How Donatos uses machine learning to retain customers | New Era Technology

Converting New Customers to Returning Customers

Donatos wanted to identify customers who were at risk of leaving so they could take action to win them back. Without the ability to derive analytics from their data and gain insight into consumer behavior, strategic planning was unsuccessful.

Achieving Company-wide Growth

With over 160 stores, Donatos Pizza had many that were succeeding, but weak customer retention meant that many stores were struggling. The leaders needed a solution to increase sales in these stores to achieve consistent and company-wide growth. 

Donatos Pizza wanted to look at data solutions that would lead to increased customer retention. Through a prior statistical analysis, company leaders knew that if customers returned within a specific time frame from their prior visit, they were likely to become long-term or loyal customers. Those who didn't return were unlikely to return at all. They wanted to identify at-risk customers and target them with successful strategies to regain their business and loyalty. 

Fusion Alliance sat down with Donatos to discuss their goals and determine the right course of action. Because they had a vast quantity of customer data readily available, we knew machine learning would offer the best opportunity to identify patterns and predict consumer behavior. 

Fusion Alliance created a three-month pilot program to implement and apply machine learning models in specific stores. We predicted that at the end of the program, the stores using machine learning would retain 30 percent of the identified at-risk customers. 

Goal: Identify the Algorithm that would Produce the Most Accurate Predictive Analysis

The value of results is directly dependent on the value of data the platform receives. With so much data available, we needed to be very specific in what we used and which algorithms we identified as the most accurate. 

Our solution: 

• Assessed the quality and quantity of data
• Developed predictive models
• Cleaned the data to use as a training set and use it to identify the best algorithm for accurate predictive models. 

Goal: Accelerate the Process while Reducing Costs

Machine learning and predictive analytics are often very expensive and risky. Fusion minimizes risk and cost through a use-case driven strategy where we start with the use case, pull in only necessary data, and create an iterative approach to deliver a working solution that brings more value to the client on a faster timeline.

Our solution: 

• Started with the use case. Specifically, identifying at-risk customers
• Performed an ETL (extract, transfer, and load) on their data to a cloud platform to improve efficiency
• Evaluated and selected the data most likely to provide accurate insight including pulling in source data, aggregating it, and removing aberrations

Goal: Hone in on a Specific Question to use in Machine Learning Models to Identify at-risk Customers

Machine learning offers the best outcomes when there is a specific, defined problem to solve. 

Our solution: 

• Formulate and test questions
• Iterate over the model to achieve a desired state
• Run the previous day's sales against the model to produce a list of at-risk customers