Deterministic vs. Probabilistic Tracking: Grasping Data…

Deterministic vs. Probabilistic Tracking: Grasping Data Across Devices

Multiple DevicesThe National Sleep Foundation recommends that we get eight hours of sleep each night. The average American falls short, logging a nightly average of 6.8 hours. Something we do spend eight hours doing each day, however, is consuming media.

According a 2015 ZenithOptimedia study, Americans spend 490 minutes with some sort of media. Given the proliferation of mobile, those 490 minutes are spread across various devices. Perhaps you check your phone in the morning; use the Internet at work all day, sneaking quick phone breaks to text or check Instagram; and then watch Netflix on your tablet before bed.

Marketers are naturally interested in what we’re doing as we transition across devices, in order to serve us the most relevant possible ads. With the rise of programmatic buying, marketers can strive to provide tailored—and even sequential—advertisements that support the consumer wherever they are in their journey. Whether the shopper just discovered a brand, or is considering buying a specific product, marketers use data to measure how they move through the purchase funnel across devices.

In order to provide a linear, supportive experience to the consumer, cross-device tracking becomes all the more crucial. Just as there are multiple methods for collecting data, such as cookies, data onboarding or data marketplaces, there isn’t a single way to track data across devices. Let’s break down deterministic vs. probabilistic tracking, and what makes these approaches different.

Deterministic data tracking has long been considered the gold standard of identifying consumers; the term “deterministic” refers only to data that is verified and true. The deterministic tracking method offers obvious value; there’s no limit to what you can track when you own the data.

Examples include personal information that customers enter themselves – name, address, phone number – when making an online purchase. You can then assign that information to a specific person and know who they are every time they log in.

“For example, let’s say you’re logged into Facebook across every device you own. That’s connected to a single email address. [A marketer] could connect the desktop cookie, and then connect that to a mobile ID,” said Adam Berke, President and Chief Marketing Officer at AdRoll at Collision 2016. “That’s great for Facebook, but all of us aren’t Facebook—so we need to look at these other ways to create cross-device IDs.”

The main case made for deterministic tracking is its accuracy. However, it also has limitations. Big players like Facebook and Google tend to take a “walled garden” approach with their data, and don’t give outside marketers the ability to leverage their data for external campaigns. At the same time, it’s impossible to truly create deterministic device graphs to correspond with every single internet user out there.

Even Facebook has to rely on the other major approach to device tracking: Probabilistic data tracking, which is focused on defining the unknown by leveraging data to determine probability. This method involves speculating and extrapolating based on what’s happened previously. The difference between probabilistic and deterministic tracking is that the former can be affected by outside variables while the latter is set in stone.

“This looks at the behaviors that you make on one device, which are probably in some way similar to the behavior that you make on another device,” explained Berke. “Your devices probably tend to travel around together, so you can pump all of these data points into a predictive model and create what’s called a ‘probabilistic ID.’ Meaning that it’s probably a match.”

One example of probabilistic analysis is linking devices; meaning, you might extrapolate that because two different devices use the same internet connection, they’re part of the same household. It’s a guess, but it’s a good one. Weather reports are another example of probabilistic analysis; meteorologists can make very educated predictions, but at the end of the day there are a million other factors that could change the outcome. The analysis of weather patterns may be highly accurate, but it’s still only predictive—we’ve all been caught in the rain without an umbrella before, thanks to a mistaken weather report. Compare this with deterministic analysis, which might be something like predicting what day of the week your birthday will fall in 2016—this is determined based on an immutable fact.

Perhaps surprisingly, accuracy isn’t always an issue with probabilistic data. A Nielsen test with cross-device software solution Tapad found that the accuracy of identifying a customer through probabilistic data can be greater than 90%.

The main limitation of probabilistic tracking is how complicated it is. A lot of data goes into those extrapolations, which means that to make assumptions with an accuracy above 90% requires the talent, tracking, software and data science to continue improving your analysis. At Connexity we’ve spent decades understanding and supporting consumers on their shopping journey—no matter where they are. To learn more about tracking consumers across devices, contact us here.