The Big Debate: Probabilistic vs. Deterministic

11月 20, 2016

Some time ago we passed a tipping point where marketers realized that targeting by device didn’t make much sense and a cross-device “people-focused” approach worked better.

With that matter settled, the next big debate was how to go about executing that cross-device targeting. There are two methods: Deterministic and probabilistic.

Much of the debate has revolved around the merits of each tactic. Generally, deterministic is more accurate while probabilistic offers more scale.

While the industry sorts out which is better, a more pragmatic line of question to pursue: Which one works best? The answer is both, depending on what you’re trying to do.

A quick primer on cross-device targeting

Years ago, consumers were mostly on their desktops. In that environment, cookies worked pretty well. However, sometime after the introduction of the iPhone in 2007, mobile began consuming more of consumers’ time.

In 2016, consumers spend an average of 3 hours, 6 minutes on their mobile devices versus 2 hours, 11 minutes on desktop, according to eMarketer. The average consumer flits between 7.2 devices during their day, Adobe estimates.

To get a full picture of those consumers, marketers need to connect those consumers’ desktop and mobile usage. One way is deterministic. When you sign on to Facebook on your laptop, phone and tablet, then Facebook and advertisers know for sure that you are the same person, who uses those three devices interchangeably throughout the day. Facebook’s not the only source of deterministic targeting. Publishers like The New York Times offer the same ability.

How do you measure consumer activity outside of Facebook and publishers’ sites? That’s where probabilistic targeting comes in. Probabilistic uses anonymized data including device type, browser type, IP address and OS to make an educated guess that connects users to devices.

In practice, many companies, including Oracle, use elements of both for a hybrid approach.

Targeting differences for marketing

For marketers, one primary difference between the two methods is - at least we have found - that deterministic tends to work better for quotidian purchases while probabilistic is more effective for big-ticket purchases.

For instance, if you’re selling paper towels or clothes, then you will probably want to base your media buying decisions on a consumer’s regular purchases. That type of data will be available via deterministic because you can be sure that this particular consumer is following a predictable path of purchases. In that case, you have already seen conversions and the goal is to try to recreate them.

On the other hand, if your target customer is in the market for a big-ticket item like a car or a vacation, then they might be moving into somewhat uncharted territory (for them) where their path to purchase is less predictable. In that case, you want to follow a path that has been set by other consumers in the same position. A good signal is if they’re over-indexing on content that indicate that they’re going to make a concerted purchase in the future, like auto reviews. In other words, there is no tangible evidence that this person has made a purchase like this before, but we can make assumptions about their behavior by taking into account their demographics and other factors.

There are gray areas, of course. A consumer might upgrade their car every year or so, which would make an auto purchase a fairly regular purchase. Another consumer might travel fairly regularly for work, so searching hotels and destinations isn’t necessarily a signal that anything unusual like booking their once-in-a-lifetime dream vacation is occurring.

Overall though, the two approaches have advantages that go beyond perceived accuracy and reach. Like Oracle, savvy marketers should conclude that the answer to the debate over deterministic vs. probabilistic isn’t an either/or proposition but rather a yes/and.