Its 4 in the afternoon and a load planner is in the office trying to plan the next day’s deliveries. She has one thousand orders to plan for the whole day. First, she needs to keep volumes of nearly 10,000 Stock Keeping Units (SKU) in her head. Then she needs to visualize all 1000 delivery destinations on the map, some of which she has seen for the first time or for occasional clients. Then she needs to memorize delivery windows for the 1000 locations. She knows that timely deliveries are the key to the company’s success, so her main goal is making these 1000 deliveries happen as accurately as possible, otherwise there will be a real-world impact.

So, what’s wrong with this process? Isn’t this how its always been done? Well yes but this job requires hours of focused work for even an experienced planner who has spent years in a firm (imagine what happens if she is inexperienced). Additionally, a planner primarily takes care of the level of service for her clients. What she is not able to focus on to the same degree, or even at all – is the efficiency of her own company’s operations. This results in unnecessary miles driven and inefficient utilization of trucking capacity.

**What is the solution?**

**Routing optimization**. Routing “optimization” is a buzz word used very loosely nowadays. For the majority of cases, people understand it as the sequencing of truck stops to get the best driving time. At most this is a partial automatization of a planner’s work, but it doesn’t mean that the route plan is “optimal”.

The most widespread way of modeling multi-stop delivery operations is by solving the so-called Vehicle Routing Problem (VRP’s). In the majority of cases, the VRP goal is to optimize driven mileage, which, in formal terms, means getting minimum possible mileage for the given plan. Combinatorial optimization is the mathematical apparatus to solve VRPs. There are two major ways to solve this problem – using exact methods or metaheuristics (approximate solutions). The discussion of combinatorial optimization is a topic for another post, so we won’t go deep into it now. What we can say is that the exact methods are limited to 100 nodes (deliveries), hence if one wants to get a plan within a reasonable time the use of some kind of metaheuristic is absolutely necessary. A consequence of this process is that no solution can get an “optimal” result but rather should try to get as close to optimum as possible. For example, if the theoretical minimum for the mentioned 1000 deliveries is 10000 miles with 40 trucks, different algorithms will certainly get more miles than that (could be both 11000 and 14000). In fact, as no one knows what the theoretical optimum is, she can’t realistically assess the quality of the solution through a comparative process. In our experience, manual planners get somewhere between 60-80% to optimal. So, what is the best way to assess how good the solution is? We must compare the firm’s historical data (if that exists) or run it in parallel with manual planning to see the difference in optimization rates. As an industry standard, good routing solutions should give results close to 95-97% to optimal.

While there are a number of solutions for coping with this problem producing varying levels of success, there are huge differences both in (a) optimization quality and (b) application usability. Some of the distinctive characteristics of a “good” application are described below.

**1. Complexity. **The main problem of VRP is time complexity or, in other words, the fact that the amount of time needed to solve the problem grows exponentially when the number of deliveries or nodes increases or when the algorithm incorporates more constraints. To give you an idea of the problem encountered when scaling up nodes/deliveries consider that solving a Capacitated Vehicle Routing Problem with Time Windows (CVRPTW) for 100 nodes takes about 10 minutes using current commercial optimization solvers but takes a whopping 8 hours for 300 nodes. Why, because solving for optimization goals for the first set of nodes takes tens of millions of permutations and for the latter – hundreds of millions.

Below are a few more examples of variables dramatically adding to the numbers of permutations and thus increasing the time for a potential optimal solution.

**1.1. Time complexity** rises when an optimization algorithm tries to incorporate more real-life data whose absence would massively degrade the usability of the solution. A primary example is that many firms have **a variety of trucks** and equipment. An average firm can use 4-8 types of trucks with different capacities. The majority of solutions either don’t include the capacity constraint into calculations or are using just one capacity constraint.

**1.2. Delivery windows** is another variable that adds complexity to the problem. The planner should make sure that capacity is maximally leveraged while ensuring ETA’s are within delivery windows.

**1.3. Service requirements** are different for different clients. Dwell or loading/unloading time can vary for each of these 400 deliveries and the optimization algorithm should be able to incorporate it into calculations.

**1.4. Priority orders.** There are orders which should be prioritized over the others. Thus, the optimization algorithm should be able to keep strict delivery requirements for the high priority deliveries while still assuring the best optimization goals.

**1.5. ELD’s (Electronic Logging Device) **are mandatory nowadays and **Hours of Service (HOS) requirements** should be considered from the planning phase. This should go as a constraint variable into the algorithms as well.

Industry constraints that need to be considered when examining the role of optimized planning:

**2. Product data availability**. An average distributor operates with 1000’s SKUs and a large distributor with 100’s of thousands. Each SKU has a different volume and weight characteristics which should be taken into account when planning deliveries. Not all applications incorporate product-specific accurate data into calculations or use capacity as a constraint at all. This results in a lot of capacity violations and too many routes becoming obsolete when it comes to filling and **dispatch**. There is also a highly variable human factor in play because many companies don’t have all the needed information in their Warehouse Management Systems (WMS). As a result, algorithms should be able to work with incomplete data. Needless to say, the software should be able to seamlessly integrate with a firm’s order and warehouse management systems or other Enterprise Resource Planning (ERP) modules

**3. Remote work capacity**. As the routing problem is complex and requires lots of data and computational resources most of the routing applications available on the market should be installed on-premise. Conversely, it is becoming imperative that people who are not in the field (and even people who are in the field) should have access to multi-user plug and work opportunities. This creates a high demand situation requiring a novel solution. The routing solution should offer robust multi-user functionality. At the same time, it should be able to process billions of combinatorial permutations in seconds. An advanced enterprise-grade cloud-native architecture is needed, otherwise, the firm will be creating additional problems rather than solving old ones.

**4. Big data**. Imagine what happens when there are not one but 10’s of planners in the firm who plan data for multiple warehouses and distribution centers. An optimization solution requires multiple installations in different warehouses resulting in massive investments and further support cost. Even without that, the majority of solution applications in the market are limited to planning 500 simultaneous deliveries and are only suitable for small firms. One can plan routes in batches but that will substantially degrade the quality of optimization for the total plan. Growing and competitive firms should have the opportunity to run and compare multiple instances simultaneously, plan tens of thousands of deliveries, and have other collaborative functionality which is only possible with the enterprise-grade cloud-native architecture.

**5. Additional features**. There are a number of additional features that should be embedded in any forward-looking solution such as a) data interoperability b) creation of dispatch documentation c) monitoring dashboard e) functional analytics f) dynamic planning, etc. We will return to these topics in other blog posts.

**Current Solutions in the Market **

There are different solutions in the market which help to plan loads and dispatch them. We will list some of them and group them following the descriptions copied from their websites.

**Maps**

Mapping applications offer route optimization, which is limited to getting the best sequence for the given data points. The number of points is limited to 10-20 points. Apparently, maps don’t do VRP optimization for multi-stop deliveries. Maps won’t provide any product and shipment management functional either. Routing functionality would be a good addition for personal or occasional usage.

**Free routing software**

Free applications are one step ahead of current maps routing functions. They can be used by small firms, but it is not clear if these applications use any kind of mathematical optimization. They apparently miss the majority of the ideal functionality previously touched on.

**Paid routing software**

All paid solutions state that they have some optimization algorithms under the hood. Routific and OptimoRoute have straightforward pricing while Route4Me has a relatively low base price but additionally charges for different features. Workwave has several products for different industries. All these solutions incorporate delivery windows in the planning and build multi-stop loads.

**In choosing the most suitable software one should consider a couple of other factors **

First, one should look deeper into **what has been optimized and how good it actually is**? As we have noted above, results can be different (sometimes substantially) depending on the types of algorithms and optimization methodology used. The best way to compare is either to test the same dataset using different applications or compare results with the optimal values for that dataset.

Second, does software **optimize the delivery plan for the day or just updates sequences on fixed routes**? Running CVRPTW is hard and requires substantial resources and appropriate algorithms, so many applications just help you to build a route plan once, without taking into consideration specific order details and location time windows. You would assign orders manually afterward by figuring out sequences. Later is a much simpler approach but it substantially reduces optimality. I.e., Less^{®} Platform runs CVRPTW for every group of orders from the scratch, thus assuring the most optimal route plan for that specific order list.

Third, data requirements for foodservice wholesale distributor and cleaning services, for example, can be substantially different. It is not the same to run the **Capacitated VRP optimization for the several thousand different SKUs and simple stops routing**. Wholesale distribution, trucking, and logistics management companies need a comprehensive product and freight management function. These industries also combine inbound and outbound operations, can have simultaneous pick-ups and deliveries and offer multi-depot operations.

Forth, the capacity of the platform for handling the number of deliveries should be appropriate to the size of the company. Due to the time complexity of the problem, larger companies should pick an application (more than 300-500 deliveries a day), which is capable to work with large data sets without compromising the solution quality.

**Conclusion**

Transportation, distribution, and logistics management firms operate in a highly competitive and increasingly uncertain environment where margins are tiny and even a small disadvantage will keep the firm out of the highest levels of competition. Delivery planning optimization sometimes lags behind other, seemingly more important operations when it comes to decision making. Crucially, this is where your short-term profitability and long-term efficiency rests. The cost of choosing a poor solution will be hundreds and millions of dollars. We have seen many cases where suboptimal planning and poor execution (which are highly interconnected) sink both growing and established firms.

**Less® Platform*** can turn your data into loads within 10 minutes and compare it with your historical results! You can also test all the necessary features for your distribution model and consult with our experts on your challenges.*

Author

Vardan Markosyan is the CEO at Less^{® }Platform

MBA from the University of Chicago Booth School of Business

PhD in Economics from the Institute of Economy of NAS RA

He spent decades of research and consultancy on business process optimization and system design