Teradata VantageCloud Lake creates a winning fan experience

Machine learning K-means cluster models help identify and understand fans

Personalizing fan engagement across channels requires a deep understanding of the various behaviors and traits of specific fans. No two fans are the same, but patterns can emerge to enhance League-provided cluster models with new team-specific segments.

“The first use case we tackled with Teradata was around building cluster models. We’ve seen the league do this successfully to understand different types of NFL fans, and we saw a similar opportunity to understand Giants fans,” says Scibetti.

Using ClearScape Analytics, data scientists and business analysts leverage teradataml for Python-based unsupervised machine learning (ML) K-means cluster models. Separate models group fans into multiple clusters across both season ticket members and non-season ticket members.

Season ticket members are grouped into three clusters based on primary ticketing, secondary ticketing, attendance, merchandise spend, email, demographics and digital behavior:

1. Premium experience 
2. Traditional die-hards 
3. On the rise 

Non-season ticket members are grouped into six clusters based on ticket and merchandise spend, attendance, age, email, sweepstakes, and NFL fan data:

1. Top of the market
2. Annual outings
3. Dabblers
4. Shopping-centric
5. Digital-first
6. Casual engagement

“Seeing the different types of behaviors, transactions, and engagements inside the clusters allows us to see how fan audiences vary. We might have one audience that's more fashion and merchandise-centric, where they skew much more toward buying merchandise on a regular basis, versus someone who buys a ticket once or twice every year. Seeing these different patterns, how someone is in one cluster, and what cluster they might grow into as they continue to engage with the team is really important information for us to understand our fans,” explains Scibetti.

ClearScape Analytics provides powerful, open, and connected ML capabilities. Using the in-database analytics provide flexibility for data scientists and analysts to use their preferred tools, speeding up time to insights and activating value.

“ClearScape Analytics has been great for us in terms of executing on our model building because of the flexible nature. We have the built-in capabilities with ClearScape Analytics. My background happens to be more in R, whereas my primary analyst’s background is more in Python. We can bring those models into VantageCloud Lake, test, and deploy really fast,” Scibetti elaborates.

Identifying games for which fans are most likely to purchase tickets

Having identified clusters for season and non-season ticket members, the New York Giants address ticket purchase behaviors on the primary and secondary ticketing market. With only 10 games played at home per regular season, personalizing messages based on fan clusters can help to drive additional ticket sales and attendance. 

ClearScape Analytics’ time-series and nPath analytics identify patterns and paths of various purchase and attendance combinations. 

“Understanding the fan journey of purchase to attendance means understanding the different patterns of when people buy tickets, which games they do and don’t attend, and whether they resell their ticket or buy from us,” describes Scibetti. “These help us target messaging, better predict attendance, and optimize pricing.”

Insights uncovered challenge common conceptions. 

“Even for a sport with only 10 home games, we identified more than 8,000 different combinations of purchase to attendance. There really is no one-size-fits-all fan journey,” says Scibetti.

The 8,000+ combinations mean the Giants no longer focus solely on a common “next game” mentality.

“Using these analytic techniques, we learned that for [individual game buyers] who attended the home opener, they were six times more likely to skip the next game and go to a game further on in the schedule. This is really important because we can use that to shape our marketing,” Scibetti shares.

Dynamic ticket pricing optimizes single-game pricing

It’s not enough to treat a game as a single product. Each game creates new variables that impact future games. Market demand for tickets is dependent on factors such as when the game is held in the season (e.g., early or later in the season), team performance, competitive matchups (e.g., division opponents or rivalries), and day and time of games (e.g., Sunday afternoon, Sunday night, Monday night, and Thursday night). 

“In sports, a common practice is variable pricing and dynamic pricing. Essentially, not every game is going to have the same demand. Understanding the different demand curves for our games is absolutely critical to have a good pricing strategy,” outlines Scibetti.

To address variable product (game) demand, demand curves are created in VantageCloud Lake. A demand curve groups historical games based on willingness to pay (WTP). Additional demand curves are created for different time periods—for example, May to August time periods versus full-season time periods.

“We make pricing changes throughout the season. When a team wins a couple of games, that can shift a demand curve pretty quickly. As the team’s doing better, demand goes up. However, that may only affect certain games. It’s not just about being able to cluster similar games together as part of the demand curves—it’s also identifying which games are most likely to shift from cluster to cluster as team performance changes,” details Scibetti.