particular conversion (or multiple conversions).
At the moment all web analytics tools, like SiteCatalyst, WebTrends, Google Analytics, CoreMetrics, and others, by default attribute a conversion to the channel immediately prior to the conversion. This is also known as last click attribution.
With MCA-ADC you are trying to go beyond the last click and get this, complete, picture of all marketing activity prior to the conversion (in this case from Google Analytics):
For this website, 767 conversions came from people who visited the site in the above precise order starting with social then a direct visit then an organic search then a referral click-through and finally one last direct visit which lead to the conversions.
The attribution bit here is the burning desire inside all digital marketers to figure out how to dole out credit for the above conversions. Should Direct get 50%? How about Social? 100%? What about Organic? 2%? But let's put that delightful thought on the back burner for just a minute while we understand a critical, often hidden, nuance. [Analysis Ninjas are magnificent at understanding nuance!]
When people talk about MCA-ADC they are still just talking about one device. Because in very close to 0% of the cases do any of these analytics tool have an idea about the behavior of one homo sapien across multiple screens (AMS).
So what you are seeing above are all the conversions that can be tied to multiple visits by a unique browser (notice I did not say person) to your website/digital existence. BTW it is fantastic that GA does this because most other tools don't even show you this.
Say, the Organic Search above had happened on a mobile phone… regardless of the digital analytics tool used, to most websites today that visit would be invisible in the above chain (cookies!). #omg
Hence it is important to separate out MCA-AMS (across multiple screens) from MCA-ADC (across digital channels) – at least for now, until the cookies, ids, privacy policies, government guidance and user habits problem is solved.
When it comes to measuring MCA-AMS you'll use the guidance provided in the above section. For MCA-ADC you'll use a different set of reports (multi-channel funnels ) and attribution models.
I'm sure you are already familiar with nuance number two when it comes to MCA-ADC. A blind-spot if you will.
The above picture does not capture what the impact of this behavior was on your offline existence (O2S). Web Analytics tools are not awesome at that. Ok, they stink at it.
So it is possible that an additional 3,835 people went and made purchases in your stores or via your phone channel (taking the Verizon numbers from above). That would also be invisible from the above report. None of the channels above, whether glorious social, beloved direct, magnificent search, sweet referral, would ever get "credit." Unless you are willing to use the methodologies outlined in the MCA-O2S section above.
When you talk about MCA-ADC, ensure that you are aware and communicate to your leadership, that you are not reporting on MCA-O2S (online to store) and it is extremely unlikely to be reporting the impact of MCA-AMS.
Here's one last Post-It sketch. The red dots are what you are likely measuring when you attempt MCA-ADC…
And if I wanted to be pedantic I would say it is really MCA-ADCFOD. Multi-channel attribution across digital channels for one device.
Now it is true that with sufficient analytical skills, time, patience, and God's direct blessing to you, it might be possible to do complete multi-channel attribution analysis where the multi-channel includes multiple online ad channels, behavior of the person across devices and the impact online and offline. Sadly, that is incredibly hard to do as a whole. And when I say incredibly hard, I mean almost impossible. And when I say almost impossible, I mean only attempt that after you know you've fixed all other problems with your advertising, your online and offline existence and your people. All three.
I know that sounds like a bummer, but a dose of reality is particularly needed in this discussion. There are simply too many fake promises being made by vendors, consultants, tweeters, gurus and fairies. That is unhelpful to the entire ecosystem.
To close this section…
Next time you hear someone utter the words multi-channel attribution, the single greatest gift you can give yourself is to ask in your sweetest possible voice: "Are you referring to MCA-O2S, MCA-AMS or MCA-ADC?"
You'll earn their respect for knowing that there are three types, and you'll be able to put into context what they are asking for and proceed to have a career and business-enhancing discussion.
Multi-Channel Attribution Models.
For MCA-O2S and MCA-AMS, it is a complex undertaking to identify "which advertising/marketing vehicle deserves how much credit." It requires patience and skills. And it requires your execution of multiple of the 16 strategies I've outlined for tracking online impact of offline and offline impact of online. Even more, it requires an ability (people + skills + desire) to execute controlled experiments.
So the question "who deserves how much credit" is tertiary at best.
With MCA-ADC that quest is a little bit easier. We have the multi-channel funnel reports at our disposal. Additionally in some tools we also have an ability to apply attribution models to the behavior you see in the two pictures above in the MCA-ADC section. #sweetness
The most common attribution models bundled into even the simplest web analytics tools are: Last click, first click, and even distribution.
If you are lucky, you have access to a more sophisticated tool which would include: Adjustable, based on mathematical algorithms, time decay model.
If you are among the chosen few, you'll likely have access to a digital analytics tool that allows you to create a customized attribution model.
Each of these models are applied to MCA-ADC (still without benefit of O2C or AMS) and provide you with incrementally better understanding of your digital media spend.
Each of these models comes with its own pros and cons. [If you have my book Web Analytics 2.0 please jump to page 358.] Some of them have more cons and barely any pros. Those should be avoided like the plague.
A couple of them pass the common sense test, and hence will put you in a better place than staying with last click attribution.
But most of what you'll get out of playing with these models is a deep and profound appreciation for how they'll, even in their most shining moment, give you directional guidance how to adjust your media spend (shift dollars/euros/pesos from Search to Display or from Display to Email or… other combinations).
You'll realize (even if you use the greatest customized model created by your most magnificent consultant at a equally magnificent cost to you) that success then will come not from that rough output, but rather from your ability to take that rough output, make changes, observe the impact (over weeks, or months if you are small sized), identify insights and be less wrong over time.
If you happen to be in a larger company, say you spend more than $10 million on digital marketing per year, you'll quickly see, having learned to be less wrong over time, that the question you want to answer with multi-channel attribution modeling is not "who gets how much credit" but rather "how can I optimally balance my digital marketing portfolio."
That will then drive you to seek solace in the arms of the only solution that actually works. The solution that is hard. The solution that requires unique people skills and an undying desire to scale un-imagined heights of glory. Media Mix Models. Executed via persistent controlled experiments.
When you reach that point, fame, fortune and happiness will be yours.
Multi-Channel Attribution: Closing Thoughts
This is a tough challenge. Simply because reality is complicated.
Customer experiences are ever more complex, influence channels intersect a lot more, content consumption is fragmented, the three-step "attract, acquire, retain" model is now broken into 37 different pieces.
So, you don't have a choice. You are going to have to deal with the multi-channel attribution problems, all three of them, if you want your company to have an effective advertising and marketing strategy.
Here's the good news: You don't have to try to boil the ocean in one go. In fact, that might be hazardous to your health if you attempt to do that. Take gradual steps. Increase your sophistication over time.
Here's what I recommend:
1. First clarify what problem you are solving for your management team. O2S or AMS or ADC.
2. Use the appropriate set of solution (see sections above). If MCA-ADC…
3. Get really, really good at understanding your multi-channel funnel reports. They are free. They are awesome. Use the Venn diagram in the Overview report to display reality to your management team. They'll love you, and stop wasting money.
4. Start to experiment with the simple models. You are moving away from last click, you'll abandon first and even very quickly. Spend some love and attention on the time decay attribution model (ideally with several mathematical options to apply).
5. Experiment with changes in your digital portfolio based on your time decay results.
6. Measure outcomes. Go back. Analyze the data. Change some more.
7. As you master that, shift slowly to playing with media mix modeling type controlled experiments.
If at any step you notice diminishing margins of return, go back to the previous step and optimize that one some more until it is truly worth the incremental company investment to take the next step.
If you understand the frameworks, if you internalize the challenges, if you define your company's immediate unique problem clearly, and follow a step wise approach outline above you'll not just do fine. You'll be fantastic.
As always, it's your turn now.
Which multi-channel attribution problem are you solving in your company? Do you distinguish between the three outlined in this post? Is there a fourth one not covered in this post? Which one do you find to be the most challenging? Are you more optimistic that we'll solve AMS (across multiple screens) than I am? Which MCA-ADC attribution models do you swear by? Who's your BFF? Do you have a attribution model that's not covered in this post?
Please share your thoughts, feedback, critique, and brilliant new ideas via comments.