June 8, 2016 | Written by: Michael Wong
Categorized: Merchandising & Supply Chain
Share this post:
Preface: Per IBM’s 2015 C-Suite study, “…CEOs put technology at the top of the list, as they have for the past four years. But now, for the first time ever, the other members of the C-suite also see technology as the main game-changer.” This Q&A is part of a series that secures ideas (of C-Suite leaders and professors) on innovative Business/IT solutions which have the potential to be deployed across a variety of different industries for positive change. In this article, Dr. Kris Ferreira explains how machine learning and optimization techniques can help drive revenues in under 18 months.
Q. While many C-Suite study executives are rightfully excited about the potential of machine learning and optimization techniques, how can CIOs convert this enthusiasm into purposeful decisions and ideally positive business results?
A. What’s the value to a company if a CIO can introduce a solution which recommends price increases that elevate revenues without jeopardizing profits or the company’s brand value proposition? At one retail firm, initial findings have demonstrated what happens when a company deploys a fairly sophisticated pricing tool that is integrated into their daily operations without causing undue administrative burdens on staff.
Rue La La is in the online fashion sample sales industry, where they offer extremely limited-time discounts (“flash sales”) on designer apparel and accessories. One of the firm’s main pain points is pricing and predicting demand for products in the highly cyclical fashion industry. To tackle this challenge, we used machine learning techniques to estimate historical lost sales and predict future demand of new products [Ferreira, Kris J., Bin Hong Alex Lee, and David Simchi-Levi. “Analytics for an Online Retailer: Demand Forecasting and Price Optimization.” Manufacturing & Service Operations Management 18, no. 1 (Winter 2016): 69–88]. First, we used data mining and exploratory data analysis to identify hypotheses on what attributes of their products and/or sales events drive sales. Second, we used machine learning algorithms that used these attributes from their historical sales data to predict future demand of new products, and we developed a multi-product price optimization tool that set prices to maximize profitability. Third, we ran field experiments to test the impact of our tool which showed an approximate 10% increase in revenue with little impact on cost for the products whose prices were changed. From start to finish, the project duration was approximately 18 months.
What should pique the interest of your readers is this retailer’s business complexity of having to predict the demand of products which have never been sold in the past. If Rue La La’s preliminary results showed an increase in revenue of 10% with costs contained; imagine what a similar business-IT solution could mean for those companies which have robust demand curves of historical product sales?
Q. Of the several machine learning techniques evaluated, why do you think regression trees performed the best and how might their application in the retail/fashion industry be applied in other Fortune 500 sectors such as CPG, healthcare or utilities?
A. In some respect, regression trees are able to determine – for each new product that Rue La La sells – the key characteristics of that product that will best predict demand, and the trees use only the demand of products sold in the past that also had those same key characteristics to predict future demand. The trees essentially give Rue La La a way to automatically identify similar products they’ve sold in the past for each new product that they want to sell in the future, and then they use the historical sales of these similar products to predict demand of the new product. Regression trees have actually shown to be very powerful predictors in many settings, certainly not unique to the retail/fashion industry. That said, the best machine learning technique to use depends on the problem at hand, the data available, and the importance of interpretability. Finally, the CIO needs to constantly think about how the IT solution potentially impacts the overall business. For example, at Rue La La, while they wanted to elevate profits, a key business requirement was to also consider their goal of growing the total number of customers. So they had a different business goal compared to some firms which are purposely segmenting target clients for higher profits even if they potentially are going to lose some from their total customer base. For Rue La La, given their ambitious growth goals, they had to balance profit increases with this other key performance indicator.
Q. So besides Fortune 500 companies, can smaller entities such as non-profits or start-ups leverage these business-IT solutions in their operations?
A. One requirement for using regression trees and many other machine learning techniques is data. Without enough data, these techniques are unable to learn valuable insights and make predictions. Therefore very small companies and start-ups may not be able to unleash the full power of these machine learning techniques until they grow to the size where they can generate enough data. With that said, it’s not just the quantity of data that matters: storage, accessibility, and data integrity are also key pre-requisites of a data-driven business strategy. It’s prudent for any company – including non-profits and start-ups – to think carefully about their long-term IT strategy to make sure that they’re capturing the appropriate data in the best way possible for later use.
Dr. Kris Ferreira is an assistant professor of business administration in the Technology and Operations Management (TOM) Unit at the Harvard Business School. In her research, Professor Ferreira focuses on online markets and retail, seeking to build operations management models that inform practice, as well as to provide theoretical insights. In particular, she employs a combination of machine learning and optimization techniques to help companies use their data to make better tactical and strategic decisions. Her work with the online retailer Rue La La received the 2014 INFORMS Revenue Management and Pricing Section Practice Award and was named a finalist in the 2015 Innovative Applications in Analytics competition.
Professor Ferreira earned her PhD in operations research at the Massachusetts Institute of Technology and her BS in industrial and systems engineering at the Georgia Institute of Technology. Before entering graduate school, she was a supply chain consultant for Alvarez & Marsal and a project manager for UPS Supply Chain Solutions.
Transform your retail business with engaging brand experiences that strengthen your customer relationships. Check out IBM Retail Industry solutions at http://ibm.com/retail