Divvy Bike-Share Success

Divvy Bike-Share Success Thumbnail

Inspiration

As an avid cycling, my project is inspired by the real-world challenge faced by Divvy, a Chicago's bike-sharing under Chicago Department of Transportation (CDOT). As urban mobility continues to evolve, bike-sharing has emerged as a sustainable and efficient transportation option. However, the company's financial success depends on converting casual riders into committed annual members.

After analyzing Divvy's business model, I identified a critical insight: annual members are significantly more profitable than casual riders. This sparked our curiosity to understand what distinguishes these two user groups and how data could reveal the path to conversion.

The project's mission aligned perfectly with my passion for sustainability, urban planning, and data-driven decision-making related to cyclists. By helping Divvy increase membership conversion, we could contribute to reducing carbon emissions, alleviating traffic congestion, and promoting healthier lifestyles in Chicago.

What it does

Divvy Bike-Share Success is a comprehensive data analytics solution that transforms raw bike-share trip data into actionable marketing strategies. Dashboard ftw!

How we built it

The approach followed a structured data analysis framework:

Data Collection & Preparation:
- Acquired 12 months of historical trip data from Divvy's public datasets (incl 2023 & 2024)
- Cleaned and preprocessed over 5.4 million ride records
- Created a unified database with consistent schema across time periods
- Enriched data with weather information, public events, and geographical data
Exploratory Data Analysis:
- Analyzed temporal patterns (time of day, day of week, seasonality)
- Mapped geographical usage patterns and popular routes
- Identified demographic correlations where available
- Built rider profiles based on usage frequency, duration, and patterns
Statistical Analysis & Modeling:
- Developed predictive models to identify casual riders with high conversion potential
- Created segmentation clusters of rider types
- Validated findings with hypothesis testing
- Quantified potential ROI for different marketing approaches
Infrastructure Development:
- Built ETL pipelines for continuous data integration
- Deployed dashboard for real-time monitoring of KPIs
- Created reproducible analysis workflows
Marketing Strategy Formulation:
- Designed targeted campaigns based on rider segments
- Created personalized messaging frameworks
- Established metrics for campaign effectiveness measurement

Challenges we ran into

The journey wasn't without obstacles:

Data Quality Issues: Missing values and discrepancies required developing robust data cleaning procedures.
Scalability Concerns: Processing millions of ride records demanded optimization of the analysis pipeline. I implemented parallelization techniques and cloud computing solutions to handle the data volume efficiently.
Privacy Considerations: While working with user behavior data, I needed to ensure compliance with privacy regulations and ethical standards, leading to implement anonymization techniques for PII.
Feature Engineering Complexity: Creating meaningful features from raw ride data that captured user intent and behavior patterns was challenging. I iterated through multiple approaches before identifying the most predictive variables.
Cross-functional Communication: Translating complex analytical findings into actionable business recommendations required developing clear visualization methods and narrative techniques.
Time-constraint: Obviously, the recurring theme in a hackathon is time-constraint. I'm not sure if I'll be able to fully implement all features I have planned within the time limit. I'd be happy if I get to implement 50% of them within this hackathon and will continue the rest afterward due to the complexity and comprehensive features for an end-to-end data engineering pipeline.

Accomplishments that we're proud of

What we learned

This project was a journey through the complete data analysis lifecycle, offering invaluable lessons:

Data Integration Challenges: Working with diverse datasets from different systems taught me the importance of data normalization and validation. I discovered that integrating temporal, geospatial, and user behavior data requires careful planning and robust ETL processes.
Behavioral Pattern Recognition: I learned to identify subtle patterns in rider behavior that strongly correlate with conversion potential. For example, I found that casual riders who use bikes during commuting hours (07:00 - 09:00) and (16:00 - 18:00) for 3+ consecutive weekdays show a 62% higher likelihood of converting to membership.
Geospatial Analysis Insights: The spatial distribution of rides revealed distinct usage patterns between casual riders and members. This taught me techniques for effective geospatial visualization and analysis.
Statistical Rigor: I learned to apply appropriate statistical methods to ensure our findings were significant and actionable:

$$\chi^2 = \sum_{i=1}^{n} \frac{(O_i - E_i)^2}{E_i}$$

Where I tested the independence of rider categories and behavior variables.

Marketing Strategy Development: Translating data insights into actionable marketing strategies required cross-functional thinking, teaching me to bridge the gap between data science and business strategy. ## What's next for Divvy Bike-Share Success Oh well, first of all, I want to complete the full implementation of feature engineering initially planned. Only then I'll think about further extending its features such as:
Development of a recommendation API
Expansion of the analysis to include socioeconomic and demographic data to identify underserved communities and develop equity-focused membership programs.
Implementation of advanced time series forecasting
Creation of a machine learning pipeline that continuously improves conversion predictions as new data becomes available.
Design of A/B testing frameworks to systematically evaluate different marketing messages and incentives.