The Pros And Cons Of Probabilistic Attribution

Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.

Today’s column is written by AJ Brown, CEO and co-founder of LeadsRx.

The mother of all “what ifs” for enterprise marketers is: What if all consumer tracking went away, eliminating the possibility of identifying anonymous markers? What if the most horrible predictions came true: third-party cookies – gone; local storage – removed; first-party cookies – blocked; IP addresses – cloaked.

Don’t sit on your laurels hoping the tech giants will delay the “what if” scenario again. Take control of your marketing futures now; your job and your brand’s survival depend on it.

It’s critical that marketers own their data-driven marketing ambitions and become proficient at mastering data and analytics. Among the important things to consider is whether you take a deterministic or a probabilistic approach to marketing. Which methodology is best for cross-device measurement, and how can you prepare for the possibility of an apocalyptic change to tracking consumer behavior?

Marketers have long relied on deterministic analytics because, they rightfully argue, it provides the accuracy they need to make sound business decisions. But new “probabilistic” methodologies are starting to emerge as attribution vendors reduce their dependence on cookies. In the apocalyptic scenario of losing capabilities, there simply isn’t a technical way to remain deterministic.

For attribution, a “deterministic” approach simply means that a unique identifier is used to unify customer activity across ad campaigns, browsers, devices and apps. Anonymity is maintained, and the key benefit to deterministic approaches is that marketers have a great degree of confidence knowing all activity has been correctly unified to the same individual. But moving forward, that will be more difficult to achieve.

Marketers must now learn probabilistic approaches. Mathematics must be used to work backward from conversion points to calculate the probability of linked activity. Experiential data is used to drive the probabilistic models. But while this sounds like a reasonable approach, the issue is: Where does that experiential data come from?

The Good and the Bad of a Probabilistic Approach

The trick is to use deterministic tracking while you can. The data collected will later fuel probabilistic models in the proposed apocalyptic new world. Here’s a breakdown of the pros and cons of a probabilistic approach:

Pros of probabilistic attribution

  • Works in a cookie-deprived future. Provides an excellent assessment of overall marketing performance.
  • Provides data for analyzing direction trends and the influence of touch points, which can be used to optimize marketing spend.
  • Delivers the “big picture” of attribution results without the need to capture large volumes of data.
  • Offers a quick, top-down assessment of the marketing programs that are working well.
  • Works in the future when deterministic tracking may no longer be available.

Cons of probabilistic attribution

  • The data relies on a “probability” and is likely less accurate in terms of empirical data points.
  • As consumer buying patterns change, these models will require new training to learn how to adjust probabilities more quickly and maintain accuracy.
  • Viewing individual consumer paths to purchase are not supported, leaving them a mystery.
  • Unreliable counts of impressions or ad clicks.

How Probabilistic Attribution Works

There are many complexities when dealing with probabilities, human behavior and the ever-changing nature of marketing programs. But let’s look at a very simplistic example of how probabilities can be used to determine attribution.

Let’s start by collecting deterministic data across as many marketing touch points as possible; this includes impressions and clicks on digital ads, exposure to broadcast radio and television spots, podcast and OTT ads.

Next, using attribution, build a probability map of these touch points leading to conversions, or customer acquisition events. While not 100% accurate, these probabilities can be used in the future when deterministic tracking may no longer be available.

What Marketers Should Know

It’s obvious by now that probabilistic attribution, just as deterministic attribution does, requires quality data.

Marketers looking to win should measure their probabilistic models against industry benchmarks, or aggregated data that can be used to further train probability maps. It’s one thing to have your own data for your own business, but when this data can be compared and analyzed against broader industry benchmarks, probability accuracy can be improved.

Fortunately, we are not yet in our hypothetical apocalyptic world. Deterministic tracking is still possible. Even with the total demise of third-party cookies (as Google suggests it is doing), there are many other ways to continue deterministic approaches. But if the last year has taught marketers one thing, it’s that worst-case-scenario planning should be a top priority.

Follow LeadsRx (@LeadsRx) and AdExchanger (@AdExchanger) on Twitter.

Enjoying this content?

Sign up to be an AdExchanger Member today and get unlimited access to articles like this, plus proprietary data and research, conference discounts, on-demand access to event content, and more!

Join Today!

 

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>