The new buzzwords
Thereās two new buzzwords taking over the DTC scene: ādeterministicā and āprobabilistic.ā While these are old statistical words with deep histories, theyāre starting to appear in DTC circles. So what do they mean? Consider this your primer on deterministic versus probabilistic.
In measuring ad performance, thereās two ways of doing it: tracking what actually occurred, or modeling what you think may have occurred, using statistics. In an extreme simplification, this is ādeterministic vs probabilistic.ā
Deterministic: an exact science?
Deterministic measurement is using explicitly-known data points to make a conclusion about what happened. In marketing, weād use user-level data to attribute conversions to touchpoints. Itās rule-based and relies on hard, verifiable data. Cookies, user IDs generated by device and identity graphs, and clicks.
In plain english: āWe know for sure a user clicked an ad and email on their path to purchase, so we give credit for that conversion to those channels.ā
Most growth marketers make a living on deterministic measurement. For a decade itās been the fulcrum for all creative experimentation and campaign scaling. All major ad platforms (Meta, Google, etc.) use a deterministic model in their in-platform attribution.
But deterministic measurement isnāt perfect: as privacy laws expand and as the walls of the āwalled gardensā get higher, deterministic measurement of performance marketing becomes more difficult. The iOS 14.5 update was the single biggest degradation of deterministic performance measurement on Meta in history. Overnight, a huge swath of deterministic datapoints were removed by Apple. Googleās now-shelved cookie deprecation plan would also remove a critical deterministic measurement point.
This is where probabilistic measurement comes in.
Probabilistic: weāre pretty sure?
When datapoints become scarce or further removed from empirical touchpoints, we have to use probabilistic measurement.
In probabilistic measurement, marketers are using statistical modeling or machine learning to assign credit for conversions to touchpoints - even when you cannot definitively tie the conversion to a specific event.
But why would you want to use this? More often than not, you donāt actually have deterministic data for some of your ad channels. Think TV, radio, or podcast - attaching conversions to those channels can be challenging. You have to āmodelā their impact on your revenue because thereās no such thing as a definitive touchpoint for āradio,ā for example.
Probabilistic measurement includes things like media mix modeling, linear regressions, or correlation analysis.
How it works: āThis model cannot see every step of your path to conversion, but based on patterns across your customer base, this journey looks like one where all the touchpoints attributed contributed to the conversion. This model will assign a value to each of those touchpoints.ā
Which one is better?
Trick question: thereās a time and place for each. Truth is, both deterministic and probabilistic have a place in your analysis. As marketers we must be scientists, and no good scientist would limit themselves to using only one tool for studying their craft.
Ask any scientist out there: the scientific method is just a way of adding order to chaos. Human beings are endlessly variable - our paths to purchase are influenced by our psychology, physiology, culture, history, learned behaviors, and straight up chance.
As a marketer, all you can try to do is use probabilistic and deterministic measurement as a way to approximate what actions your customers take on their path to conversion. In this way, you can strategize about what works and what doesnāt.
Probabilistic measurement in the form of MMM is useful when making sweeping forecasts about performance across channels, for example, as you cannot deterministically measure the future. Statistical analysis is valuable for measuring channels that lack deterministic touchpoints, like billboards or radio. It helps create actionable assumptions from the chaos.
Deterministic is more useful when youāre doing explicit experiments. Finding ways to get deterministic data for views, clicks, and other empirically-measurable datapoints gives you a sense of whatās āworkingā right now. Youāre probably already doing this in your reporting. The problem is that most marketers limit themselves exclusively to deterministic measurement when they should be using both. Legacy brands often exclusively use probabilistic measurement because their budgets and channels are so huge and lacking deterministic touchpoints.
Fact is, the best marketers on earth use both measurement techniques equally. True mastery comes from knowing when you use each. Sometimes itās worth taking a step back and looking at the big picture: as marketers, what are we really trying to do here?
Since weāre committed to marketing mastery, you can access both probabilistic and deterministic models in Northbeam, and theyāre woven into tools like MMM+ and Metrics Explorer.
As always, chat with us and we can help you achieve this mastery. Book time here.
Huge thank you to our own John Max Bolling for contributing to this weekās newsletter.
.png?upscale=true&width=1120&upscale=true&name=Group%2049737%20(6).png)
%20(1).png?upscale=true&width=1120&upscale=true&name=Blue%20and%20Orange%20Vintage%20Illustration%20Cocktail%20Menu%20(LinkedIn%20Post)%20(1).png)
