In this article, we will discuss the different types of Markov models and how they are used for attribution modeling.

Markov models are a mathematical approach to measuring attribution probability that is particularly useful when analyzing customer journey.

## What is a Markov Chain?

Markov chains are a useful tool for predicting the likelihood of an event occurring based on previous events. This is extremely useful for predictive marketing to represent a consumer's journey through different websites, with the probability of moving from one website to another changing as the journey progresses.

A Markov chain has three important parts.

**The first part** is the state space, which is the different states that something can be in. In this example, the state space is just sunny or cloudy.

**The second part** is the Markov assumption, which says that something only depends on what happened before it. In this case, the weather only depends on the weather from the previous day.

**The third part** is the transition probabilities, which tell you how likely it is to move from one state to another.

## Simple Markov Model

The Simple Markov Model is a way to figure out which website gets credit for a sale in affiliate marketing by looking at the probability that someone will buy something from a website based on their previous actions.

To figure this out, the model uses something called the Removal Effect Approach. This means that the model removes one website at a time and looks at how it affects the overall conversion rate. The conversion rate is the percentage of people who buy something after clicking on a website's link.

So, if the conversion rate goes down a lot after removing a website, then that website deserves a lot of credit for the sale. On the other hand, if the conversion rate doesn't change much, then that website doesn't deserve as much credit.

## Hidden Markov Model

The Hidden Markov Model is used to predict the probability of a consumer purchasing at different stages of their journey. The observed data would be the actions taken by the customer, such as clicking on a link, visiting a website, or making a purchase.

The HMM would then be used to predict the most likely sequence of states that a new customer will go through based on their observed actions. This information can be used by affiliate marketers to optimize their campaigns and better understand their customers' behavior.

The goal is then to find the combination of hidden states that maximizes the probability of the a particular customer behavior or at least identify patterns in customer behavior that may not be immediately apparent, allowing them to create more targeted and effective marketing campaigns.