Apr 13, 2020
Case Study

How can you overcome distribution challenges as a bycicle sharing system?

Maximilian Zachow
Apr 13, 2020
Case Study

How can you overcome distribution challenges as a bycicle sharing system?

Maximilian Zachow

When I left the house to grab a beer with some friends in the evening of  September 29th, I saw myself in an unexpected situation. I usually take advantage of Madrid’s rich offer of mobility sharing systems to go places, but this time, the availability of the scooters was completely shifted. The majority of them were clustered around the Estadio Santiago Bernabéu waiting for people who were watching the local football match between Real Madrid and Atletico. People like me in other areas of Madrid faced a shortage of available vehicles. I thought that many shared mobility providers probably face this kind of distribution challenge and that it could be interesting to analyze the vehicle distribution data with Graphext.

STEP 1: WHAT ARE WE ANALYZING?

With Graphext we analyzed the vehicle distribution data (from Kaggle) of a bike sharing service based in Minnesota. One dataset contains information about every trip (duration, user type, date, start station, end station, etc.), the other one about the bike stations (location, number of docks, etc.). We created a project in which each node represents a station and each link represents trips between two stations. Stations that share many trips together form the clusters you can see in the image. We can identify exactly five of these communities in our network graph. Let’s have a look on how well the bike sharing provider maintains the distribution of its bikes..

STEP 2: DOCKS VS. DEMAND

In order to check how well the provider maintains its stations with an appropriate amount of docks, we set a filter for the stations with the largest capacity ranging from 25 to up to 40 docks, and plotted them against the number of trips completed. We found that the largestations are also the most demanded ones (The blue bars represent the distribution of our selection of the biggest stations and the grey bars show the distribution of the whole dataset also including the small stations). At a first glance, it seems that the provider is well aware of its distribution challenge and always meets the customer’s demand.  

But are all areas with a high demand already sufficiently covered? To answer that question we created a selection of small stations with no more than 25 docks, and used a heat map to show the number of trips between these stations. The more yellow the node is, the more demand there is, the more purple the less. We would have expected to only find purple nodes in our selection because we filtered for small stations only. However, we can see in the graph that there are at least five nodes marked yellow. To be clear, that means that at these stations the capacity is completely maxed out. The provider should consider to increase the amount of available docks and bikes. Otherwise it won’t meet a raising demand and waste potential revenue, or even worse, in these areas the company can lose customers to its competitors.

STEP 3: SEASONALITY

Another aspect that we need to consider in our analysis is the existence of seasonality in demand. The aggregation variable number of total trips doesn’t say anything about the change in demand during special events. It can make sense to manipulate the distribution of bikes over the city by manually transporting them from one area to another during particular daytimes or weekday such as when a big event like a local football match takes place. Fortunately, Graphext can help to detect the moments and locations for which these manipulations should be put into action.

For example, we suspected that there are differences between weekdays and weekends. To investigate, we created a collection of the stations that have most of their demands during the weekends and another one of stations that are in general very popular, regardless of the time. We can plot those two collections against each other and compare differences for each variable. Orange represents the most demanded stations over the weekend,  green the collection of the overall most demanded stations. We can directly see that there is a major difference between the collections in their relative distributions for the number of docks. Apparently, the demand for bikes varies based on the locations of the city throughout the week. Most of the stations that are mainly used during the weekend only have very few docks: 85% of the weekend’s most popular stations have 20 docks or less. Whereas generally the most demanded stations have between 20 to 28 docks. Thus, one conclusion the bike sharing company could draw is to temporary expand the capacity of some stations during the weekend. Perhaps it needs to manually transport some bikes on Fridays from one area to another to be prepared for the initial increase in demand on Saturday morning.

STEP 4: FURTHER USE CASES

With our analysis, we covered several challenges a bike sharing provider can face regarding vehicle distribution. However, there are many more issues to address. For example, to detect suitable times for promotion campaigns we could investigate the different behaviors of one-time users and frequent customers. If there are certain stations that are – at certain weekdays – more favored by one-time users than others, the provider can consider to offer discounts for subscriptions or bonus packages if you purchase minutes in advance. As you can see, the actual impact of Graphext on your business’ performance also depends on your individual use case. We demonstrated in this issue how you can use Graphext and its feature variety to develop business strategy or to identify sources of missed revenue.

If you like what you read and think Graphext can also help your company to improve its decision-making process, feel free to ask us for a product demo. We are also always happy to receive any kind of feedback or ideas about future data analysis projects with Graphext.

Want to learn more? Request a demo!

Want to do something like this?

Start now

Other stories

Las 2 Españas a Palos de Golf: quiénes y cómo son los que reparten tweets a izquierda y derecha.

La semana pasada, millones de personas en España acabaron imaginando en su cabeza algo que realmente nunca pasó: a un señor rico del barrio de Salamanca destrozando mobilario urbano con un palo de golf. Analizamos con datos y Graphext cómo se originó este bulo, y sobre todo quién hay detrás a izquierda y derecha propagando estas nuevas narrativas políticas

see more >

Las cifras de muertos en España por COVID-19 en contexto.

¿Quiénes son y dónde están los excesos de muertes en España en los últimos años por edad, sexo y Comunidades Autónomas?

see more >

How to look good on video calls: analyzing 1K rated Skype & Zoom rooms. COVID-19 Voyeurism.

Collecting 1000 thounsand screenshots from video calls on TV to find patterns of what makes you look good or bad

see more >

Clustering countries by Covid-19 Mobility Trends changes

We used the Apple data to cluster with Graphext countries that are experiencing similar behaviour of changes in people walking outside: 🔴Italy, Spain and other suddenly decrease by more than 90% 🟠Nordic, Germany slowly going back to the streets. 🟢Slowly reducing to -50% since February

see more >

Ready to try Graphext?

Want to know more about our smart data discovery tool?
Ask us for a demo!

Start now