How AI enables agile business operations, elevating analytics to board-level conversations.
While the world watches the many applications of machine learning unfold — Google’s new AI assistant and Uber’s driverless cars, to name a few — a fundamental shift in retail operations is taking place. This shift is driving a bifurcation of the retail industry: companies that will still be around in a few years and those that will not. Advancements in artificial intelligence (AI) have led to a regime change from agile business operation to “just-in-time” operation — the ability for a business to use improved forecasts to go from concept to consumer (material sourcing, production, marketing, shipping, etc.) at exactly the right time.
The moment to either adopt just-in-time operation or perish is rippling out from the major players and into the entire ecosystem.
Big retailers have seen the water rising for a while, subsequently shifting their investments to bolster their data science teams. Amazon announced “anticipatory” shipping, which predicts products you might want to order and sends them to a distribution facility near you. The moment to either adopt just-in-time operation or perish is rippling out from the major players and into the entire ecosystem. In its ongoing battle with Amazon, Walmart now provides their suppliers with historic inventory data and has “told suppliers it ultimately wanted orders delivered on time 95 percent of the time or they would pay a fine.” (Reuters, January 2018)
Need another example of how better forecasts lead to better business operation and more market share? Take a look at the clothing brand, UNIQLO. By pushing for just-in-time operation, they are securing record profits after turning products that used to be released once seasonally, into A/B tested commodities. (Reuters, January 2018)
The adoption of just-in-time operations is not limited by interest, but rather by ability. Walmart can hire an advanced data science team, but what about their suppliers?
Just-in-time operation is driven by forecasting future values (sales, number of customers, how many widgets needed, etc.). This class of problems is commonly referred to as time series modeling, which has been very resistant to automation. And classical approaches to time series are often simplistic (limited to just the quantity of interest and the date - with no other drivers), difficult to implement, and time-consuming because finding the right methodology for your problem can be challenging, even for an expert data scientist.
Most organizations struggle to hire enough data scientists to keep up with their internal demand.
To make matters worse, time series efforts for retailers are plagued by two major challenges: scale and system change. The challenge of scale is the sheer number of predictions required for each product, color, store, etc. A change of product lines, market conditions, and business strategies mean that these problems require constant effort and must be revisited over and over again. Most organizations struggle to hire enough data scientists to keep up with their internal demand. Companies are searching for ways to scale their internal forecasting teams, or are starting to supplement this shortage with software and/or consultancies. Retail has the added problem of competing with organizations like Google, Uber, and Facebook for the same data science talent.
What do we do after seeing the pattern of a retailer improving their ability to make predictions, getting closer to just-in-time operations, and benefiting from annexing their competitors’ market share? Admittedly, my world view has been impacted by hundreds of conversations with hedge funds and banks over the past few years, but my first step was to start reading SEC 10-Q filings to see if retail boards are talking about forecasting as a competitive advantage. I leave it to the reader's own curiosity to explore more, but here is a snippet from The Home Depot’s most recent SEC filing:
“We continue to drive productivity and efficiency by building best-in-class competitive advantages in our information technology and supply chain. These efforts are designed to ensure product availability for our customers while managing our costs, which results in higher returns for our shareholders. Given the changing needs of our customers, our goal is to create the fastest and most efficient delivery capabilities in home improvement.”
At DataRobot, we are meeting this challenge head on with automated machine learning for time series problems. Our approach is to create competition among different algorithms, and quickly identify the best one to get the forecasts required to drive just-in-time operations. Learn more about DataRobot Time Series here.
About the Author:
Jay Schuren is the General Manager of our Time Series activities at DataRobot. Jay joined DataRobot through the acquisition of Nutonian in 2017, where he lead the customer facing data science team and focused predominantly on time series use cases across all industries. Jay has over 10 years of experience and a PhD from Cornell University.