Machine Learning 1: Segmentation Report
0 likes119 views
The 2017 customer segmentation report analyzes approximately 2,000 trip reservations to identify three customer clusters based on their travel behaviors and demographics. These clusters include young professionals, young and married passengers, and family-life passengers, each with unique characteristics and preferences that inform tailored marketing strategies aimed at maximizing customer retention and revenue. The report emphasizes the importance of understanding each segment's needs and timing to enhance loyalty and promote upselling opportunities.
Machine Learning 1: Segmentation Report
- 1. Presented by: Bruno Cervantes Quino Bruno Gobbet Gianini Federico J. Garcia Lopez George Kofi Akanza Kamal Nandan Melody Ucros Peder Viland OUR CUSTOMER CLUSTERS SEGMENTATION REPORT 2017 To: Marketing Director
- 2. Executive Summary…….….…………….…………….…………….…….…. 3 Analysis Design ……………………………..…….………………………...... 4 Data Selection …………………………...………….…….………......4 Data Manipulation ………………………………….…….………...... 4 Segmentation …………….…………….…...……….………………. 4 Analysis Results ...……….…………….……………….………..……….…... 5 Customer Clusters……………………………………...……………. 5 Visual Representation………………………..………...……………. 5 Analysis Interpretation.………….…….……………..…………….…………. 6 Young Professionals…………………………………………………..6 Young & Married Passengers ……………………………………….6 Family-Life Passengers ………………………………………………7 Conclusion……………………………………………………………...7 Annex …………………….…………….…….……….……………..………… 8 TABLE OF CONTENTS Segmentation Report 2017 2
- 3. EXECUTIVE SUMMARY In order to create a marketing plan that yields maximum ROI in 2018, we were requested to run a segmentation analysis on our existing customer base. The data analyzed is a sample of trip reservations containing approximately two thousand observations. It shows the departure date, return date, purchase date, if the people had children, if they had a car, if they booked a hotel during their trip, if they had insurance, their gender, age, the number of seats and finally the customer type that shows whether they were registered customers or not. After careful data manipulation and algorithm selection, we found 3 clusters: April-May Sept-Dec • Travel April-May • Early 30’s • 3 day trips • 10% with children • Buy 10 days before • 70% on Weekdays • Travel Sept-Dec • Early 30’s • 4 day trips • 17% with children • Buy 8 days before • 24% on Weekends • Travel July-Aug • Late 30’s • 10 day trips • 35% with children • Buy 25 days before • 5% on Weekends July-Aug Although the first two have similar age and travel habits, each customer cluster has a very clear label that describes them. The third segment represents the oldest and most consistent customers; the most profitable and loyal ones. It is safe to say that within a couple of years one cluster replenishes the other, making their retention a key component of future revenues. The marketing initiatives suggested on these report will focus on the retention and growth of the existing customer base. By tailoring our offers to their individual needs, our goal is to increase customer loyalty, referrals and trip frequency. At the same time, we discovered who to upsell, when, how, and with what, in order to increase our annual revenues. Segmentation Report 2017 3 Young Professionals Young & Married Families
- 4. ANALYSIS DESIGN Data Selection The data analyzed is a sample of trip reservations containing approximately two thousand observations. It shows the departure date, return date, purchase date, if the people had children, if they had a car, if they booked a hotel during their trip, if they had insurance, their gender, age, the number of seats and finally the customer type that shows whether they were registered customers or not. Data Manipulation Before we could start working on the segmentation, some date manipulation was required. Data handling is critical since every decision we make while working with the data will impact our model’s result. As a part of the data handling we: parsed and formatted all the dates; changed measurement types to Boolean for the fields children, car, hotel and insurance; changed the gender field and recoded as Female=0 and Male=1; checked for outliers for age and removed row using (1.5 Inter Quartile Ranges). We also created the following features: trip duration in days; month and day of departure; summer (traveled between July and August); weekend; Christmas; purchase lag (which shows the time between the purchase date and the departure date); alone labor, that shows if someone took just one seat for Labor Day; flag/dummy variables for the three categories of customer type. We binned the variables age, duration and lag using quartiles as binning criteria. We produced standardized versions and generated mathematical transformations of the variables age, duration and lag. We “unfolded” first customer type and then replaced the missing values with “0”. Finally, we recoded customer type into a new flag variable called “Registered” considering those registered customers (either first time or regular). Segmentation For the segmentation, we used the K-means clustering algorithm. For the number of clusters (algorithm) we chose 3, since selecting 4 or 5 would result in clusters with similar attributes that would eventually need to be merged. As for the sampling, since out data wasn’t that large to begin with we used the Dataiku Default (First Records, 100.000), that is, we were able to analyze the whole dataset. For Outliers, other than the detection we had previously made on data manipulation we followed the same steps we did in class and chose not to detect any outliers. We tried using PCA analysis with 10% and even though we got interesting results, we eventually agreed that would be better not to use it since we didn’t cover it in class. When we did use it, we could clearly see a boost of Silhouette performance. As for the features, after testing with several different variables we came to the conclusion that departure date, purchase date, purchase lag, duration and age were the ones that would result in a better segmentation allowing us to prepare an optimal advertisement plan. Segmentation Report 2017 4
- 5. ANALYSIS RESULTS Customer Clusters The first cluster (cluster 0, approximately 28% of the whole dataset) consist of people in their early 30’s (average age 30.71) who travel for an average of 3.62 days, and buy their tickets an average of 8.10 in advance. This group travels between April and May, and only 10% travels with children. They usually travel on week days too (70%). The second cluster (cluster 1, approximately 36% of the whole dataset) are very much alike the first cluster with one major difference: they usually travel between September and December. They are on average 31.51 years old;, their trip last an average of 4.3 days, and they buy their tickets an average of 7.79 days in advance. Only 24% travels on weekends and 17% travel with children. The third cluster (cluster 2, approximately 36% of the whole dataset) consist of people who are older, compared to the first two groups. They are 36.63 years old on average, travel for an average of 10.61 days, and 99% of them travel between July and August. They usually plan ahead and buy their tickets an average of 25.9 days in advance. 35% of them travel with children and only 5% travel on weekends. Visual Representation On graph #1, you can clearly notice that the cluster 0 travels in the beginning of the year for a short period, while the cluster 2 travels in the summer for a larger period . Cluster 1 travels after summer until December for a short amount of time. On graph #2, it is possible to notice that the longer the trip, the farther in advance they tickets are bought, and it is also possible to notice that cluster 2 is really the only one who plans in advance. Graph #1 Graph #2 Segmentation Report 2017 5
- 6. ANALYSIS INTERPRETATION Young Professionals (Cluster 0) This segment is to be viewed as a “replenishing batch” for cluster 1 members who would move to cluster 2 in a year or two, since age gap is almost non existent. Therefore, we need to adopt a pricing strategy and customer-service approach that helps to draw customers in, feed the evolution cycle, and secure future revenues. As the youngest of the three clusters and buyers of basic economy tickets, they are likely the most price- sensitive passengers and initially make purchasing decisions solely on price. They’re not likely to carry above-weight luggage since their travel duration tends to be shortest among three groups, have no kids, and probably single. These people have the luxury of time and would likely prefer to scout for tourist sites and hotel by themselves since they’re themselves curious and adventurous. Once we know exactly what they want, we may attempt to sell packages to them though the success rate may not be that high. They may probably be saving towards marriage, getting a suitable home for their future families, among others. Besides, since a huge chunk of their leisure expenditure budget may already be hard-pressed due to regular after work and weekend outings, they’d likely want to save on travel cost. Any attempt to upsell to them before and or on the day of purchase may not succeed. It is recommended therefore that upsells be considered at check-in instead when there’re still spaces in the premium sections of the planes, as they may just suddenly realize the restrictions that come with basic economy and be enticed to upgrade. Young & Married (Cluster 1) This segment appears not to be exactly different from Young Professionals based on age and purchase lag, however this is far from reality. With children still very young, they will obviously carry more load and travel on different dates. With the synergy of combined couple wealth, they are mostly likely to have a lot more disposable income and would be interested in upscale services to match their newly acquired status as couples or good income earners. Therefore, they would be open to any offer that affords security, convenience, ease and reliability. In view of this, the company should offer a full-service package via its website. Services may include extra luggage, more comfortable seating (premium), additional inflight services, accommodation, car rentals, tourist places, among others. The company could then add an additional mark up on prices. Promotional activities should however precede the season they start buying and traveling so that we could present ourselves as a viable option during the pre-decision stage for consideration and subsequent subscription. This segment would likely migrate / upgrade into cluster 2 as their earning power continues to improve and settle into marriage after a while . Segmentation Report 2017 6
- 7. ANALYSIS INTERPRETATION Family-Life Passenger (Cluster 2) Higher prices and cross selling opportunities are more viable with this group of customers, since they are more mature and have more disposable income. They are more settled in life; marriage, job, children, among others. Most likely at this stage, they would be a few steps away from occupying relatively high level managerial roles in their companies and can afford a lot more luxury than cluster 1. Children would obviously be more mature with varied needs and preferences both inflight ad off flight. Due to the added complexity of moving with two or more kids and the fact that they tend to travel during summer and for longer periods means that they would be more open to any form of prestige or premium service that would reduce their stress levels considerable. Even more, due to high demands on them due to their roles in their work places, they would relish not having to sweat over choice of flight tickets, destinations car rentals, among others. Upgraded service offering above that offered to Cluster 1 would afford the company the chance to exploit the upselling potential here. Besides, these customers may have been with the company for an extended period and would have become more trusting of our recommendations and happy with our services. This would give us leverage and allow us to succeed with our upselling strategy. It is instructive to mention here that we are by no means insinuating that all customers in this and or other clusters have necessarily been with the company for so long. It is just a possibility. It would be crucial to keep these people happy and loyal in order to derive maximum benefit from them. Additionally, seeing that summer times are when they travel most, it would provide good basis for further price discrimination by suggesting premium service tickets or packages to them using algorithms and other approaches. Again, work out discounts with partner service providers and also add a mark-up figure for this extra services provided. Conclusion When trying to decide how to target which cluster, keep in mind the reason why they are buying from you and when they are doing so. Young Professionals, just want the cheapest and simplest option that gets them where they need to be. Young and Married Passengers, want an affordable option of getting where they need to go, but also one that offers convenience. Family-Life Passengers are more focused on how comfortable they can be for their annual getaway, and traveling preferences reflect on their social status. Each group, ideally, replenishes the other. Therefore, each customer must be treated well at the stage they are in, so that they grow with us as their income does. Segmentation Report 2017 7
- 8. APPENDIX Summary Segmentation Report 2017 8 Purchase Lag Distribution Duration Distribution Cluster 0 Purchase Data Distribution
- 9. APPENDIX Age Distribution Segmentation Report 2017 9 Departure Date Distribution Cluster 1 Summary Purchase Lag Distribution
- 10. APPENDIX Segmentation Report 2017 10 Duration Distribution Purchase Date Distribution Age Distribution Departure Date Distribution
- 11. APPENDIX Summary Segmentation Report 2017 11 Purchase Lag Distribution Duration Distribution Cluster 2 Purchase Data Distribution
- 12. APPENDIX Age Distribution Segmentation Report 2017 12 Departure Date Distribution OTHER Variable Importance
- 13. APPENDIX HeatMap Segmentation Report 2017 13 Metrics