Seculus Uses AI to Predict Inventory Requirements
“Getting a product from concept to shelf can take months. By the time we realize we haven’t stocked the right amount, it’s too late to fix. Our top priority was to improve the certainty of our forecasting models. Kepler was able to help us do that.”
– Oswaldo Moreira Neto, Director of Marketing & Innovation
Seculus is a watch manufacturer that prides itself on contemporary design, versatility, and the sort of care to detail that one expects from a high-end watch company, but at a mid-level price. The company has been in business since 1960 and began a portfolio diversification process in 2005 in order to bring their offerings to a larger audience who value both form and function when it comes to their timepieces.
The schedule that Seculus used to manufacture watches is based on what they believe the demand for each specific product will be. With a catalog of more than 50 different watch styles that all require different parts, the business was dealing with more than 1,400 SKUs, making it incredibly difficult to manage that amount of data in order to advise the production decisions.
The in-house data team at Seculus was tasked with building predictive models that could determine which of their watch models would have the highest sales volume nine months in advance so that they could order the parts needed to fulfill those orders.
Especially important in this process was identifying holiday sales volumes as, in the past, they had either underestimated their stock requirements, or were left with overstock at the end of the season.
In both situations, they needed to optimize supply chain processes that directly impact gross margins. For this company, precision was not just important to their products, but to the means of creating and distributing them, as well.
The issue that Seculus had was not a lack of data, but rather the amount of data that they needed to manage and understand in order to make the right decisions. In an effort to use that data to its fullest, the Solutions Team at Stradigi AI worked closely with Seculus in order to better understand the nuances of their requirements, as well as the data that they had available.
From there, the teams on both sides worked together to develop four prediction models that could use their data to uncover outcomes that could be used to look forward. These models were based on previous sales and watch feature data. Seculus was then able to add their data to Kepler, which ran models for them that accounted for seasonality, product lifecycle, and swings in demand. Just a few weeks later, Web and brick-and-mortar retail data was added in order to further enhance SKU demand predictability.
The results began making themselves evident very quickly. Within the first week, a product demand forecasting model had been put in place that was returning results with 86% accuracy. They began to prepare for the upcoming holiday season with a lot more confidence in the decisions that they had made to accurately order parts and fulfill production demands.
The data output allowed Seculus to actually reduce the total number of SKUs to a much more manageable number, allowing them to more clearly understand their requirements with less of an investment in time.
What has been very rewarding for the data analysis team at Seculus has been their ability to build their own machine learning models, which is now allowing them to undertake additional use cases in order to further optimize processes over a larger part of their organization.
> Watch Manufacturing
> Tabular regression with demand forecasting