How We Used Machine Learning to Help a Ticket Retailer Fill Empty Seats With Predictive Modeling
This Ticket Retailer (“the Retailer”) is an online ticket marketplace where users can buy and sell tickets to a variety of live entertainment events, from hockey games to theater performances.
But like all ticket retailers, the Retailer was facing challenges with empty seats and no-shows, or those who purchase or reserve seats but ultimately do not attend the event. They wanted to find a way to optimize event attendance and fill as many seats as possible, despite the risk of no-shows.
The Retailer had an innovative idea that would allow them to monetize no-shows, but they needed the help of a predictive modeling guru. Centric Consulting and our machine learning experts immediately came to mind. So together, we formed a partnership to develop a new product that uses predictive modeling to estimate the number of empty seats for a given event, based on a variety of factors and audience demographics. With our expertise, this could help fill stadiums, theaters, and venues across the country, and help the Retailer earn more revenue off ticket sales.
Enter Centric: Using Machine Learning to Monetize No-Shows
Together with the Retailer and Centric India’s offshore experts, we set standards and established matrices while building out their strong predictive analysis framework. Leveraging Python and AWS Sagemaker, Glue, and Lambda, Centric helped deliver the product, capable of maximizing event attendance and ticket sales.
In the first step, we worked with the Retailer to define and assess criteria with certain predictive power. We pulled these criteria from ticket scans from previous events, and they help estimate the rate of event attendance for future events and the chance for no-shows. They include factors like:
- What day of the week is the event?
- How many tickets did each customer purchase?
- How far away the customer lives from the event venue?
- When did the customer purchase the ticket?
All of these factors can influence an attendee’s decision to actually make it to the event and can potentially increase the opportunity for ticket resale.
After defining the criteria, we moved on to preprocessing data obtained from partners to ensure it was a fit for feature engineering. Then, we extracted advanced features out of the provided data to feed into the predictive models.
After running the predictive models for six weeks, we pulled data to determine its effectiveness. This would help determine how well the model could predict the rate of no-shows based on the criteria we gave it. Results from these tests allowed us to improve the predictive model and better select the data with the most predictive power.
With our help, the Retailer was able to build a battle-ready product capable of predicting event attendance based on selected criteria. Using the predictive modeling architecture, powered by Amazon’s cloud offerings, the Retailer will be able to increase profits through ticket resales.
The DevOps strategy we employed minimizes the amount of maintenance required to keep the model up and running, allowing the Retailer to focus their effort on other endeavors. In addition, throughout the project, Centric India developed a stronger system of communication for the Retailer, especially around feature engineering outputs and deployments.