Consumers are progressively adopting healthy and wellness-related practices. As a result, companies operating in the fitness centers business now see their customers base grow at rates never seen before. However, managers are also conscious of the volatility of consumers’ desires, particularly those signing up for this kind of services. This leads to a high churn rate.
For this reason, any strategies geared towards understanding and successfully managing the business become a significant competitive advantage. In this context, an analytics exercise was done using data from a chain of fitness centers in Madrid (Spain). Obtained results are presented next.
The ultimate goal of this exercise was to answer key questions related to the fitness centers business. The first question was: how my current customers behave, and what factors influence that behavior? The second question was: is it possible to predict that a customer will churn, in order to convince him or her to stay before the decssion is made?
With this purpose, a dataset of 1.547 customers of one fitness center in Madrid was analyzed. Features included demographic, membership, and attendance details.
Understanding how different types of customers behave is the first step in making decissions to better serve them. The first step in our exercise was to create a visual representation of the dataset, in order to quickly spot behavioral aspects and customer preferences.
In figure 1, descriptive statistics related to customer demographics are depicted. From those, it turns out that this fitness center is more successful among young people, and that gender doesn’t make a difference. Also, no clear pattern seems to exist when plotting customer sign-ups and churns; however, looking at the number of active weekly clients, the trend is clearly descending, with just a slight spike at the end of the year.
Lastly, attendance significantly drops along summer holidays. That might ring a bell: launching a special offer in that period will stabilize attendance and help generate more revenue.
Fig. 1. Demographic, sign-ups, churns and attendance details.
In figure 2, attendance metrics are displayed. Peak and valley hours are quickly spotted, as well as how attendance is distributed over the week (decreasing from Mon to Sat). Also, most active customers are those in the range of 20 to 30 years, and 40 to 50 years, and time spent in the center ranges from 45 to 75 minutes.
Also, the visualization is interactive: a number of parameters can be set (age, gender, type of membership) to check how different customer profiles behave. By doing so, it become pretty obvious how membership features as term contract, prepayment, duration or group discounts relate to customer behavior.
Fig. 2. Members attendance indicators.
Lastly, membership features were compared against demographic details, in order to understand what features are more compelling for each customer profile. These insghts are useful to design customized marketing campaigns, in which the prospect is offered the type of membership that will likely convince him or her to sign up.
Fig. 3. Membership features vs. Demographic profiles.
The next step was to build a predictive model capable of detecting what customers would churn soon. With this information, it would be possible to contact those customers before they churn, identify the causes for their dissatisfaction, and make them an offer to stay.
This model’s results were the top 50 customers more likely to churn next month. To assess model performance, it was measured how many of those customers did actually churn, and how many of the actual churns were identified by the model. Data from 2015 was used to predict churns in January 2016; then, data from January 2016 was used to re-train the model and predict churns in February 2016. And the same approach was followed to predict churns in March 2016.
Table 1 shows results for the first three months. Key predictors were attendance frequency, attendance duration and membership features.
The following table shows the obtained results along 3 months.
Table 1. Model performance.
LDA (Linear Discriminant Analysis) was the classification algorithm that consistenly showed higher accuracy among several techniques (logistic regression, QDA, KNN, Naive Bayes). Figure 4 shows the model parameters.
Fig. 4. Churn model in R.
This exercise attempted to prove that Big Data Analytics can shed light on business improvements to successfully manage fitness centers by mining data that can be easily acquired and processed.
Both methodology and data had a number of constraints: number of observations might be too low for inference, the sample is biased (one fitness center in a location of high-income citizens), and the model accuracy might improve with more advanced techniques.
Turning Analytics into business-as-usual for fitness center management is a great step towards replacing intuition with fact-based decissions. Under this paradigm, managers can spot unsatisfied customers before they churn, and try to convince them to stay. They can also create custom offers for potential customers, based on what similar customers chose in the past.
However, the possibilities go beyond the aforementioned scenarios.
With accurate and up-to-date information about customer profiles and their behavior, fitness centers can define new business strategies geared towards improving efficiency and performance, design new product features or anticipate whether a new fitness center will be profitable.
Big data analytics has opened the door, and fitness center managers will decide how much they want to leverage this tool to maximize their success in business.
Rafael San Miguel Carrasco
Data Analyst, University of Malaga
Miguel Santana Barroso
Fitness Specialist, Infinit Fitness