Deterministic and probabilistic data models in email marketing

God doesn't play dice

How we see and interpret variation in the world affects the decisions we make. Part of understanding variation requires an understanding of the differences between deterministic and probabilistic models.

What is a model anyway?

The word ‘model’ is itself problematic. There are lots of ways to use the word, two of which are especially relevant to this discussion. The first definition is “a mathematical model as a decision-making tool”. The second definition is “a way of thinking or representing an idea,” which is more akin to linguistics.

Our world is full of models:
• A customer database is a model of a company’s customers
• A bus timetable is a model of where buses should appear
• A map is a model of an area, which can help us get from A to B

Deterministic models

Deterministic models presume certainty in everything. Examples of deterministic models include age, gender, and RFM – all of which can easily be ‘sliced and diced’.

Probabilistic models

Probabilistic models include customer preferences, category & product recommendations, and customer loyalty.

The concept of probabilistic models might sound complicated, but they begin with the simple idea that each customer has unique and evolving tastes and interests. Given the nuances of customer preferences, most customer data models should arguably be probabilistic rather than deterministic, but this can be hard to achieve in a traditional CRM database.

So how do probabilistic recommendations work in practice? There are two broad approaches: collaborative filtering and content-based filtering. Each approach has its own strengths and weaknesses, but both allow the marketer to automatically rank products for each specific consumer. This results in better personalization, faster campaign deployment, and higher ROI.

Models and email marketing

To address the personalization requirements of the modern email marketer, new technologies combine both deterministic and probabilistic models, allowing marketers to really dig into what a consumer likes, or doesn’t like. Then to use that data for product recommendations or campaign timing.

There are important roles for both deterministic and probabilistic models in email marketing, and as more marketers become comfortable with probabilistic optimisation technologies, these models will become a regular part of the savvy marketer’s playbook.

P.S. Here’s a handy pre-populated Tweet to help you share the new post 🙂

Cover image courtesy of Welder at Deviantart

Comments are closed.