Exploring the inner workings of Ibex: A deep dive into the Eligibility Matrix & Deposit Models

Connor Coutts
March 21, 2023
Data Science
Back to blogs
Welcome to the second instalment of our series on understanding how Ibex, iGaming’s first self-driving retention tool, makes decisions.
In our last article, we outlined all the different components that make up Ibex’s decision-making process, including the Eligibility Matrix, Deposit Rate Model and Deposit Amount Model. In this article, we will delve deeper into the inner workings of these three components, and explain how they work together to determine the next best action for each player at each point in time, and play a crucial role in maximising player retention and engagement. So, let’s dive in and unlock the power of Ibex!
Eligibility Matrix
The eligibility matrix is the first key component of the decision making process which ascertains the customers eligible to be contacted with marketing actions.
The output from this step of the decision making process is a matrix of customer ID and action ID pairs for everyone who could be contacted, rather than everyone that will be contacted.
  1. Initial matrix creation: The starting point of this process is the creation of all possible combinations of contactable customers and active actions. This forms the base of the eligibility matrix. From here, the list is refined based on specific criteria, such as brand, country and site, and further narrowed down through the use of audiences, which target specific customer types or customers with specific habits (such as days since last deposit).
  2. Locked customer removal: The next step in the eligibility process is to remove any customers who have been locked out of Ibex for various reasons, such as being a bonus abuser or still in the analysis period from the last action they received.
  3. Inactive player removal: This reduces the contact rate for players who have been inactive for more than two weeks. With the analysis period set to two days, a player would expect to be contacted every other day. However, if they have no activity for two weeks, this duration will start to increase.
  4. Event stagger: Finally, to prevent new customers and brands from having all their customers contacted on the same day, Ibex uses an event stagger. If the output of the eligibility matrix on a given day has too large a percentage of the customer base, it will remove some customers with the intent of smoothing out the number of events sent out on a given day.
Deposit Rate Model
The purpose of the deposit rate model is to predict the likelihood of a player making a deposit in the short-term window after receiving an offer.
This model takes into account the players past behaviour combined with information relating to the offer.
The offer, or more accurately an action Ibex takes could be a monetary promotion, an informational communication, or simply the action of doing nothing at all. By comparing the outcome of taking a specific action and not taking any action, we can determine the predicted impact in deposit rate.
Deep neural networks (DNNs)
Feed-Forward Neural Network with one hidden layer
To model the deposit rate, we use Deep Neural Networks (DNNs) due to the large volume of training data and the complexity of the dataset.
DNNs are a family of machine learning models commonly used for supervised learning problems where there is a large amount of training data. They consist of multiple layers of nodes that can model non-linear relationships in the data and the Ibex data science team uses industry-standard tools such as Keras and TensorFlow to train and deploy these models.
Ibex has dozens of input variables at its disposal to model deposit rates but they can be broadly categorised:
  1. Player demographics (age, gender, location etc.)
  2. Timeseries transactional data (deposits, withdrawals, losses etc.)
  3. Time and date (day of month, time of day, etc,)
  4. Action metadata (offer type, cost parameters etc.)
  5. Historic action outcomes (how players responded to previous offers)
Deposit Amount Model
The deposit amount model predicts the amount a player is likely to deposit after receiving an offer.
The process of calculating predicted deposit amounts is simple:
Expected Avg Deposit x Multiday Deposit Factor x Deposit Amount Factor
The multiday deposit factor accounts for the fact that customers who receive an offer and then deposit, tend to deposit on multiple days.
The deposit amount factor is a value calculated based on the offer, the players value compared to the rest of the player base, and the players location, among other factors. This factors value is a comparison of the deposit amount between our campaign (players who received a marketing action) and control (players who didn’t receive a marketing action) groups.
In cases where the model does not have access to all the necessary information, an average factor is calculated from the available data, which is then included in our training dataset to improve the models accuracy in future cases.
At this point in the decision making process, Ibex can determine:
  1. Which players can be contacted with marketing actions on a given day
  2. The predicted deposit rate impact of sending marketing actions
  3. The predicted deposit amount uplift
However, this is only 1/3rd of the Ibex decision making process.
In our next article we will discuss the next three components; the GGR prediction, bonus cost model and other cost estimates.

Grow your business

Use Ibex to drive customer engagement and deliver the right messages at the right time
Start Now