In their research paper “Optimizing apparel production order planning scheduling using genetic algorithms“, Wong and others propose leveraging genetic algorithms to address uncertainties and random variables associated with manufacturing order scheduling at the factory level.

The paper investigates the challenges of order scheduling problem at the factory level, specifically a manufacturing setup in which each production process corresponds to a unique shop floor comprising of one or multiple assembly lines. 

From the mathematical perspective of an objective function, the primary objective is to maximize the total satisfaction level of orders’ actual competition times while minimizing the total throughput time. These are achieved by determining which assembly lines to use for an order and the optimal time to start the production process of each order. 

I started thinking about extrapolating this into other supply chain and logistics scenarios. As you can imagine, opportunities to use it beyond this specific use case in the supply chain world are plenty. An example can be warehouse fulfillment flow optimization.

But if you go through the mathematical formulation of the problem in the paper, you will find some interesting applications beyond supply chain and operations. Examples are financial service and marketing campaign management. Let us explore the application in marketing campaign management.

We tend to primarily think about marketing campaigns from an A/B testing perspective, as far as analytics goes. These tests are primarily meant to capture the effectiveness of a campaign. But how about optimizing the right time to launch and campaign and leveraging the right marketing levers?

If you analyze the high-level components of the factory level problem in the paper, the two key variables are: 

• (a) Which assembly lines to use (Which marketing levers will be useful/impactful)
• (b) When to start processing these orders on the factory floor (What is the right time to launch the campaign)

The algorithm can be granularized to accommodate additional aspects like the optimal duration that the campaign should run. The challenge in this, or other similar applications, is not the choice of the algorithm of formulating the problem at the high level. Or identifying objectives and variables.

The challenge in this specific example is identifying and quantifying some of the variables. Like marketing levers. If you launch a campaign for a new desktop monitor, which levers will be more effective across which channels, demographics, regions……the list can be long. Defining, collecting, and harmonizing this data set will be the most difficult part of this experimentation.

Nevertheless, this can bring some good solid science into marketing campaigns across industries that offer innovative products. This will work for functional products as well, but the benefit vs the investment might not be attractive.