For some years, the terms AI and ML have been on the rise, as well as their interest. The decrease in costs of cloud services, the massification of information, and the proper use of these make a significant competitive advantage for many companies, who started their career being data-driven, and of course, seeing big companies like Microsoft, Apple, Google, Amazon, Netflix, and others taking advantage over their competitors with great use of data. They have considerable volumes of data and lots of models leveraging it, which aligns with all their business processes.
For my part, I have always been passionate about data, and so I started my working life a few years ago. I started as a web developer, followed by epidemiology data consolidation, and later, I spent some years in revenue management data for an air cargo company. Also, I dedicated some years to the financial sector and later arrived at the real sector, a very special one: fast fashion retail. The most exciting thing when I arrived here was to have a large amount of information to exploit and take advantage of sales information of thousands of SKUs per day, more than 500 points of sale, complete logistics and supply chain processes, in other words, the Disneyland of data.
“The most exciting thing when I arrived here was to have a large amount of information to exploit and take advantage of sales information of thousands of SKUs per day, more than 500 points of sale, complete logistics and supply chain processes, in other words, the Disneyland of data”
One of the challenges faced here was to design a store-level optimization model that will help maintain an ideal stock, suggesting that SKUs be replenished in each store daily so that the points of sale always have the necessary stock of products to minimize sales due to lack of inventory. As a company of considerable size, there were many different sales patterns and factors to consider for an ideal sales forecast. We ran multiple models by store clusters and had terrific results. Everything was perfect until the peak season arrived when sales were the highest of the year regardless of store type, location, and other variables. The model would emerge very well the restock; however, the logistics companies collapsed and needed the operational capacity to send that large number of garments to each store in the reduced time expected. Indeed, our beautiful model failed, not because of a wrong forecast, but because of a result that could not be executed in the expected time to external factors.
We manually adjusted the model by adding an incremental offset before reaching those shipment peaks, and that is how we solved it. However, we were left with a great lesson: leveraging data to improve the business will always be the way to go, but it must start by fully understanding the business and finding a solution, not vice versa.
The same thing happens today in many companies. It is believed that the best data model, the most accurate, and the most powerful is the one that will help solve their problem. Nevertheless, in the eagerness to obtain results, many factors must be mapped and linked to a human reality, which should not be ignored.
