Deterministic vs. probabilistic attribution

In modern marketing, understanding user behavior is essential. Knowing what consumers want and how they interact with various touchpoints empowers marketers to engage them effectively.

To do that, marketers aim to attribute conversions and sales to the marketing tactics that drove them to action. It isn’t easy, though. There are many methods to choose from, each with its own pros and cons. 

So, which attribution models are most marketers using? Rather than method, let’s first discuss the category of attribution, for which they are two choices: probabilistic and deterministic. What are they, and which is ideal? Glad you asked. 

What is deterministic attribution?

Deterministic methods of attribution, quite simply, determine how the customer journey ended with them buying a product, signing up for an account, or subscribing to a service. They do this by identifying the touchpoints that led to a conversion. What was the customer last doing, and where was it before they bought your product or service? That's what receives the attribution, whether that's an ad network or a marketing email.

Attribution points can include:

• Paid ad networks
• Particular campaigns, e.g., a social media post run or blog series
• Website browsing
• Smartphone apps

And so many more. Deterministic models rely on accurate identifiers to highlight the single touchpoint that led to a conversion event.

Let's take a break from tech and head to the great outdoors. Think about a ranger keeping track of wildlife in a state park. That ranger can see raccoon footprints near the bins, so they know raccoons are present and probably responsible for causing the latest litter-filled mess. The ranger also spots bear prints and where deer have been nibbling at trees, so the evidence is clear that these animals are thriving in this area.

But what if the raccoons did come near the bins, but it was the bear that raided the trash? The ranger might need to move the bins or warn visitors, but based on the evidence in front of them, they won't know to take either of these actions.

That's one reason why deterministic attribution is fundamentally flawed. It doesn't tell you the whole story and relies on multiple assumptions. Plus, deterministic methods need data points from every touchpoint, which aren't always available.

What happens with app tracking transparency (ATT) and other tracking obstacles?

While deterministic attribution relies on comprehensive tracking, growing privacy restrictions (like Apple's App Tracking Transparency or ATT) make it difficult to capture the full customer journey. From iOS 14.5 onward, for example, Apple's privacy-centric options allow customers to opt out of tracking as soon as an app opens.

Preventing tracking means that deterministic attribution models will only have data on visible touchpoints. This can skew campaign performance results, leading marketers to believe that some aspects of the campaign are underperforming or over performing. In truth, without that all-important data from the hidden marketing touchpoints of the customer journey, attribution becomes largely guesswork.

Of course, these developments are good for everyone. It's important to protect user privacy and ensure your company complies with data protection regulations like the GDPR and CCPA. However, it means marketers relying on a wholly deterministic approach must employ other attribution methods to see a more complete picture.

Pros and cons of deterministic attribution

Deterministic attribution is simple to understand but doesn't reveal the whole picture to marketers.

Pros

• A simple model that's easy to implement
• Highlights clear, factual data on customer journey touchpoints, e.g., a Google Ad or e-commerce website pop-up

Cons

• Oversimplifies the customer journey
• Potentially skews marketing performance data
• Doesn't provide marketers with access to data about other touchpoints that could have been vital in influencing the customer journey

What is probabilistic attribution?

Probabilistic attribution moves away from relying on existing touchpoint data. Instead, it takes a statistical approach, using probability theory to consider all the touchpoints a customer may interact with and the likelihood of each contributing to a conversion. The primary advantage is that, while more complex, marketers get a complete picture of campaign effectiveness and data on the true impact of each aspect of that campaign.

Back to the analogy: our ranger got fed up with dealing with nature and made a move to the big city to become a private investigator. Suddenly, the evidence is much more complex instead of clear footprints and claw marks. A stray conversation here and a blurry photo there require all his deductive skills to come to the right conclusion. He constantly learns more about his clients' behavior and becomes skilled at figuring out "whodunnit" every time.

A probabilistic attribution model works in a similar way, gathering what evidence is available and then using statistical analysis combined with AI and machine learning to fill in the gaps. Machine learning algorithms can deal with huge volumes of data and, like our PI, learn a little more about consumer behaviors and trends all the time. Over time, the accuracy of these models improves, helping marketers generate more revenue.

Probabilistic attribution in e-commerce: an example

If your e-commerce store gets a hit, it could be via an ad click. But is that the only source of attribution? Let's take a look at the true customer path:

1. They search on Google for a place to buy the goods they need.
2. They find a blog on your website about the history of those goods.
3. They read this and, from here, download your app, which has more interesting articles.
4. Later, they search directly for your online store.
5. The top result is your ad, so they click through and make a purchase.

Using the deterministic attribution model, the only data collected on attribution would be about the online advertisement. However, the probabilistic attribution method will assign credit to each of the touchpoints on that pathway.

Marketers can keep paying online and mobile advertisers to keep their ad campaigns running. However, they also know that their blog campaign performance is high, and SEO is helping their content rank highly. They also understand that their app helps cement the company name in users' minds. All these data points empower marketers to channel their efforts (and money) toward the right channels.

Pros and cons of probabilistic methods

Probabilistic attribution has many advantages over a purely deterministic model when it comes to campaign measurement. However, like most marketing techniques, it's not perfect.

Pros

• Creates a complete picture of consumer behavior
• Ensures marketers know all the consumer touchpoints leading to conversions
• Can consistently improve when provided with continuous, accurate data

Cons

• Complex to understand and potentially tricky to implement
• Relies on data points like IP addresses and device details, which aren't always accurate and may be hidden, depending on the user's operating system.

Why we prefer probabilistic attribution

Experienced marketers have long been cautious against relying too heavily on touch-based attribution when pitting activities with different goals against each other, like demand generation and demand capture. 

Due to the deprecation of third-party cookies and the restrictions imposed on cross-app tracking, deterministic attribution models of the past have become less accurate. In the future, marketers will need to rely more on probability than certainty to measure the effects of their marketing efforts and determine where to allocate their marketing budgets.

With the decline of third-party tracking, marketers must adapt by embracing probabilistic models and sophisticated techniques like marketing mix modeling. These approaches, increasingly accessible to all, offer a more accurate path forward for measuring success and optimizing campaigns.

That path uses a blend of marketing mix modeling, multi-touch attribution and incrementality testing (a method we call triangulation) to determine what tactics are most contributing to results and where marketing spend is best placed. See the Funnel website for more information on accurately measuring campaign performance and attributing as accurately as possible.