Boylston Street, Boston: A Case Study on the Effects of Food Delivery

In the not so distant past, the word "delivery" generally referred to pizza delivery, and it was a service only a handful of pizzerias with their own dedicated drivers could provide. Orders were generally placed over the phone and there was no telling exactly when the driver would show up. Now, thanks to technology and platforms like Doordash or Grubhub, delivery of just about anything from duct tape to cheesecake to a margarita is possible — all at once in the same online shopping session, paid with a single click, tracked in real time, and carried out by a freelance delivery driver with an E.T.A. algorithmically predicted down to the minute.

As the 2010s came to a close, the popularity of online food delivery was growing steadily thanks to technology and the increasing breadth of delivery services, when suddenly the COVID-19 pandemic began. By March 2020, nationwide “stay-at-home” orders were issued and millions of people began to work from home, setting up the perfect storm for the delivery industry to thrive. Millions turned to delivery as a safe way to get food, supplies, and groceries without having to risk exposure to COVID-19 or step away from work. The delivery industry grew at the fastest rate ever seen, and all delivery platforms reported a massive increase in orders. Yet, even now as we return to “post-pandemic” normalcy, the food delivery industry continues to thrive and shows no signs of slowing down. In short, the delivery industry was not just a fad or a band-aid during the pandemic; it is here to stay.

The term "convenience economy" aptly describes the paradigm shift. For just a few extra dollars, getting a meal from a restaurant, going to a department store to buy supplies, or even grocery shopping can all be accomplished from the comfort of home. Rather than going for a drive or walk to obtain goods, all one has to do is simply open their door to find their delivery order waiting for them. There is much good in this, such as allowing people to have more time for friends, family, work, exercise, leisure, etc., but the luxury is not without negative externalities.

The increased delivery activity was not immediately noticeable during the pandemic in many cases, as streets that were once used for consumer parking and commuting were being far less utilized than normal. But now, many are returning to work or are simply more willing to leave home in general, meaning the road must once again be shared by more users.

This fact, combined with the continually increasing demand for delivery, has resulted in drastic consequences on transportation and, in particular, parking. The incredibly valuable resource of curbspace is becoming increasingly occupied by delivery drivers, vying for space that was already saturated pre-pandemic. As we put it at Vade, the demand for curbspace allocated to delivery activity has changed enormously while the supply has not. Most cities' leaders have recognized the dilemma and some have tried to reform their parking policies in problematic locations, but in many cases the issues are not resolved, nor is the extent of the issues fully understood. You can’t improve what you can’t measure, so the symptoms persist or even worsen. The result is double parking, idling, and congestion local to the delivery hotspots. Zooming out further, this leads to many undesirable implications: decreased road safety, traffic, increased emissions, less parking access for non-delivery drivers, slower delivery times due to inefficiency, etc. In short, the increased delivery activity is becoming a significant challenge for many cities.

For example, you may have seen recent articles or news segments from The Boston Globe, CBS Boston, NBC Boston, or WHDH about what has been described as “chaos”, “a dilemma”, “a recipe for traffic”, and most notably… a “logjam”. Boston’s Chief of Streets Jascha Franklin-Hodge himself declared the situation a crisis.

In the words of Eli Rosenberg of NBC, live on the scene: “the issue is double parking…[on] this stretch of busy restaurants… Food delivery drivers [are] clogging [the road]”. That busy stretch is the 500 block of Boylston Street, located in the Back Bay of Boston, whereon a concentration of popular fast-food and fast-casual restaurants reside — namely, Chick-Fil-A, Chipotle, Smashburger, and Clover Food Lab — all of which offer pickup/delivery options via online platforms. During lunchtime and dinnertime, the road is flooded with delivery vehicles and the disorder described before ensues.

The story was first covered (to my knowledge) by Anissa Gardizy of The Boston Globe in December of 2022. In her article, Gardizy explains that the issues due to food delivery have been ongoing in several areas throughout Boston, Boylston Street being the “most egregious example”, and that it has become bad enough that city leaders have made it a top priority. The City of Boston by that point had taken several measures to ease the problem. First, metered curbside parking was replaced with free 5- or 10-minute food pickup zones (5-minute on Boylston Street), sacrificing meter revenue estimated to be worth “several hundred thousand dollars” per Chief of Streets Franklin-Hodge in the name of safety. The problem persisted, so enforcement was ramped up by having officers monitor the pickup zones, but due to the difficulty of catching offenders who may only be double-parked for a few minutes, traditional enforcement methods were deemed ineffective. Also, as Franklin-Hodge points out, traditional enforcement punishes the drivers who are working to make a living, rather than the company (i.e. Grubhub, Doordash, Uber Eats) dispatching the drivers. It became apparent to city leaders that more creative solutions would be needed.

In an effort to understand the problem more, the city brought representatives from delivery companies and restaurants together to discuss solutions. In Boston Board of Licensing meetings (available on youtube), city representatives can be seen asking restaurants about their delivery projections and other related metrics. As a result of these meetings, some delivery platforms went as far as to limit the number of orders certain restaurants on Boylston Street were able to take. Specifically, on Boylston Street, Chick-Fil-A was limited as it was recognized as the most popular restaurant on the block -- more to come on that in a followup blog. As of late February 2023, evidenced by ongoing coverage of the story, the problem persists on Boylston Street and elsewhere, and it likely is still a top priority of city leaders. The following is a story and case study of that problem.

…

…

…

Shortly after The Boston Globe’s article was published, it was shared internally amongst Vade employees as an interesting read that pertained to the bread-and-butter of what we do and what we try to solve as a Curb Management company. Reading the article, I had an idea: let’s put some cameras up and see what’s really happening on Boylston Street. Of course, we can’t just go put up cameras on street poles or other city infrastructure at will, so I had to get creative. I made a few calls to some businesses located in the vicinity to see if anyone with a good view of the street would let me install cameras looking out from their window. Fortunately, one business (who wishes to be kept anonymous) was empathetic of the cause and kind enough to let us install cameras on their property. In fact, they seemed rather glad to receive my unusual phone call, and after an email or two they were happy to help as the traffic and lack of parking on their street was affecting them, their business, and their customers.

With a plan in place, I sent my new colleagues two cameras to install with suction cup mounts on two windows to capture a wide swath of Boylston Street. A week and change later, we had 24/7 footage from Friday, January 13 through Thursday, January 19, amounting to over 40,000 images between the two cameras, covering around half of the blockface (approx. 275 feet).

The images were retrieved and processing began to obtain data on curbside parking and double parking violations. There are five lanes on Boylston Street: two curb parking lanes, two travel lanes, and a bus lane. The near-side curb parking lane is the free 5-minute food pickup zone (formerly metered parking but was adapted for the increased delivery activity); the far-side curb parking lane is 2 hr metered parking. We looked for parking sessions in the two curb parking lanes and two travel lanes as there was negligible parking activity in the bus lane.

For processing, the footage was pushed through Vade’s proprietary computer vision pipeline, but we found that the unusual camera mounting location made for inconsistent results. Namely, as we were imaging through a window, there was significant glare during certain times of day on the window pane. Also, Vade’s AI model is trained using images from perspectives in the 10-20 feet above the ground range; these cameras were ~45 feet above the ground, meaning the cameras were much farther away from the vehicles and had an atypical, almost overhead view. Cars on the extremities of the frame where these factors were most extreme were sometimes missed by the model. Thus, to ensure accurate data, all sessions were validated and missed sessions were inputted with Vade’s human-in-the-loop pipeline.

As the data processing came to a close, we found ourselves stunned by the results.

In one week, the section of Boylston Street we captured produced over 8,300 parking sessions across the four lanes, an almost unfathomable percentage of which were double parking violations. On some days — more so on weekdays — almost half of the sessions were double parking violations. In terms of dwell time, the vast majority of sessions fell into the 1-15 minute dwell bucket. In terms of vehicle type, 95%, of the sessions were passenger vehicles, i.e sedans, minivans, SUVs, etc versus commercial vehicles, i.e freight trucks, service vans, box trucks, etc.

You may have noticed another outlier: Sunday. The difference between Sunday and the other days is so drastic that we thought there may have been a bug somewhere throwing off the numbers. It was in fact not a bug. More to come on that in another follow up blog post… (Hint: Eat More Chicken!)

Given Boylston Street’s reputation as a delivery hotspot, we wanted to dig further into the delivery related activity. Looking at the dwell time distribution of food pickup zone sessions and double parking sessions (assumed to be related to food pickup activity), we can see that there were plenty of sessions that extended well beyond 5 minutes. Many lasted 15-20 minutes or even longer. I limited the x-axis range shown in the chart for clarity, but there were well over 20 sessions throughout the week lasting over 3 hours in the food pickup lane, the longest of which lasted over 11 hours. Based on the distribution, delivery sessions in this study will be defined as sessions 1-15 minutes long, although there is evidence that some food pickup sessions last longer.

With the length of food pick up sessions defined, let’s assume that all sessions on Boylston Street in any lane lasting 1-15 minutes are food pickup related. The figure below shows the count of short sessions (assumed to be synonymous with food pickup sessions) by day and lane type. We can see that the vast majority of short sessions are in the food pickup zone or double parking violations, while very few occurred in the far side metered curb parking lane. In filtering for short sessions, it is also revealed that 7,004 of the 8,310 total sessions (nearly 85%) were short sessions.

Looking at the short sessions by hour in the figure below, it can be seen that the peaks occur, as would be expected, around lunchtime and dinnertime, during which hours as many as 100 short sessions is not unusual for any day other than Sunday.

To get a fuller picture of the demand profile for short sessions by lane type, the data separated by lane type is shown on a timeline view below.

As seen in the table view, there are two discernible spikes corresponding to lunchtime and dinnertime for each day. The trends are generally congruent across the different lane types, suggesting that the food pickup demand during the peak hours spills over to the other lanes. To understand why food pickup parking is spilling into other lanes, the peak occupancy and utilization of the food pickup zone can be found, calculated using an estimated maximum number of vehicles that can fit in the section of the food pickup zone we covered.

The peak occupancy and utilization tables show that the pickup zone becomes most occupied and utilized during the lunchtime and dinnertime spikes. However, despite high levels of occupancy, utilization is seldom above 30%, suggesting that the pickup zone is not highly occupied consistently even during peak delivery hours. If this is the case, it seems curious that there is so much double parking as this would suggest there are actually spots open in the food pickup zone. I poured through images to find the answer and came to the following conclusion.

There are three reasons why food pickup vehicles are double parking. The first and most obvious reason is that there are in fact no parking spaces available in the food pickup zone. For example, in the image below, the food pickup zone is packed with cars, besides a small space in the lower right hand corner of the frame. Due to this lack of parking, the circled white sedan and black SUV, apparently delivery vehicles, have resorted to double parking.

The second reason for double parking is that there are no good spots available. For delivery drivers, speed and efficiency is the name of the game. The faster they work, the more money they make. If the spaces close to the restaurant they are picking up from are not open, delivery drivers may be more inclined to double park to cut out time spent walking to and from a curb parking spot that may be down the block or across the street. The image below shows several delivery drivers double parking presumably to be closer to their pickup restaurant rather than parking in the open curb parking spots in the food pickup zone. The restaurant just out of frame where those drivers are double parking in this case (and the vast majority of examples of this type of double parking) is Chick-Fil-A.

The third reason is — for lack of a better word — indifference. This is evident in the event of drivers choosing to double park even though there is a food pickup zone spot open directly next to where they double parked. In digging through images, I was surprised to find many instances evidencing the case for this third reason, especially outside of peak delivery hours when space in the food pickup zone is available. But in theory, it is logical: if delivery drivers ideally are only going to be parked for a few minutes, they may be less motivated to spend time parallel parking. There is likely a chain reaction or peer pressure effect contributing to the phenomenon as well, where delivery drivers see other vehicles double parking, so they are more likely to do the same themselves. The three consecutively captured images below show an example. In the first image, there is open and unblocked space in the food pickup zone, yet the circled vehicles that arrive and park in the second and third images choose to double park rather than back into the open food pickup zone spots. Seeing that others are double parked in the travel lane, they were likely less motivated to park legally, as it would have cost them their time and the risk of getting blocked into the food pickup zone by another vehicle double parking.

In light of the hypothesized causes of double parking, the overarching question can be addressed: what is the root cause of the problem on Boylston Street?

Looking at the parking in isolation, the problem is one or both of the following at face value: a problem with the parking policy or a problem with the parking policy adherence. In the first case, the parking policy is being followed, but does not meet the demand. In the second case, the parking policy is not being followed, but if it was followed, the parking supply would meet the demand. The case where both problems exist is one where the parking policy is not followed, but even if it was, the issues would persist.

To answer the question of policy adherence, we can look at the dwell time distribution in the food pickup zone that was shown in a figure before, which revealed that a significant number of vehicles are staying much longer than 5 minutes. In terms of compliance in a binary sense, the pie chart below reveals that over half of the sessions in the food pickup zone are not following the policy during policy hours.

To answer the question of whether the existing policy is sufficient, the maximum throughput of the section of the curb analyzed in this study is calculated. The section of the food pickup zone covered can hold at most ~12 normal sized cars. Given the 5 minute time limit in the food pickup zone, each spot can support 12 sessions per hour, meaning the stretch of curb theoretically allows for 144 food pickup sessions per hour.

Looking at the count of short sessions by hour, reproduced again below, it can be seen that the number of short sessions — across all the lanes — never exceeds 144. Even if the section was only 75% space efficient (meaning the section could only hold 9 cars due to excessive distance between parked cars), the section would be able to handle 108 short sessions per hour, which is only exceeded in two individual hour blocks throughout the week.

Therefore, if the parking policy was followed, the food pickup zone would be able to support all the short sessions. At face value, the root cause is found; it is a problem with the adherence to the policy. If drivers parked for only 5 minutes, there would almost always be spaces available and drivers would not be double parking, right?

Probably not — this would only help reduce double parking due to the first reason mentioned: the case where no spots in the food pickup zone are available. Besides, ensuring that the policy is followed means enforcement, which brings us back to the point made by Franklin-Hodges: enforcement punishes the drivers and does not fix the system. Also, if there were enforcement officers present at all times to ensure the food pickup lane had more turnover, other issues would likely arise. For one, delivery drivers may be less willing to pick up orders from the area and restaurants would see a drop in revenue.

To me, the second reason mentioned as to why drivers are double parking is an important clue to the solution. This reason for double parking was that drivers want to be closer to their destination rather than walk to and from a parking spot far away to grab their order. As mentioned, I noticed in looking through images that the vast majority of double parking due to the second reason was right outside of the Chick-Fil-A. Without spoiling too much of the follow-up blog, it is evident that the Chick-Fil-A accounts for a very large percentage of the activity on this section of the street. What if a second Chick-Fil-A opened towards the other end of the blockface? There would suddenly be twice as many good spots (that is, in close proximity to a Chick-Fil-A), meaning double parking to avoid parking far away from the restaurant would lessen. I’m not suggesting Chick-Fil-A opens a second store, but multiple food pickup points located along the blockface could aid in motivating drivers to utilize the full curbside pickup zone by making more spots good spots and may make food pickup more efficient in general.

There are likely a myriad of other solutions involving micromobility, but I will digress as that is outside of the scope of this blog, though I agree with Chief of Streets Franklin-Hodges take that “There are a whole lot of reasons why driving someone’s chicken sandwich around in a 4,000-pound SUV is a terrible thing for the city”. On that note, a case study which explores how many orders are being delivered by bikes, mopeds, robots, etc. versus passenger vehicles would be interesting.

See the follow up blog for more on the interesting trend related to Sundays!