Under the Hood: A Look at Sidecar’s On-demand Logistics Infrastructure
On-demand deliveries may be the loudest space in tech right now, but they have also become a topic of debate. Fans love the convenience of an instant mobile experience for everyday tasks like getting a ride and ordering dinner.
Critics however, pan companies in this space as unregulated VC-funded fictions or Webvan 2.0. To be fair, more than a little of that sentiment is deserved. However, I think the following quote (and the stature of its author) illustrates how split we are on the long term outlook here perfectly:
“I think there’s just an urban mythology out there that an [on-demand delivery app] somehow changes the basic cost input of the logistics business or changes the patterns or the underlying business situation and that’s just not–that’s just incorrect. So great company, great concept [in reference to Uber], but I don’t think it’s…likely to be a major player in the logistics business”
– Fred Smith, CEO, FedEx
This is certainly a very damning statement from someone who ought to know. So what’s the root of the disagreement here? After all, it is uncontroversial that on-demand delivery services are about half the cost (and half the time) of the quickest comparable delivery from FedEx. You can get food from Caviar, EAT24, DoorDash, Postmates, etc. (some of which are actually delivered by Sidecar Deliveries) for about $7, but a FedEx same-day delivery will cost you $16.
But is that simply because VCs love losing money to buy growth? Are these prices a fiction?
Mr. Smith is correct in stating that our delivery app doesn’t change the cost of taking something from point A to point B, no more than our ridesharing app does. However, once there is a network of such app users in place dispersed around a city at all times, they are able to do wonders, especially when the cloud can predict, observe, and optimize every step of the way. When everyone is connected to the Internet all the time, people can work together in shockingly more efficient ways than was previously possible.
As I intend the newest iteration to our logistics infrastructure to demonstrate, the on-demand economy represents a fundamental change to how much these services cost, the reliability and speed with which they can be deployed, and in general, what is possible when it comes to getting things on-demand.
With that, let’s take a closer look at what’s going on here:
An Inter-modal Delivery Network Despite our roots in peer-to-peer ridesharing, our fleet is not exclusively made of cars. As we began to scale deliveries, we realized that cars are actually not optimal for many deliveries in an urban environment. Today, depending on a number of factors, when you place an order with one of our partners, it is delivered to you via some combination of cars, bike couriers and walkers. Each of these modes has predictable strengths and weaknesses: bike couriers can’t carry a huge volume of food or drink, but are FAST over short distances, walkers take zero time to park but can’t move distances greater than 0.25 miles (they must hand their goods to a different mode at one point or another to get the item to its destination). The ability to intelligently decide which mode works best at which time, and path-find between multiple modes, is a powerful cornerstone of the Sidecar network. While other startups also do this type of optimization, Sidecar recently announced we combine walkers and cars or bikes for a particular delivery, which was even greater impact on efficiency.
Yield Prediction and Location Efficiency Management
The next piece is more subtle, but equally important. Every city we operate in is broken up into 0.25 x 0.25mi “grids”, and our data and operations teams work together to predict how many orders are going to come in over each grid, each day, every 15 minutes.
Additionally, the data group observes the locations at which parking is difficult (i.e. any restaurant along Market Street), which our algorithms then take into account to determine routing and which delivery mode(s) to effectively utilize.
Our Routing Algorithm
The routing algorithm is the heartbeat of our on-demand infrastructure. In compsci terms, it is solving a new class of problem: a traveling salesman problem modified with ordered pickups and dropoffs and deadlines. Most importantly, however: in our context the problem is not static. An optimally-routed system will become sub-optimal once the next person orders a ride, a burrito, an emergency auto part, etc. This obviously has tremendous impact on complexity of solution, and continued optimization on this front is where we spend a LOT of time.
This is the stuff of an algorithmist’s dreams (or nightmares). Every car, bike courier and walker on the Sidecar network is not on one ride, or one delivery per se. Rather, they are assigned to an ordered set of multiple waypoints each with individual deadlines. They might be picking up two orders at the first location, dropping one off at the next, picking two more up at the third, and so on and so forth.
These waypoint sets can be assigned or reassigned on the fly, as optimality is recalculated.
As new orders come into the system, the algorithm assigns or reassigns them while taking into account prescheduled orders still needing to be dispatched. The goal of the system is to maximize driver efficiency and minimize confusion due to reassignments, all while targeting a 95% on-time percentage.
Take for example a San Francisco driver with someone’s groceries in her backseat who is headed to deliver them in Lower Nob Hill. An order is placed for a hot food delivery and the pickup is roughly along the driver’s route. The algorithm now has a decision to make: offer this order to an available driver in another neighborhood, or dynamically insert it into the occupied driver’s itinerary. In this case and in many like it, the already-busy driver ends up being the best choice for this new work as they gain efficiency while only adding marginal incremental drive time to make the pickup.
Putting it All Together
So what’s the punchline here? Well, efficiency translates into price, and a low price for on-demand deliveries means that you can get more and more things delivered on-demand without being cost prohibitive. Let’s walk through a peak demand evening and see what happens:
Let’s say a particular 0.25 mile x 0.25 mile grid has a number of on-demand enabled restaurants, bakeries, medical marijuana dispensaries, etc. on it. Let’s also suppose that, over 60 minutes, the sum total of all orders placed to locations inside this grid is 20 deliveries.
If this isn’t a first time spike in demand, our forecasts would indicate that walkers should be placed in this zone in anticipation of this volume. In this situation, let’s say that eight of the orders were headed north of the grid in a relatively clean group, and eight were headed south in another clean group.
In this case we should expect two walkers each to pick up eight orders as quickly as they can, and hand each batch of eight to a different driver, who will arrive once the walkers are projected to complete their last pickups. Additionally, for the outliers, one or two bikers will come and pick these up and drop them off.
All this works with the labor economics of the local market. In San Francisco, walkers need to make $16/hour, drivers make $22/hour and bikers make $19/hour (and they are all in the area already). At peak, a fully loaded car can drop off six packages/hour while a biker can do four/hour. Therefore, fully loaded everyone gets paid including our 20% commission at $6.90/delivery. It seems too cheap to be feasible, but it’s not; the key is that 0 effort is wasted by any person in the network, and additionally, no vehicle spends ANY time doing something inefficient for that particular type of vehicle. Drivers don’t park, walkers don’t move large distances, bikers don’t take too many packages, etc.
Now that these pieces are all stood up together, all of sudden, you can transport things across the city from any point in the city to any other point under the right circumstances for $7, in under an hour, which despite what my colleague at FedEx previously stated, is a marked reduction (again, the cheapest two hour FedEx delivery in SF is $16), in cost. These costs aren’t a fiction, they come from increased efficiency of all parts of the network, as well as being able to utilize a network of drivers, bikers, and walkers that are ALREADY in the area for simply the portions of the deliveries they are suited best for (i.e. making the pickup, driving a batch, etc). Contrast this with the status quo of driving dedicated cars into the area and then driving them (a “dead-head”) all the way back and it is clear what is happening. Simply put, the on-demand contractor model, especially when different types of contractors are working in tandem, is a much more efficient one when it comes to these kinds of hyperlocal deliveries.
This is far more than a convenient experience with a slick interface. Sure, some of the fundamentals are the same. People need to make the same amount of money for the same amount of work. However our technology enables us to work together more efficiently than was previously even imaginable. When we take a moment to look, we can see that the changes in our technological capabilities are driving a fundamental shift in what is possible in the on-demand logistics world, and therefore consumer behavior at large. The larger question is, once this kind of network exists everywhere in the world, what breakthrough services will the world build on top of it?
Originally published at www.side.cr.