In one of my jobs early in my supply chain analytics career, I did a rotation in Transportation planning and analytics for a leading workspace solutions manufacturer. The project I was assigned to entailed optimizing inbound shipments from suppliers within trucking distance from our manufacturing plants. One key lever of this initiative was transportation route optimization.

The project

Many of our raw material suppliers frequently shipped small volumes (2-3 pallets) to our plants, leveraging modes like LTL, expedited cargo, etc. Hence, the idea was to plan these inbound as multi-stop FTLs, thereby significantly cutting the inbound logistics cost. As you can assume, the key lever in this entire plan/architecture was transportation route optimization.

The outcome: Resistance to change led to the slow and challenging implementation

The year was 2008, and the American economy had just hit the trough of recession. If you can recall, the manufacturing sector in MidWest was brutally hit. Companies were laying off in numbers across departments. It was a challenging year, and any efficiency improvement initiative focused on manufacturing, or transportation was being viewed as an effort towards workforce reduction.

One of the biggest challenges with the conventional approach of transportation optimization is that for drivers and fleet managers, who are averse to change, it is easy to blow holes in the solution. They can find hundred different ways to “demonstrate” that the routes generated will not work in the real world.

A traditional fleet or route optimization solution only incorporates a few essential inputs in the optimization process and creates routes based on those criteria. While those aspects are essential inputs for route or fleet optimization, unfortunately, the world is not Black or White. There is a multitude of factors that may impact the daily routes of your fleet that are not incorporated in classic route optimization solutions.

When we started the implementation pilot for the program, looking at the GPS data, we could see that most drivers were not following the prescribed routes. We started interviewing these drivers daily and received inputs like the following:

(1) Supplier not ready when they get there, leading to delay and that forced them to skip few pickups.

(2) The pickup details specific to dimensions were not provided accurately by the supplier (ex: Supplier had four pallets instead of 2 entered on the portal)

(3) Traffic events

(4) Vehicle breakdowns

(5) Weather-related issues

(6) Last-minute or urgent pickup requests that drivers need to accommodate. In such circumstances, drivers used their tribal knowledge to accommodate those on their route. This accommodation, however, is not always optimal.

What could we have done better during the planning and/or implementation?

The answer is -Nothing. This is precisely why I do not advocate for a majority of popular off-the-shelf VRP tools in the market. There are so few input variables in these tools, and the architecture does not allow for incorporating some of the issues identified above.

Are these off-the-shelf solutions helpful at all?

Yes, they are, but I think only for strategic-level analysis. These solutions may be suitable for initiatives like fleet optimization, driver headcount reduction, transportation network optimization, or a high-level evaluation of the impact of network redesign on last-mile transportation. However, as far as periodic route optimization goes, I suggest leveraging a customized solution if you have the resources to do so.

I find it very frustrating when I see companies leveraging an off-the-shelf tool for periodic route optimization in today’s age of easy access to pertaining technologies. Technologies that can incorporate most of the sh*t the world can throw at your routes. Technologies that can help you build customized solutions. And even powered by Generative AI.

What will a customized solution look like, though?

The Big data age solution: Intelligent dynamic route optimization

The IoT era allows us to harness real data generated by various sources, and this data, combined with other real-time data points related to traffic, weather, customer preferences, etc., can allow the creation of dynamic routes-routes that evolve in real-time and optimize as the real-time data comes into the system. An example is shown in the illustration below-a last-minute urgent pickup request comes in, and the ML algorithm recalculates the route.

Computing power, availability of data, and connectivity technologies have opened the door for developing Intelligent route optimization solutions. These solutions are created at a very high level, pairing real-time data generated from various sources with machine learning to optimize real-time routes. Examples of data points that the ML algorithm can look at are our truck speed, GPS location, route traffic, weather, destination, customer requests for changes to delivery time, etc.  The critical aspects of a solution like this will be:

• Self-learning system
• Optimal fleet selection
• Optimal delivery assignments powered by artificial intelligence
• Considers real-world and the ground fuzziness
• On the road driver app