Recommendation system
How to win Customer Loyalty with Big Data Technology?
Who are they, the heroes of our time? They are different, but they all have something in common - this is daily activity and regular sports. How to become a favorite fitness club for these people for a long time with the help of big data technology is told by David Melkumyan, director of development at World Class, and Sergey Marin, founder of Data Studio.
What do we love most about service? When we're treated individually. This increases both our loyalty and engagement, and the average check grows behind them. But to what extent is such an attitude possible if the service is streaming and is designed for a large number of clients? Big Data technologies allow us to study them and make the best customized offer for a specific person. World Class took advantage of this technique and received a positive result.

Sport has long been something more than a workout, it's a way of life. This idea was also confirmed by World Class studies. After analyzing its customer base, the company came to the conclusion that people who, in addition to a subscription, also buy other fitness club services, often return to renew their annual membership. Further, the company considered the structure of purchases and identified priority areas for proposals: group training, personal training, participation in club sports events and SPA-salon services. It was these services that were chosen by customers who renewed their sports membership.

Collaborative filtering in practice
The next step is big data. As a contractor for the development of the recommendation system, World Class attracted the Data Studio.

The team was tasked with creating a recommendation system based on the accumulated data on the use of additional fitness club services: SPA services, individual and group training, sports events.
Collaborative filtering was used in the construction of the recommender system. This approach is based on the fact that if one person likes products A, B, C, D, and another likes products B, C, D, E, then there is a high probability that the first person will also like product E, and the second one A.
In practice, this approach is implemented through matrix solving using Python libraries.

To solve the World Class problem, a giant matrix was built, each row of which is a client, and each column is a service from a fitness club. The Python library then ran the matrix factorization formula, resulting in client and service vectors. Further, in order to make an individual offer to the client, collaborative filtering compared the vector of the selected client with the vectors of other fitness club members, and found the most similar ones. As a result of the comparison, a person received as an offer services that were used by people with vectors similar to his.

But, of course, the project was not without its challenges. For a certain number of clients there was no information about the services used. Therefore, the system took into account not only past customer purchases, but also compared profiles in the personal account: gender, age, membership type, visits. Also, for a certain sample, the proposals were made randomly, and as a result of using the service, the system was retrained as a result.
From high-tech to real life
The recommendation system gives recommendations, but they still need to be conveyed to customers correctly. Various channels were used for the proposals: a personal account on the website, a mobile application, Wi-Fi in clubs, verbal recommendations in the fitness club space itself and at the reception desk. Negotiating with electronic channels turned out to be easier than with people.

From curious cases: the system recommended a hair coloring service for a man, the employees rejected the offer, and later it turned out that the man really resorted to hair coloring. So, in order to motivate the staff, the KPIs of the club employees were changed.

Now let's move on to numbers, a little dry, but pleasant. The consumption of value-added services increased by an average of 10% per customer. Outflow decreased by 14%. Most importantly, the club's clients who used the recommendations began to use 58% more services of the company as a whole.
Case was presented at Big Data Conference. The responses of the author of the column to questions from the audience that were asked during his speech at the conference are published below.
  • Question:
    What is NPS stands for?
    Net Promoter Score is a generally accepted customer satisfaction index that measures the willingness of a customer to recommend the company's services to other people on a ten-point scale
  • Question:
    The goal of the world class is to become a love brand? How long does it take to get to that goal in the fitness industry? What are the main qualities you need to have in order to become a love brand?
    We believe that we are talking about Lovemark here – it is a brand image aimed at creating an emotional connection with the client, increasing customer loyalty.

    We have, in fact, a whole range of activities that have already been implemented (a loyalty program, feedback, and so on, for the future - a customer journey map, the development of a mobile application, etc.). There are also quite a few media personalities, cultural figures, and professional athletes among World Class clients. All of them are a kind of reference point in the lifestyle and the role of sports.
  • Question:
    What parameters are used to determine the churn probability? Only the frequency of club visits?
    As in any machine learning model, we used a fairly large number of parameters, more than 100. The most significant of them are the frequency of using additional services, the presence of a personal trainer, the remaining number of days of freezing, the average duration of a visit, the consumption of spa services, the duration of the contract and the season. As you can see, the main significant predictors are behavioral and show the level of involvement in the club's services.
  • Question:
    How has LTV or its outlook changed? Or what are your goals for that?
    There is a year on the observation horizon (this is how long the recommendation system in World Class is currently working) LT has grown by more than 30% on those clients who accepted the recommendation
  • Question:
    How is outflow defined? No renewal immediately after subscription ends?
    Outflow - the absence of prolongation within 3 months. To train the model, a period of 1 month was used: that is, no prolongation for 1 month
  • Question:
    What communication channels are used by the results of the recommendation system? Only email newsletter?
    All active and passive World Class channels: reception, WiFi in clubs (when accessing WiFi, a targeted advertising banner is displayed), personal account, application and yes, email newsletters
I agree to the Data Usage Terms
+7 (967) 215-75-05