top of page

PathFinder: Harnessing Consumer Transportation and Environmental Data to Drive Business Decisions

The Rise of Micromobility

Think back to your college days, where you had 10 minutes to walk across campus to a class that typically takes significantly longer to get to. For the average student without accessibility to alternate forms of transportation, the options here tended to be either running to class and showing up sweaty or walking to class and showing up late – neither of which were ideal. Micromobility companies such as Bird, Lime, and VeoRide have set out to address issues such as these while revolutionizing transportation across college campuses and major cities alike by providing instant accessibility to eco-friendly electric scooters and e-bikes. Whether it is for getting somewhere on time, avoiding showing up sweaty, or simply just for fun, these platforms are changing the way we travel while providing an environmentally conscious solution in dense urban areas. As the popularity of these companies and services has continued to grow, so has the amount of incoming data generated by customers.

Every time a customer takes a ride on one of these platforms, companies can monitor a variety of information such as the vehicle type, battery levels, start location, end location, times of travel, and even continuous GPS monitoring to depict the path that was taken. Reliable GPS in this instance is especially crucial because if a platform is not returned to a designated charging station, employees must manually pick it up and return it to a station to ensure better accessibility and battery power for the next customer. With this collection of diverse data being generated by each individual ride, the insights and trends that can be developed both from a business and population oversight perspective are truly limitless when viewing ride data in aggregate. This is exactly what we aimed to accomplish through our new globally accessible tool: PathFinder.

Introduction to PathFinder

PathFinder provides a descriptive, predictive, and real-time analytics dashboard for deriving key insights from VeoRide’s trip data through College Park, Maryland. VeoRide is the University of Maryland (UMD) College Park’s primary micromobility partner, and as such they provided a dataset for inclusion as part of UMD’s 2021 Data Challenge via UMD’s Department of Transportation . This dataset consisted of individual ride data from October 2019 and October 2020, totaling approximately 40,000 rides. Cleaning this dataset of ride data with erroneous values yielded a final dataset of approximately 35,000 rides. This served as the data foundation for PathFinder. Therefore, while PathFinder is an independent Boulevard project, the foundational ride data is fully owned and operated by VeoRide. In addition to VeoRide’s initial data, PathFinder also accounts for associated weather data using World Weather Online’s weather data API to better understand the influence of environmental factors on ride patterns.

PathFinder Feature Overview

PathFinder is based in Python and Microsoft Power BI, and at a high-level. It empower users in solving both large-scale and small-scale business and population oversight problems rooted in resource allocation, inventory optimization, investment strategy, and population oversight by deploying advanced analytics and machine learning (ML). With direct applicability to traditional work streams of Data/Business Intelligence Analysts, Inventory/Product Managers, Ride Pickup Teams, and Law Enforcement, PathFinder paves the way for both, responsive and real-time decision-making ahead of time.

Descriptive Analytics

The dashboard provides users with the ability to construct customized complex logistical and weather-based queries, allowing the user to identify patterns and better understand the historical data from a variety of perspectives. By specifying query selections, an end user can immediately answer questions related to start/end locations of individual rides, aggregate pickup/drop-off statistics at locations, top ride paths, ride volume over time, utilization of scooters vs. e-bikes, average distance/speed/time of matching rides, and much more. By driving better understanding of ride density and other associated statistics, PathFinder’s descriptive capabilities will allow end-users to make optimal macro level decisions for optimizing inventory usage and maximizing profit.

Predictive Analytics

The PathFinder dashboard also includes a predictive analytics feature, using ML to predict utilization (number of rides) given temporal and environmental factors at all relevant locations. Rather than making responsive decisions, this feature allows for users to make calculated inventory decisions ahead of time to prepare for expected demand. By allocating inventory to accommodate expected demand prior to the need, end-users can maximize both platform availability and profit while dynamically accounting for the subtlest of temporal or environmental changes.

Real-Time Analytics for Anomaly Detection

Finally, PathFinder’s last major feature couples ML with synthetic data to deliver a real-time analytics capability for anomaly detection. This feature allows users to monitor pickup and drop-off rates in real time while being able to dynamically detect statistically uncharacteristic levels of either high or low activity. As a result, users can identify and understand rapid demand shifts in real-time, enabling dynamic planning and resource allocation while developing key understandings of root-causes for activity fluctuation. This ability to process large volume of data in real-time and detect anomalous behavior lends itself as a useful statistical tool for population oversight.

Case Study

Current data shows that the smallest proportion of ride counts occurs between Sunday and Wednesday. The limited number of rides during these days constricts the potential revenue stream for e-scooter/bicycle companies providing for College Park, Maryland residents. Proper resource allocation of electronic scooters and e-bikes would enable VeoRide to position their units in an optimal time and location to meet consumer demand and maximize revenue at cheaper overhead cost. This case study will explore how PathFinder can directly be used to develop a concrete inventory strategy for optimizing this allocation process Sunday-Wednesday by dynamically answering a variety of questions such as:

  1. Location of popular pickup/drop-off “hotspots”

  2. Frequent times of travel

  3. Expected customer demand over time at popular locations given weather forecasts

  4. Real-time customer demand

By using Pathfinder to filter the data to drill down on the rides between Sunday and Wednesday, the end-user can immediately see that, agnostic of weather factors, the most frequent pickup and drop-off locations occur at either primary on-campus class/dining areas or primary on/off-campus living areas as seen below:

  • Kirwin Hall / Glenn Martin Hall (Math and Engineering buildings, primary classrooms, North Campus)

  • LeFrak Hall / South Campus Dining / Point of Failure (primary classrooms and dining, South Campus)

  • Varsity / View Apartments (primary housing and shopping, adjacent to North Campus)

  • South Campus Commons (primary housing, South Campus)

  • CP Towers / CP Shopping Center / Leonardtown (primary housing and shopping, adjacent to South Campus)

PathFinder’s Descriptive Analytics, Filtered Sunday-Wednesday

PathFinder’s Descriptive Analytics Cont., Filtered Sunday-Wednesday

Further analysis of the “Top 10 Ride Paths” graphic indicates that not only are these locations the most frequent pickup and drop-off locations, but a vast majority of rides occur among these locations from start to finish. By interpreting the “Ride Counts by Hour” graphic as well, general ride activity/customer demand tends to increase sharply between 11AM – midnight while significantly dropping off outside of these hours. At this point, PathFinder’s descriptive analytics capabilities have successfully answered our first two questions (Case Study Questions 1 and 2). By identifying hotspot regions and frequent travel times, dedicated efforts to optimize processes among these locations during high demand hours will encompass a significant proportion of customer rides, increasing platform availability at lower costs. Due to the added insight that the most common paths typically occur among these hotpots, in the event of rapid vehicle reallocations, the company could reasonably target any of these locations for available vehicles because of the expectation of higher incoming customer traffic.

With targeted hotspots and high-volume times identified, the natural question becomes how much resources are needed at these locations over time to promote optimal processes and platform availability. Too few resources indicate missing out on potential customers while too many resources indicate over-investment and/or inefficient use of current resources. Therefore, the challenge becomes finding that balance. PathFinder’s Predicted Utilization feature answers this question directly by forecasting the expected number of outbound rides given location, time, and potential weather factors (Case Study Question 3) at user-defined levels of specificity. Being able to predict customer demand inherently shapes the resource requirements needed to meet this expectation, therefore this feature can be used to develop a specific numeric resource allocation strategy. In line with the theme of developing an optimal resource allocation strategy for Sunday-Wednesday, consider the scenario of using PathFinder to forecast the expected number of rides out of the Varsity / View apartments on a Wednesday given a specific weather report:

PathFinder’s Predicted Utilization Feature

PathFinder Generating a Resource Allocation Strategy

Not only does PathFinder develop a specific baseline forecast of exactly how many vehicles to allocate at the Varsity / View apartments location over time on this specific Wednesday, graphing the forecasted results above depicts a very similar ride volume pattern to the insights learned using PathFinder’s descriptive analytics capabilities.

Graphing PathFinder's Forecasted Utilization

The similarity between the forecasted results and the previous insights gained not only adds further confidence in the ML model’s accuracy, but also highlights the model’s ability to pick up on subtle factors and patterns such as the influence of time on ride volume and customer demand. By driving allocation strategy because of predicted demand, PathFinder enables business to develop strategies that have true balance while avoiding the consequences of having too few or too many resources at any given place at any given time.

While PathFinder’s Predicted Utilization capability provides a great baseline resource allocation strategy for vehicle placement ahead of time, sometimes reality does not always align with predicted patterns and outcomes. Whether the weather report abruptly changes, unexpected events take place in College Park, or other unforeseen circumstances, sometimes outside factors result in unpredictable outcomes. In response to potentially rapid shifts in consumer demand, PathFinder’s Real-Time Anomaly Detection feature allows the user to immediately identify uncharacteristically high or low amount of activity coming out of all relevant pickup or drop-off locations (Case Study Question 4). Through this feature, the user can actively monitor rapid demand shifts in real-time and make split-second decisions to accommodate these fluctuations if reality deviates from predicted patterns. To provide an example of this feature in action given the current case study, view PathFinder’s Real-Time Anomaly Detection graphic below, depicting the current pickup/drop-off rates on a given Wednesday.

PathFinder's Real-Time Analytics for Anomaly Detection

Quickly reviewing the results above indicates a pickup rate that is statistically significantly higher than average out of the Chapel Field / Ritchie / Frat Row (marked by 1). Likewise, the drop-off rate at Stamp Student Union (marked by 2) is also statistically slightly higher than average while the corresponding pickup rate out of this same location is statistically visually insignificant as indicated by the bubble size alone. If that there are not enough resources stationed at the Chapel Field / Ritchie / Frat Row location to accommodate this spike in outbound rides, then platforms can be re-allocated efficiently in real time from Stamp Student Union with a minimum expected consumer consequence.

Path Forward

Using advanced analytics and ML, PathFinder provides a showcase of leveraging geographic, temporal, and environmental data to drive business decisions rooted in a variety of problem sets such as resource allocation, inventory optimization, investment strategy, and population oversight. In its native state, PathFinder’s capabilities have a direct applicability to government and commercial transportation organizations, law enforcement, energy, community planning/development, and more. While PathFinder in its current state provides an incredibly impactful analytical platform, we aim to continue development and integrate new features soon to provide an even more robust solution. Things to look forward to include:

  • AI-based Resource Allocation: Built on top of PathFinder’s Predicted Utilization feature, automating this current capability will allow PathFinder to generate tables like what is shown in the case study above with minimal user interaction and faster resource allocation strategy development time.

  • Automated Logistics Planning: Currently, PathFinder can be used to dynamically develop specific numeric resource allocation strategies, so an automated logistics planning feature would take this one step further by also depicting the most efficient way to execute this strategy.

  • Financial Modeling: By including financial modeling capabilities, PathFinder will utilize this additional feature for consideration in developing resource allocation strategies that will aim to directly maximize revenue at the cheapest overhead cost. As a result, end users can achieve complete transparency of the fiscal implications for any specific strategy in question.

  • Maintenance and Repair Analysis: To reduce fleet maintenance costs, PathFinder will assess the factors (e.g., geographical location, time of day, severe weather) that increase likelihood of equipment failure or a negative rider experience (e.g., scooter not at the expected location, physical damage to the unit). Understanding these factors can allow end-users to predict and allocate resources to quickly address maintenance issues and develop strategies to prevent incidents from occurring. In addition, PathFinder will track repair costs at both the individual vehicle and regional level to provide understanding of overhead costs associated with equipment maintenance.

  • Enhanced Scalability: PathFinder is currently built primarily through Python and Microsoft Power BI, the latter of which is a commercial off-the-shelf (COTS) framework, alongside VeoRide’s historical ride data. While this does well to demonstrate the effectiveness of our algorithm, we are in the process of integrating this capability within a readily available (and downloadable!) solution-set for you to use to enhance your business decisions.

With these planned features in mind, PathFinder is intended to become a fully automated and comprehensive analytical strategy development tool that bridges the gap between optimal resource allocation, logistics, and cost. Whether it is identifying macro level patterns, resource allocation ahead of time, or demand shifts in real-time, PathFinder in current and future state will empower organizations to make the right decisions consistently, efficiently, and with complete transparency on financial impact. We at the Boulevard Consulting Group pride ourselves on providing not only the technical expertise to utilize the latest in ML and analytics techniques, but also the strategy and operations experience to craft solutions customized and optimized for each client.


Join us as we explore the future of business through the lens of machine learning & AI, data-driven analytics, cloud computing, and operational transformation.





bottom of page