I have seen the Emperor walking naked for too long, and I wish I could be that naive kid in the crowd. I do not believe in marketing “attribution”. Beyond the combined power of algorithms, data, software and professional know-how, the concept is — at its foundation — flawed.
Here’s an attempt at explaining my standpoint, although it’s worth noting that more scientific avenues have already been explored with similar conclusions.
1. A BLACK HOLE
It all started with a beautiful idea. Cross-channel attribution (or “multi-touch attribution”) became a popular concept at the time when web analytics had just completed its journey from IT to the marketing department (circa 2008).
What was it about? Essentially, our ability to assign a specific value to each touchpoint or event — often a paid ad view or click — contributing to a final business outcome or conversion. This would allow us to stop giving undeserved credit to the last or first campaign or touchpoint logged in the chosen system of record.
Increasingly more sophisticated techniques for the integration of owned, paid, and earned media touchpoints within first-party analytics environments have subsequently increased our capabilities, eventually spinning off a software category of its own.
What is not to like about the whole concept? It embodies everything marketers ever wanted to get out of data!
The cross-departmental synergies it creates cannot be ignored either, starting with a translation of marketing outcomes/touchpoints/reach into dollars — a long dreamed bridge between the CMO and CFO camps — and continuing with an engineer-friendly understanding of the marketing process, which is music to the ears of CIOs and marketing technologists. Plus CMOs can now be held accountable, making CEOs happy. In other words, killer fuel for company politics.
But there is of course an ultimate promise: a mastery of the formula results in consistently getting more money out of it than the amount originally invested. Which is truly unbeatable.
Why question it, then?
Well, attribution — even when solely focused on digital channels — places a very tall order on prerequisites:
- It requires linearity. There is no single, common timeline sheltering all possible initiatives in the vast realm of marketing. Even less so one in which such investments and our desired business outcomes coexist. A fictional “snapshot” is required when cause and effect reside in parallel realities — one that is defined by our even more arbitrary decision-making milestones.
- It requires causality. Causality does not happen in aggregate, but instead at very granular level. More specifically, at customer, user, fan, visitor, lead level. This means cross-device and cross-media integration at a user level are imperative if attribution is to work. In other words, attribution requires the famous 360-degree view of a statistically significant amount of potential customers. More on that separate fiction in a minute.
- When pertaining to humans (always the case thus far), the understanding of attribution is limited by the understanding of the human mind. Our culture, life experiences, perceptions… affect a system (human brain) which we do not entirely comprehend. Now, can the human mind ever understand its own intricacies?
Okay, this was somewhat vague and abstract — highly relevant picture of the lizard trying to understand the lizard brain aside — but the ground is now paved for a more straight-forward explanation.
So here’s a second try. Attribution will not happen because:
- There is no common timeline. Many campaigns, channels, or media have a longer term impact on true business outcomes that we can even measure. Are you prepared to maintain never-ending conversion funnels so as to properly take into account the long-term impact of social reach and cultural associations? Will you really be able to compute all possible conversions?
- There is no single customer view. People are less “digitally unique” and traceable every day. First because they choose to be. Second, because they scatter their attention and touchpoints across multiple, isolated environments. Third, because privacy standards or data protection laws prevent further integration.
Sure, I will have to explain myself much better on this latter point:
- Useful as it is for its primary purposes, cross-device identification is not enough to get us to a single view of even a fraction of our potential customers. And it will only get worse unless we bring supercookies back into the picture, which is highly unlikely given the next point below.
- We easily brush privacy compliance aside, but the EU’s upcoming General Data Protection Regulation (“GDPR”) [PDF] will draw a red line in the sand (or the Atlantic Ocean), with the many US companies participating in the Privacy Shield program most likely dragged into a much harsher reality. Furthermore, the FTC has identified the compliance of cross-device identification activities as one of its top priorities, while the Federal Communications Commission (“FCC”) has just introduced an opt-in approach in its own Broadband Consumer Privacy Rules.
- Social trends run counter to a single identity. Surely much more important than regulatory limitations, as these only follow social unrest. But society keeps finding much more effective and enforceable means of defending itself against lack of transparency.
“But what about new models combining deterministic information (truly integrated at granular level) with probabilistic data? Have we not overcome technical and legal constraints with smart algorithms?”
For starters, even though I proposed this myself as a solution at the time, probabilistic models are now facing the same legal challenge: the EU’s GDPR will label this non-PII data as “pseudonymous” (rather than anonymous) if it can be used for profiling purposes, and the collection or processing of such data will be subject to the very same limitations/burdens as of May 2018. And this month’s ruling on IP addresses by the European Court of Justice will ensure that the very concept of PII as a threshold for compliance becomes a thing of the past well before that.
Secondly, does it really matter that you put together the best sounding algorithms and weight distribution alternatives when all you ponder are touchpoints within your sphere of control? The core limitations have not changed one bit, and yet we place our faith on the more sophisticated blend. Do we simply want to believe in magic?
Now, as convinced as I am that attribution does not work in itself, I can surely appreciate that attribution efforts (i.e., investments in the pursuit of such nirvana) do in fact produce positive and tangible results.
2. THE LEGITIMATE PURSUIT: FINDING ROI
Taking it from that point, if attribution is impossible, useless, even illegal (!)… then why do we spend fortunes on this mission? Look at the amount of effort gone into the said blending of space and time:
eMarketer estimates that over 50% of US marketers are using digital attribution models in 2016, with over 60% expecting to expand attribution to offline channels in 2017.
A recent report by eConsultancy concluded that 43% of organizations reported having a single customer view — yet only 12% claimed to have the required technology in place. As pointed out by by David Raab (see his comment on Scott Brinker’s post), the companies who say they have built a complete view without the technology “are either magicians or fooling themselves.”
Perhaps more relevant are the findings of the multiple workshops and roundtables on attribution taking place across uncountable marketing events worldwide. Having attended a few of them myself over the years, I can confirm that the following summary from a recent (eConsultancy) Digital Cream gathering could have perfectly been transposed from any other:
“Marketers found that offline data was very difficult to match up with online data as there was a lack of customer identification at offline touchpoints. This meant that that measuring ROI for online campaigns in the offline space was nearly impossible.”
Has anybody actually got this thing to work? (Successful readers, please do comment!)
From a very cynical point of view, were attribution achievable or had it ever been attained by anyone alive, the never-ending impact of eternal ROI would have drained every other source of income on the face of the planet. Those in possession of such formula would drive unlimited resources towards one offering after another making the legend of Midas pale in comparison.
All of which leads me to conclude that there is no such Holy Grail.
A few recent studies would confirm that I am not alone: Gartner’s recent Hype Cycle for Digital Marketing and Advertising (2016) saw attribution sliding down the Peak of Inflated Expectations deeper into the Trough of Disillusionment. According to Gartner, reasons for the descent range from unrealistic expectations to vendor hype. In contrast, old-fashioned, aggregate data-based marketing mix modeling was considered to remain in the Plateau of Productivity.
Where, then, is the ROI of attribution? Despite all of the above, I could never understate the beneficial side effects of this effort. To name a few:
- A welcome understanding of campaign naming conventions across the entire organization (and its multiple agencies).
- An imperative to audit the organization’s martech and adtech stack to ensure maximum interoperability, discarding legacy solutions unable to provide reliable reporting APIs or bulk access to client data.
- A push for brand-driven data governance and first-party measurement as opposed to agency-driven, platform-driven, or media-driven measurement.
- Ensuring full control over the “data layer” for normalized data collection purposes.
- Insights on the sequence/story that customers have chosen to put together with the myriad of digital experiences that we enable, providing basic visibility on previously hidden touchpoints.
Do any or all of these justify your investment? Perhaps, if you ponder the amount of money you could have wasted in useless campaigns that you cannot truly measure anyhow.
But hardly so if you compute the huge cost of opportunity incurred. An attribution project happens at the expense of many others. And the side effects listed above can become far more agile stand-alone endeavors when purposely tackled.
(To be completely fair, some of these side-effects are already identified as the final goal in certain, definitely more sensible, approaches to attribution).
3. WHAT IS THE ALTERNATIVE?
As a summary of all of the above, cross-channel attribution serves a much higher purpose than the attainment of ROI in itself. The alignment of resources and minds towards such common purpose results in a powerful driving force. Quite the paradox: emotions driving the quest to encapsulate emotions.
Should we then find a new dream? Is there an alternative source of inspiration that is actually attainable? Something that has indeed been achieved by colleagues or competitors and in fact resulted in a tangible competitive advantage?
Equally important: does turning our backs on attribution imply a denial of the possibilities of data? Should we stop believing in its ability to determine our priorities, discover anomalies, validate hypotheses, or unveil truly useful insights?
Not at all. There simply is something very different to “attribution” as the pinnacle of data-driven marketing.
How about starting with the acceptance of this new demand-led reality in which you cannot expect to shape or understand each customer journey, but instead you are finally able to obtain a single view of your own business. A single view of your brand. A single view of the experiences you provide. A “brand journey” for your customers.
Opening ourselves to this simple premise paves the ground for three more thoughts:
- Customer-level intelligence belongs in the space of data-powered automation/optimization/activation (data as an engine), not data-driven decision-making (data as a witness). In other words: Customer Data Platforms (CDPs) and Data Management Platforms (DMPs), not decision-support systems. As for the mentioned limitations of cross-media and cross-device identification, they will matter little when the focus is placed on the collection and storage of permission-based first-party data. That is, until it is customers who voluntarily store and share their own data at every demand-led interaction.
- We love to understand the inner components of every process because that is the world that many of us grew up in. But discerning the pieces is no longer needed to work with the whole. Aggregate data is not inherently inferior to granular data. And there is great power in correlations.
- The very nature of digital data (mostly unstructured or semi-structured) has provoked a database and data management revolution. The new models -or rather the fact that models are not a prerequisite for data collection- result in a peaceful coexistence between an in-depth understanding of the few (customers) and a shallow understanding of the many. This is aligned with the said new mindset to let go of the search for neverending causality/structure and, I believe, completely disrupts the traditional approach to “business intelligence.”
There are, in sum, enough open fronts to entertain the most hyperactive and ambitious of CMOs looking to reallocate a sizeable portion of their budgets — only this time backed by first-hand experience.
So, here are a few potential alternatives if you find yourself in the said Trough of Disillusionment, with the essential question being: what were you truly seeking in attribution?
- Was it all along about maximizing return on investment, understanding for “return” a very clear set of short-term measurable outcomes? Then the answer could be in marketing mix modeling (using aggregate data).
- Was it about making the most of the time and resources at hand? How about adopting agile marketing methodologies? (Get yourself a copy of Scott Brinker’s own Hacking Marketing book.)
- Were you hoping to find golden insights in the advanced analysis of multiple sources of data? Great, but why make that “data lake” project the new center of the universe when both decision-making (data-as-a-witness) and data activation (data-as-an-engine) can still fly much faster and further on their own?
- Were you instead looking for medium-specific optimization through behavioral, cookie-based analytics and testing? Digital analytics may then be what you need.
- Was it about nurturing one-to-one relationships with your customers? Do you really need ongoing, human decision-making for that? Customer data platforms (CDPs), primarily first-party data, and data management platforms (DMPs), primarily third-party data, will let you activate customer data in any content personalization or media buying platforms you may want to plug into them.
If however, what you really are looking for is the said holistic view of your owned, paid, and earned media, then we are back into the realm of human-driven decisions, information delivery, and performance management. And this requires special treatment.
4. A NEW HOPE
We have discussed the growing importance of a 360-degree view of your brand and the experiences it creates in the face of the marketing revolution that a demand-led world has brought about. This purpose is fully aligned with the promise of omni-channel intelligence and not contaminated with customer-level data integration imperatives.
It is at such decision-making level, where ideas meet data, that the most crucial things happen. For the executive decision is the one step that defines the business and cannot be automated.
This concept, somehow sitting “between traditional BI reporting and advanced analytics,” with the specific challenges of digital data and marketing agility in mind, has finally been recognized as a space in its own right by Gartner’s recent Guide for Marketing Dashboards.
And even though “dashboard” is a very generic word that pretty much applies to everything these days, I believe the Gartner team has done a great job at both differentiating “pure players” from generic self-service BI or digital analytics solutions and setting all of them apart from data visualization (or visual discovery) offerings — something long overdue.
Most importantly, the authors mention a few of the things that are unique at this intersection of data/science/performance and ideas/decisions/collective intelligence:
- Data connectivity for multiple owned, earned, paid media sources
- Built-in marketing templates and metrics
- Workflow features allowing the dissemination of insights
And I certainly believe that some of the promises of this new, growing space are just as exciting as those of attribution. To name a few:
- An aggregate summary of investments and partial outcomes. Knowing how you are doing half way into the race should be a good indicator of eventual success. Not to mention the cost savings associated to putting an end to countless hours of manual reporting work.
- Anomaly detection models that take advantage of a very large amount of behavioral and media performance data. Cost savings come in this case from even more expensive analysts’ hours.
- “Contribution models” or “journey distribution flows”, computing aggregate reach, response, or behavioral milestones along the path of online and offline experiences offered by the brand.
- A combination of social media metrics with intentional and brand equity benchmarks originated in trusted third parties better positioned to gather consumer feedback than most individual organizations.
All of them cost significantly less than attribution projects and happen within weeks. But given that I am biased — Sweetspot is both focused on the above and one of the four solutions listed as “pure players” in the Gartner guide — you should not take my word for granted.
There you have it, if you must: your ROI.
I very much expect a heated debate. Please bring it on.
Thank you, Sergio. Readers: do you agree or disagree? Share your viewpoint in the comments below, or if you have a longer rebuttal for marketing attribution, let me know — I’d be happy to publish a well-argued counterpoint. A more extensive report from Sergio’s perspective with examples is available on Sweetspot’s website.