Recently, Turnkey invested in IBM SPSS Modeler, an advanced modeling platform that enables us to build more robust and advanced models than previously possible. Turnkey’s Prospector team uses this platform to construct CHAID Decision Tree Models. These models utilize ticketing data from teams’ ticketing partners (Ticketmaster, Tickets.com, Salesforce, and/or Veritix), as well as personal and household data from our partners at Acxiom, as inputs.
Before creating a scoring model, the Turnkey Prospector team works hand in hand with our clients to determine which ticket plans are sales priorities (usually full and/or partial season tickets). Then, we isolate the current buyers of those plans using event and price codes. These segments of buyers became our “target” groups to compare against single game buyers. This process of comparison enables us to understand exactly what makes the target group(s) unique.
After completing this analysis, we build a series of tree-models. Visually, the results of this type of modeling are akin to a family tree, branching and splitting on the distinct criteria needed to differentiate between target buyers and single game buyers. For instance, these models will often split on different age groups, showing the age ranges most likely to purchase a full season plan (decreasing in likelihood as we look from left to right). We build ten or more of these trees simultaneously; then, we blend them together to create an “ensemble” model, ensuring that the model we deliver to teams is a stable and predictive representation of that team’s buyers.
After running each prospect through the ensemble model, a “raw propensity” score is generated for each lead. This score represents the likelihood that that lead will be a buyer of a particular ticket plan. We then create star rating buckets based on the percentile breakdowns of the raw propensity scores, and each prospect is sent to one of these buckets and labeled as a 1-, 2-, 3-, 4-, or 5-star lead.
To ensure that our models are highly predictive, we initially hold back a portion of each team’s current planholder list; then, after a team’s model is built, we use this list to test against the model. This process helps us determine how well the model performs with records it has never seen before. Below is the result of the same NHL client’s test data being run through the ensemble model. As you can see, this model did a great job of identifying buyers.
5 Star – 90.5%
4 Star – 4.2%
3 Star – 3.7%
2 Star – 1.4%
1 Star – 0.2%
Unscorable – 0.0%
The process outlined above allows us at Turnkey to build our models in an effective and powerful new way, delivering scores to teams and their reps overnight. We hope this makes the process a bit more understandable. Should you have any questions feel free to call us at 856-685-1450 – we’d be happy to explain this process further at any time.