Abstract

Walt Disney World uses analytics and optimization models to assist in the day-to-
day operations of its theme parks and marketing. The company uses these models
to determine things like park attendance, ride wait times, cast member schedules,
demand for costume supplies, and vacation package offerings. I propose using 3
or 4 analytical models in order to help the company advertise a FASTPASS ticket
to a particular customer. A FASTPASS ticket allows guests to skip the line at a ride
ahead of others who are waiting in the regular line to ride.

By getting historical hourly data on park attendance, day of the week, day of the
month, day of the year, whether the day falls on a holiday/special occasion,
average temperature, and total rainfall, the company can use a time-series model
to forecast park attendance for a particular day at a particular hour. This
forecasted data can then be used along with data on ride attendance, ride
demand, ride capacity, ride length, ride average rating, and the number of
FASTPASS riders in another time-series model to determine forecasted wait times
for a ride (both regular line and FASTPASS line wait times). Finally, with that data
along with a guest’s age, gender, family/party size, family/party ages, the number
of days of attendance at a park, and location, the company can use a classification
model to determine if a sales representative should spend some time advertising
the FASTPASS tickets to a guest upon their arrival.

Inspiration: https://www.informs.org/Impact/O.R.-Analytics-Success-
Stories/Industry-Profiles/Enhancing-the-Guest-Experience

Model #1: Forecasting Park Attendance

The first model involves forecasting park attendance at an hourly interval. For
simplicity, we will be using the model for just one theme park owned by Disney
and assume that the park is open at the same hours each day. The company
already has historical park attendance and other data useful for this model.
Getting historical data, we have variables like the day of the week, the day of the
month, the day of the year, the hour of the day, and whether it’s a holiday or a
special occasion. We can also get variables like average temperature and total
rainfall. Since park attendance can be quite cyclical and seasonal, we can use a
Holt-Winters exponential smoothing time-series model to predict future park
attendance hour by hour on a particular day.

We can assume park attendance is likely to grow throughout the day and die
down as closing time approaches. It’s likely to be higher on the weekends than on
the weekdays. Park attendance could be higher or lower on a holiday/special
occasion depending on the holiday or special occasion itself, which is why having
the day of the year data is also valuable. Warmer and drier days are also likely to
lead to higher park attendance.

Model #2: Forecasting Hourly Average Wait Times Per Ride

Now that we have a model that forecasts the park attendance in a given hour, we
can get the output of that model along with historical data on the number of
riders for every ride, calculate the ride’s demand (the percentage of guests at the
park that hour that got on the ride), the ride’s capacity, the ride’s length, the
ride’s average rating, and the number of FASTPASS riders. Using that data in
another time-series model, we can predict the average wait time for every ride.
Instead of forecasting just one average wait time, we can forecast an average wait
time for the FASTPASS lane and an average wait time for the regular line.

The ride’s capacity, length, and popularity/demand play a direct and highly
correlated factor in determining a ride’s average wait time. And by having two
different wait times per line can be useful information for the sales representative
to use to convince a customer in purchasing a FASTPASS ticket upon arrival at the
park.

Model #3/4: Classifying a Customer for Potential FASTPASS Promotion

Now that through our analysis and previous models we can forecast park
attendance and average wait times for regular and FASTPASS lines for every ride,
we can get additional data like the guest’s age, gender, family/party size, the ages
of accompanying guests, how many days of tickets the guest has purchased, and
the guest’s home location. We can use a support vector machine classification
model to predict if a guest is a good candidate for the sales representative to
push a FASTPASS promotion on along with providing the guest with information
on average wait times for popular rides to entice them further.
Guests of a particular age, gender, and family/party size are likely to ride certain
rides over and more than others. To get even more advanced and creative, we
could utilize a clustering model or recommendation system to predict which rides
the guests are likely to ride/be interested in riding, especially if we have historical
information on the particular guest’s previous ride history. A guest’s home
location can also be useful as out-of-state guests are more likely to purchase a
FASTPASS to make use of their limited time and relax during their vacation. The
number of days a guest plans on visiting the park can also be useful as a guest
who plans on spending several days at a park is less likely to be in a rush.

Model Frequency/Ease of Collecting Data

Since most of these models are time series models and data is changing
constantly, it makes sense to run these models every hour of park operation. As
for the classification models, they can be run far less frequently since they’re not
time series models and the variables don’t change that drastically.

Most of the variables mentioned are easy to collect through ticket information,
weather, surveys, and customer information at ticket purchase. In reality, Disney
most likely has far more variables/data collected considering their use of
technology and location tracking of guests which can add to the complexity and
maybe accuracy of models.