At any given time, tasks in a mobility market are competing with each other for a limited supply of operational resources. Hiring more employees means more completed tasks, but at the price of higher and potentially inefficient labor costs. Reducing headcount decreases labor costs, but leads to vehicles remaining out of service for longer, and the deferral of tasks, like rebalancing, that greatly improve revenue. As a result, operations teams are constantly looking for ways to do more, higher-value, tasks with fewer resources. To do this, operators’ options include:
- Maximizing the number of tasks completed per labor hour
- Focusing on the most important tasks; those that will either earn, or save, the most money
At Zoba, we’re strong believers in using modeling to improve the efficiency of mobility fleets. Though we have primarily focused on improving top-line vehicle revenue through fleet optimization, the story of mobility unit economics goes beyond top-line revenue. In a recent report about the future of micromobility, McKinsey estimates that “relocation costs’’ make up between 40–50% of the unit cost per vehicle trip and anticipates “no significant change” in relocation costs for the foreseeable future. To the contrary, at Zoba we believe that it is possible to substantially reduce per-vehicle movement costs by leveraging decision automation to maximize the efficiency of operations. In this post, we will dive into the approaches that will best position operators to substantially decrease vehicle movement costs in the near term.
In today’s mobility landscape, most operators are performing only basic operational optimization — primarily starting with route planning. For example, to increase the number of tasks a single person can complete in a shift, some operators plan sequenced routes to help drivers traverse a city as quickly as possible. If routing software can help a driver handle 30 tasks per shift, up from 25 per shift, the value of labor in the market has increased by 20%.
While route planning increases the total number of tasks that operators are able to complete in a given time window, it does not help them prioritize the most valuable tasks. At any given time, there are myriad possible decisions about which tasks should be performed and in what order.
The first step in improving task selection is understanding how to prioritize tasks that will drive the largest marginal increase in ridership. This means selecting the most critical vehicles to swap (or charge) batteries for as well as ensuring the fleet complies with all market regulations. If a vehicle is out of circulation, or the operator loses their license for failing to meet equity requirements, it’s impossible to get any rides. Second, operators must prioritize tasks that improve fleet efficiency by moving vehicles from areas with low marginal demand to areas with high marginal demand. They can further increase task value by combining elective (e.g. rebalance) and required (e.g. dead vehicle battery swap) tasks to increase the value of each vehicle touch. For example, swapping the battery of a vehicle at the same time it is rebalanced to a new location realizes higher value from the same labor input by lowering the marginal cost to rebalance.
Another example of operators increasing the value of tasks in the real world can be seen in Zoba’s customers’ use of our fleet optimization recommendations to choose high impact rebalance tasks. By choosing better locations to place vehicles, operators improve the value of individual rebalance tasks over the market’s legacy baseline. Using similar methods, Zoba can also suggest which battery swap tasks to focus on. For example, the value of a battery swap task is dependent on how many more rides the vehicle would capture if it received a new battery. Vehicles that have the highest marginal contribution to ridership generate the highest value battery swap tasks for operators and should be prioritized as such.
While every individual decision may not be particularly difficult to make on a one-off basis, it is nearly impossible to make decisions across the thousands of priority trade-offs present in a highly dynamic market using human intuition. Vehicles in need of a new battery don’t only compete for resources with other vehicles in need of a new battery; they compete with vehicles that need to be collected for maintenance, vehicles that are parked illegally, vehicles that must be rebalanced, and so on. Any time an operator spends time handling a task for one vehicle, they are giving up the opportunity to handle a task for another.
To prioritize tasks globally, operators must first determine the relative value of each task. For example, a task stemming from a 311 complaint should be prioritized over other operational tasks because not completing it could lead to a lost operating license. On the other hand, not swapping the battery of a vehicle will mean potentially lost rides, but won’t result in an operator being removed from a city. Extrapolate this kind of decision making process over large markets with thousands of tasks of different types, and it becomes nearly impossible to do using intuition or a basic rules system. The problem becomes more complex when you consider ranges of priorities within a single task type. It’s important to swap a vehicle with a battery level of 35%, but at 5% the operator risks losing the vehicle entirely from a loss of GPS signal. If that wasn’t complicated enough, vehicles with the same battery life remaining have differing priorities depending on the marginal demand where they are located.
Furthermore, routing does not exist in a vacuum. The amount of time spent traveling between tasks directly impacts how many tasks an operator can perform. Choosing better vehicle routes for drivers means an operator can handle more tasks, which results in a new set of routes, which then impacts which other tasks can or cannot be handled. A task may be valuable by itself, but that value is relative not only to how valuable other tasks are, but also to how far out of the way an operator must go to reach the task.
In a previous article, my colleague Evan discussed how we can balance tradeoffs using mathematical optimization. Just as we can apply mathematical optimization to place vehicles where they will receive more rides, so we can apply it here to determine the best use of operational resources. To find the highest-value operational strategy, we need to define the following:
1. Decision Variables: The choices we make when trying to complete tasks
- The van that should be used for each task
- The order in which each task is completed
- The priority of tasks based on their expected value to the market
- The tasks that should be skipped if it’s impossible to complete all tasks in the given time window
2. Constraints: What limits a market’s operations
- The number of vans and workers
- The number of vehicles or batteries that can fit into a van
- The length of time an operator has to complete tasks
3. Objective function: How good is the output of any given set of decision variables
We can define the “best” plan as the plan that minimizes the total cost to execute all tasks in the market. To define our total cost we must first understand the cost of each task. The cost of performing a task in a market is dependent on whether it is completed or not. If an operator chooses to complete a task, the total cost of completion is the cost of getting to the task, and of completing it:
We also need to model the possibility that the operator doesn’t have the resources to complete every task. In that case, we can skip the task and incur a penalty:
When there are not enough operational resources to complete all tasks, the solver will begin selecting tasks to skip. It will prioritize keeping tasks that have higher value, because the cost of non-completion is higher. For example, completing a 311 request is very important to keep an operator in good standing with a city, and so has a high cost of non-completion. Additionally, overall value can vary amongst tasks of the same type. A vehicle in need of a battery swap that is in a high demand area is more expensive to skip than a vehicle in need of a battery swap in a low demand area.
Therefore, now that we’ve modeled the cost of completing a task and the penalty for skipping a task, we can define the cost for a given task T as:
From this we can then define the global cost function for a market with N tasks to be:
With the constraint that:
The solution that minimizes this objective function will be a set of van trips that have the shortest possible routes that complete the most valuable tasks in the market.
This optimal solution can be reached using a constraint solver. The solver attempts to find the best solution by first minimizing the total driving duration (all of the benefits of trip planning), while also working to make sure that those trips allow drivers to complete the most valuable collection of tasks possible. If some tasks are unable to be completed in the given time window, the model will then work to drop tasks that are both hard to get to (have high cost in labor hours) and of lower value, while ensuring that essential tasks such as 311 requests are addressed as quickly as possible.
The benefit of using mathematical optimization to plan operations in a market is that it doesn’t just improve efficiency around the edges; given enough time, the solver will find the global optimum. The solution dictates not only which tasks should be completed, but how to work through them in as little time as possible. Additionally, because the solution is the global optimum, there is no better way that an operator could handle tasks in the market. Based on internal research we believe that optimizing fleet operations in this way can help lower vehicle movement costs substantially, increasing net margin per vehicle by 15% or more.
Routing tasks through a city efficiently is important, and many operators are taking this smart first step to improve their operational efficiency. But there is much more to be gained from fully automating the thousands of decisions required to move vehicles through a market efficiently in conjunction with routing. By better balancing the tradeoffs between operational demands and labor constraints, operators can significantly improve the unit economics of their fleets.