Imagine a world where every marketing dollar your company spends is linked directly or indirectly to every dollar generated.
Now ask yourself, what are the two indisputable data points in this equation? If your answer is the marketing cost and the sales revenue, your answer is right. Everything else is subjective. Every link in the chain between your marketing message and the transaction is subject to some level of probability and fractional attribution.
To challenge this assertion, let’s dissect the simplest digital advertising campaign: A $20 AdWords PPC campaign that drives one click-based e-commerce conversion of $20.
100% ROAS, right?
But wait: it turns out that the $20 soccer ball was researched on your mobile website anonymously while the consumer was supposed to be watching his kid’s game. That night he used Google to get back to your site. So yes, they clicked on your AdWord, but they would have made the purchase on your site anyway. Some of that 100% ROI should go to your rock-star web team that made a responsive site that made it possible to shop for the ball on the field.
This example shows how the current multi-channel marketing environment presents too many uncontrollable yet influential factors to trust measurement solutions that prioritize exactness over actionable breadth. To effectively measure your marketing spend you need to determine the right level of precision needed to make decisions. With cross-screen measurement, trying to be too precise carries significant risks. Statistical models, often informed by some volume of deterministic linkage, are far more actionable than data that is only available under constrained conditions.
Here are three reasons to prioritize action over precision.
- Precision can limit your marketing options.
Unless your measurement solution is ubiquitously deterministic across all devices, you will be limited to the confines of your partner’s walled garden. Case in point, recently a large Millennial Media advertiser was forced to pull their entire mobile advertising spend because their chosen fractional attribution partner could not support advertising in mobile apps. The advertiser wanted a single view of their spend across all channels but they had to give up channel flexibility and creative options. Was this decision really worth it?
Another Millennial Media advertiser tested Facebook-powered Atlas to track their multi-channel spend. The system works across channels, but the test showed clear limitations. Log-ins on Facebook’s mobile website are rare compared to usage of their mobile app. This limited the system’s ability to deterministically track the source of conversions that started outside of Facebook that ended with a transaction on the advertiser’s mobile web site.
For comparison purposes, the advertiser also leveraged Millennial Media’s cross-screen attribution product. This report showed how Millennial Media advertising contributed to some of the mobile web conversions. Transaction data from the advertiser’s ecommerce platform supported the evidence that conversion events were taking place on mobile devices. Atlas, however, was unable to attribute those conversions to a source. It was clear to the advertiser that this purely deterministic approach to measurement was not able to tell an accurate story more widely. The report significantly undervalued the advertiser’s spend in mobile.
- Precision could lead to the wrong action.
Economist John Maynard Keynes said, “it is better to be roughly right than precisely wrong.” The previous example illustrated this nicely. Had the advertiser taken the data at face value they would have concluded that all of their mobile web conversions were organic in nature and stopped advertising in mobile apps. They would have been precisely wrong.
Marketers have always used small (yet random) samples of data to allocate billions of marketing dollars. No one ever described these measurement techniques as exact, but they did provide answers to questions with a level of precision that was good enough to make decisions. Today, advertising platforms with large deterministic footprints make cross-device advertising with 100% precision and accuracy seem possible. The result is that marketers put determinism on a pedestal because of its purity and simplicity to explain to stakeholders.
However, relying too heavily on deterministic data to link users across devices introduces too much bias into the measurement sample to derive meaningful conclusions. A highly biased sample, no matter how large, ultimately leads to the precisely wrong actions.
- Precision exacerbates transparency and privacy concerns.
The idea of any single company owning and controlling a perfect map of every device with some level of visibility into every action on those devices is scary to most people. Apple recognizes this public sentiment and is trying to differentiate its products from Android devices by claiming they are more willing and capable of protecting consumer privacy. In their defense, Google told AdExchanger that “it will only use its logged-in data to kick-start a probabilistic model.” So, this leaves Facebook as the only real player left in the deterministic game.
Or does it?
Just like Google, Facebook’s size and breadth puts them under the magnifying glass from regulators. Their own marketing materials use people-based advertising as a euphemism for “We know a lot about you and are using it to target and measure you.” Facebook does not make claims about determinism for the same reason as Google. They don’t want to sound scary to consumers. You add privacy concerns to the other limitations of determinism and I am completely convinced that Facebook, even with all of its deterministic data, is using or will be using a probabilistic model to drive its people-based marketing strategy.
There is evidence to support this. Buried in the press related to Facebook’s abandoned plan to limit the collection of mobile device IDs, you can see that the original purpose for the initiative was to get people to embrace a device agnostic people-based view of marketing effectiveness. Facebook is trying to get their partners to give up their own deterministic view of attribution and embrace theirs. Throughout the controversy Facebook kept telling their partners and advertisers that they were going to get an even more informed view. I believe it. But, my interpretation is that Facebook’s new view will also include a statistical model that Facebook wants to protect. Handing all of those raw IDs and a cross-device map to their mobile measurement partners is not something that Facebook wants to do for multiple reasons, privacy being just one of them.
There is also the argument of pure necessity. Facebook’s usage numbers show that their user base is becoming less of a cross-screen audience. Their recent 10-Q shows that 44% of Facebook users never log into Facebook on a desktop computer. For Atlas to remain a relevant cross-device tracking system in mobile web and on desktop, they will need to embrace a probabilistic model of attribution to fill the gaps.
The value of statistical modeling is that it functions across channels, works around sample bias, and can protect consumer privacy. These facts are not going unnoticed. More and more industry experts are noting that fully deterministic device bridging is not a realistic solution for marketers.
What is realistic and exciting is the way that deterministic linkages will help inform and improve probabilistic device maps. These device maps will compete in the open market and power attribution models, optimization tools, and reporting visualizations. As these models improve and integrate they will provide marketers with powerful insights into how multi-channel marketing leads to sales and revenue growth.
When looking at any tool that relies on a device map to function, here are the questions to ask:
- How will this tool improve the effectiveness of my marketing dollars?
- What actions will my team make on the basis of this tool’s data?
- Does this tool expand my universe of marketing options or does it constrain me?
- If it does constrain me, can I live within the constraints?
The right measurement solution will vary by company, vertical, and marketing tactic. If you decide that you need a solution that is capable of tracking users across screens you will also need to accept that all cross-screen measurement will include some level of probability to function at scale. Business value and actionable insights should ultimately drive your measurement decision without an ideological pursuit of precision.
