I walked outside one morning and a scooter was standing on the sidewalk. It was the first time I had seen one in front of my house but I didn't think much of it. The next morning, it was still there. And the next. After about five days it was gone. I don't know if someone used it, or if the company moved it somewhere else to see if it would get used there.

This got me wondering: what led the company to place a scooter (or bike) there? It didn't strike me as an optimal location. So I set out to determine what those optimal locations were and what the characteristics of those locations are. This can be used as a building block for:

• Optimizing exising scooter operations in Washington DC
• Identifying expansion opportunities into new cities based on each city's specific characteristics
• Adjusting service in response to COVID-19 by reducing focus on tourist areas and shifting to the next-best locations

Specifically, I sought to identify locations that had high minutes of usage throughtout the day based on the scooter's location at 6:00AM. Why 6:00AM? If a scooter company needs to recharge and/or relocate a scooter, it's probably best to do it overnight to keep it available for use throughout the following day. By "high minutes of usage", I mean scooters that have above-average use. I'm not interested in locations that have a lot of scooters that are used for 5 minutes a day, but am interested in locations that have a lot of scooters that are used 60 minutes a day.

The map below divides the city into 100m x 100m squares. Each square is color coded based on the average daily minutes of usage originating in that square.

While interesting to look at, the map doesn't convey much to those not familiar with DC and its layout. For example, there are two horizontal lines with concentrated high usage just south of the White House. These border the National Mall which houses many of the Smithsonian's museums and popular monuments. Scooters are prohibited to be dropped off within the Mall, which means the companies place them across the street for tourists to use.

To generalize the findings, each square was characterized based on land use within that square. For example, if a square is comprised of 60% Residential, 20% Commercial, and 20% Special Purpose, the square is classified as Residential since this is the category with the greatest footprint. This was done at two levels of detail.

Based on this classificiation, the highest usage classifications are shown in the chart below.

If we step back to a more general level, the analysis reveals that Commercial squares on average have the highest average daily minutes of use. This is followed by Special Purpose and Residential (see note at the end on Industrial).

Commercial is a broad category. Digging one level deeper, Specific Purpose squares have the highest usage, but there are only 29 identified in the city which makes it less practically useful. Offices and Hotels are the next highest usage categories which suggests that tourists are frequent users of scooters for sightseeing purposes, and that office workers may use them during the day to run errands. It's also possible the Office usage is related to last-mile transportation from subway stations.

Similarly, Residential is a broad category. We can dig a level deeper on two fronts. First, we can classify based on the density of households in each square. Second, we can classify based on the average assessed value of a household in the square using property tax records. The heatmap below reveals that there isn't much usage in squares with a density below 21 households, regardless of assessed value. From 21 to 60 households, high usage is concentraded in areas with assessed home values of $600,000 and up. However, the greatest usages is concentrated in dense areas of 60 or more households, with the most usage concentrated in squares with average assessed home values of $200,000 - $800,000.

Keep in mind that there are households in other squares, but those squares may have more Commercial than Residential, and are therefore classifed as Commercial.

Deep diving into the last major category, we see that most of the Special Purpose usage is concentrated in the Museum, Library, and Gallery category. This may be unique to Washington DC and its many museums, and as previously mentioned, the situation surrounding the National Mall and scooter placement.

Making it Impactful

As previously mentioned, this can be used as a building block to make more informed decisions on scooter placement and market-entry strategy. To make it more practical for those applications requires additional analysis highlighted below:

Optimizing existing operations - The average total minutes of usage in each square classification was the focus of the above analysis. To make it practical for a scooter company, an additional layer of analysis is needed looking at scooter-level data to understand the capacity of each square in terms of scooters. For example, if a scooter company has 500 scooters to allocate, it matters in deciding where to place scooters whether 100 minutes of usage in a square classification comes from 2 scooters or 5 scooters.

The analysis could also be tailored to identifying squares by minutes of usage in excess of a breakeven level. For example, if a scooter has a daily breakeven equivalent to 10 minutes of usage, then the analysis could use that as a threshold instead of the present method of looking at above-average usage.

Identifying expansion opportunities - At a high level, a predictive model can be created leveraging the analysis above to assess the usage potential of different cities. This would require similar property record and tax assessment information to be available for other cities, which could be further supplemented by census data and other alternative data sources.

Adjusting service in response to COVID-19 - COVID-19 may have lasting impacts on tourism and travel. Based on the above analysis, this may reduce the usage potential of key squares in the Hotels and Museums classifications. The analysis provides a starting point for identifying the next-best squares to deploy scooters and understanding the negative impact of redeployment in terms of lost minutes of usage. The previously made point on capacity in the "Optimizing existing operations" bullet point above also applies here.

About the Data and Process

The scooter data is comprised of geolocation data points from Jump (both scooters and bikes), Lime, Skip, and Spin. This is a subset of operators in DC, so while it doesn't provide a complete picture, it provides enough of a view of what locations are relatively more favorable than others. It also has the benefit of providing a cross-sectional view of usage, as opposed to being confined to a single company's usage data.

The data covers the period from November 2019 to February 2020. This period was selected to exclude the impact of COVID-19 on usage. However, a potential drawback is that it focuses on colder months with relatively little daylight when usage may be relatively low compared to months with warmer weather and longer daylight.

Furthermore, the data is not uniform across companies with variation in the type and frequency of data. To get something useful from the dataset required assumptions on my part about what activity did and did not consitutue a trip, amongst other factors. Therefore it's my best estimate for the scooter market in Washington DC though it may not 100% line up with each company's internal data.

Data on each square's characteristics is based on information from Washington DC's Open Data portal which incldues data on the location of different lots, as well as information about what is on the lot (i.e. Residential, Commercial, etc...). Tax record data from the same source was also used to supplement the analysis.

A note on Industrial: The two highest-usage squares were categorized as Industrial. Looking on Google Maps, there didn't seem to be much to support this outcome. In addition, all scooters were from the same company in these two squares. Therefore they were excluded from the majority of the analysis under the assumption there was an error in the data.