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. Upon visiting Rue La La’s website (www.ruelala.com), the customer sees several “events,” each representing a collection of for-sale products (“styles”) that are similar in some way.  At the bottom of each event, there is a countdown timer informing the customer of the time remaining until the event is no longer available; events typically last between one and four days. Flash sales businesses like Rue La La aim to create a feeling of urgency and scarcity of products by offering great deals but for a limited time and with limited inventory.

One of Rue La La’s main challenges is pricing and predicting demand for items that it has never sold before (“first exposure” items), which account for the majority of sales. Figure 1 shows a histogram of the sell-through (% of inventory sold) distribution for first exposure items in Rue La La’s top five departments. For example, 51% of first exposure items in Department 1 sell out before the end of the event, and 10% sell less than 25% of their inventory. Department names are hidden and data disguised in order to protect confidentiality. A large percent of first exposure items sells out before the sales period is over, suggesting that it may be possible to raise prices on these items while still achieving high sell-through; on the other hand, many first exposure items sell less than half of their inventory by the end of the sales period, suggesting that the price may have been too high. Unfortunately, prior to the beginning of the event, it is very hard to predict where on this histogram a product’s sell-through will fall. This challenge motivated the development of a pricing decision support tool, allowing Rue La La to take advantage of available data in order to maximize revenue from first exposure styles.

Figure 1: First Exposure Sell-Through Distribution by Department

Our approach is two-fold and begins with developing a demand prediction model for first exposure styles; we then use this demand prediction data as input into a price optimization model to maximize revenue. The two biggest challenges faced when building our demand prediction model include estimating lost sales due to stockouts, and predicting demand for styles that have no historical sales data. We use clustering and machine learning techniques to address these challenges and predict future demand. Figure 2 shows a summary of features we used to develop our demand prediction model. Interestingly, we find that one of the most important features in predicting demand is the “relative price of competing styles,” where competing styles refers to a similar type of products in the same event; this metric is calculated as the price of the style divided by the average price of all competing styles, and it is meant to capture how a style’s demand changes with the price of competing styles shown on the same page. By including the relative price of competing styles as a feature, we are essentially considering the average price of competing styles as a reference price for consumers.

Figure 2: Summary of Features Used to Develop Demand Prediction Model

Regression trees—an intuitive, yet nonparametric regression model—prove to be the best predictors of demand. To the best of our knowledge, this is the first application of regression trees used for demand prediction. Extending beyond the flash sales industry, our intuition is that regression trees may make effective demand prediction models for (i) new products, and (ii) products whose price can be considered a signal of quality.

We then formulate a price optimization model to maximize revenue from first exposure styles, using demand predictions from the regression trees as inputs. In this case, the biggest challenge we face is that each style’s demand depends on the price of competing styles, which restricts us from solving a price optimization problem individually for each style and leads to an exponential number of variables in the price optimization problem. Furthermore, the nonparametric structure of regression trees makes this problem particularly difficult to solve. We develop a novel reformulation of the price optimization problem and create an efficient algorithm that allows Rue La La to optimize prices on a daily basis for the next day’s sales.

To implement our demand prediction model and price optimization algorithm, we developed and implemented a fully-automated pricing decision support tool at Rue La La. It is run automatically every day, providing price recommendations to merchants for events starting the next day. The entire pricing decision support tool is depicted in the architecture diagram in Figure 3.

Figure 3: Architecture of Pricing Decision Support Tool

To estimate the tool’s impact, we developed and conducted a field experiment on approximately 6,000 styles from January through May 2014 to address the following two questions of particular interest to Rue La La:

(1)    Would implementing the tool’s recommended price increases cause a decrease in demand?
(2)    What impact would the price increases have on revenue?

The results of our field experiment suggest that raising prices only negatively impacts demand for very low-priced styles (price < ~$50). For higher priced styles, it appears as though our model recommends price increases on styles where the increase in price would not negatively impact demand. This is great news for Rue La La, who was concerned that increasing prices—although potentially increasing revenue—could negatively affect long-term customer demand. Furthermore, the results of our field experiment show an overall financial impact of approximately a 10% increase in revenue on the styles in the field experiment.

Our collaboration with Rue La La shows that combining machine learning (predictive analytics) and optimization (prescriptive analytics) into a pricing decision support tool has made a huge financial impact on Rue La La’s business. We hope that the success of this pricing decision support tool motivates retailers to investigate similar techniques to help set initial prices of new styles, and, more broadly, that researchers and practitioners will use a combination of machine learning and optimization to harness their data and use it to improve business processes. Furthermore, our work has shed light on the unique challenges present in the relatively new and growing flash sales industry. As this work illustrates, there is huge potential for academicians and practitioners to work together to develop new operations management models and techniques tailored to this industry, and ultimately guide the industry’s future growth.

Click the following links to see videos of Murali Narayanaswamy, VP Pricing & Operations Strategy at Rue La La, share his thoughts on the project.

https://www.youtube.com/watch?v=ahOHAsECeIw&feature=youtu.be

https://www.youtube.com/watch?v=lc4wV6O_YDA&feature=youtu.be

https://www.youtube.com/watch?v=AzJhAxkpkEU&feature=youtu.be

REFERENCE

Ferreira KJ, Lee BHA, Simchi-Levi D (2015) Analytics for an online retailer: Demand forecasting and price optimization. Manufacturing Service Operations Management 18(1):69--88.