The future of route optimization is here!
- June 23, 2021
- 3 mins
[REPORT]
The different types of route optimization
Despite the power of optimization algorithms, route optimization solutions are usually very limited in their ability to handle most everyday problems encountered by planners: unforeseen events, incorrect or unknown data, constraints and priorities that cannot be translated into the algorithm, etc.
In fact, route optimization software has inherited from mathematics a purely theoretical notion of “absolute optimality” which has no real practical application. In reality, the best decision at a given moment, if it exists at all, can be strongly challenged a few minutes later, as new information arrives, time passes, and the context itself evolves.
As a result, although route optimization undeniably improves the efficiency of routes compared to manual planning, it makes users lose a large part of their reactivity when making decisions. Indeed, despite impressive theoretical performances on paper, they often fail to truly encompass the complexity of the real world, where everything is constantly changing. New service quality standards are emerging in all industries, such as same-day delivery/service or real-time interaction with the end customer, and make contexts even more changing than in the past.
This is why, even today, most organizations rely on manual planning, as the reality they face is far too complex for most route optimization software.
Continuous and context-sensitive route optimization
At Kardinal, we believe that it is possible to overcome the limitations previously outlined. For us, this requires designing route optimization around the following 2 concepts:
- Continuous optimization: events are happening all the time, decisions are made at every moment. A route optimization software must help continuously: before, during and after the routes.
- Context sensitive: without context, route optimization software is limited to generic recommendations, not always relevant, which often do more harm than good. Context sensitivity is the key to making the best decisions at any given time.
These two concepts cannot be considered separately. Of course, designing a route optimization software that never stops computing while being able to gather and use structured contextual knowledge in real time is quite a technical challenge, both in terms of pure software design and in terms of the mathematical algorithms needed to run the whole process. But we strongly believe that it is the only realistic way to bring route optimization to the field, alongside the men and women who make logistics work.
Below, we will show how these two concepts translate into the real world upstream of operations, when drivers are on the road and even after rounds are completed.
Continuous route optimization solution
Before operations
In most situations, the data for route optimization can never be completely and perfectly known in advance. For home care visits, for example, it is impossible to know in advance exactly which patients will need a visit during the day since this activity, like many others, involves managing urgent interventions on a daily basis.
In express parcel delivery, delivery orders continuously flow into the information system and it is impossible to wait until the last moment to launch the optimization before starting to sort the parcels.
Routes must therefore be constantly adapted, but without overlooking the context and timing: you can’t load all the parcels into a truck a few minutes before leaving without making it late. Thus, the optimization algorithms must be able to “understand” the status of each tour at a given time, in order to provide appropriate modifications whose impacts will be acceptable for the warehouse management.
During operations
Once the drivers have left the warehouse, various events can occur. The first and most impactful is traffic. Even with accurate traffic forecasts, which will be accurate on average, road congestion or public transport availability can evolve very differently from what was anticipated, threatening service quality when things go wrong, or providing opportunities when they go better than expected. During operations, the context can be described with real-time traffic as well as transportation options available to the worker.
In addition, new tasks may need to be included in ongoing rounds, such as new package pickups added on the go. For home visits, a home care worker currently at a patient’s location requiring more care than expected will cancel future visits, which can then be optimally assigned to other available home care workers.
After operations
Once the rounds are over, new rounds will soon follow. This time should be used to work on the data collected in order to improve the system as a whole. Indeed, real durations have been recorded: travel time, work time on site, etc. They can vary according to the type of work and depending on a wide variety of factors. Therefore, it is essential to understand these variations in order to use them in the next route optimization.
For parcel delivery, it is known that drivers are generally faster in the areas they know best. Can this be quantified? The type of vehicle used, the driver’s seniority, the time of day, the vehicle’s load, the time of year or the weather conditions will all have an impact on their speed, as will the type of environment in which they operate (office towers with many floors or suburban areas will have a significant impact on delivery time).
For home visits, it can be expected that junior caregivers will take longer to perform the care, due to their lack of experience. In addition, patients themselves may require different amounts of time for the same service. All these inputs make a basic approach of allocating X minutes per patient irrelevant in practice.
In both cases, Machine Learning algorithms will learn from previous data to understand how all these factors impact actual times. Over time, as the amount of data stored increases, the entire system will become more accurate. This will create a positive feedback loop, following the learning curve of the workers, which will not only optimize the rounds one after the other, but will also optimize itself over time.
As we can see, continuous and context-sensitive optimization is the next step towards a stronger connection between processes and algorithms. This will provide organizations with greater responsiveness and, above all, more efficient and realistic routes, resulting in real performance improvements and work comfort for the field teams.
