What was the impetus for this research?
The impetus for this research is twofold, both personal and professional. As a consumer, I get frustrated when my family and friends tell me about a good flash sales deal, and I cannot get it because the item sold out too fast. I want to be able to take my time to go to the website, read reviews, see what people are saying about the deal, compare prices, and then calmly make my decision—to buy the item or pass on the deal.
As a supply chain management scholar, I find the flash sales model fascinating. Online retailers can provide a lot of inventory liquidity in such short amounts of time and capture new consumers who are hunting for bargains. It is not surprising, therefore, that flash sales markets have become an integral part of online retailing. Just look at the importance of Amazon Prime Day and Alibaba's Singles Day. Those are their biggest selling days of the year! I was interested in understanding ways retailers could ensure that the flash sales deals will be available at the right price for the right amount of time so that consumers can take their time to make a decision. It is important not to alienate consumers, because, on the Internet, it is too easy for consumers to switch and never come back.
What makes forecasting and product pricing for flash sales so challenging?
The flash sales model is built on the principle of scarcity. If either inventory or time runs out, the deal is gone, and it may not come back. This generates "competition" among consumers, who will "fight" to get a good deal. So, it all depends on how strong these scarcity effects (that is, this competition among consumers) will be. It is very difficult to predict the strength of these effects, though. A price increase or decrease can significantly alter the market, driving consumers into it or out of it. And it becomes even worse when consumers interact with each other, not only by observing each other's purchases but also by chatting with each other on discussion forums. The conversation may or may not be very influential. Other sales models that hinge on social interactions among consumers do not display such strong scarcity effects because they usually involve inventory replenishment and offer consumers plenty of time to make an informed decision.
What is the Bass demand growth model?
The Bass demand growth model is a model of communication. It helps predict the diffusion of an idea, a product, or a service among a population. It divides the population into two groups, in terms of their susceptibility to be influenced to adopt the idea, product, or service. There are those people who are influenced by factors that are external to the social system, for instance, mass media communication and email blasts from retailers. And there are those who are influenced by factors that are internal to the social system, for instance, the ability to observe others making the adoption. It relates to flash sales because inventory sales, in this business model, usually follow a pattern that is typical in the Bass model; there will be those bargain hunters who will become aware of a deal from external sources and will buy the item first, and then there will be those latecomers who only get to know about the deal after observing those early purchases.
Why did you decide to focus on looking at how consumer sentiment affected demand in the flash sales market?
The traditional Bass model has a lot of predictive power but relies heavily on consumers' ability to observe other consumers' purchases. But what about the influence among consumers via discussion forum posts? If you read the conversation, you will find people displaying a huge interest in a particular deal. But you may also find others bashing that deal, for instance, saying that they can find the same product elsewhere at a much lower price or that they just hate that product. Research shows that online conversations may, or may not, be extremely influential, and that is something that was not captured well in the traditional Bass model.
Can you describe the data set that you used for creating your model?
We looked at thousands of deals offered over seven years in the FSM of one of the leading online retailers in the United States. For each deal, we had the amount of inventory that was sold, how fast it was sold, if it sold out, and its price. We also had the consumer posts in the discussion forum associated with that deal. We used a proprietary software that rates the posts as conveying either a negative, positive, or neutral sentiment. (There are many tools available out there, so one must be very careful when choosing one to ensure that they are getting a good measure of sentiment.) We asked three people to also rate a fraction of the posts' sentiments in our data set and compared their measures with the measures provided by the software to ensure the measures we were using were reliable. We then modified the traditional Bass model to accommodate a parameter (the consumer sentiments). Our intervention predicts a growth pattern that deviates from the one you would get by shifting the demand (the sales are expected to move slower or faster depending on how negative or positive the consumer sentiments are).
How can this research be used by practitioners?
We provide practitioners with a tool to calibrate a forecasting model for their flash sales. They already have their sales data. All they need to do now is to measure sentiments, using one of the numerous tools that are available out there, and then enter the data into our model. We recommend that practitioners monitor their flash sales in real time, including capturing consumer sentiments, and calibrate a growth model early in the sale. That model can assist them in making adjustments to their prices. For instance, if a flash sales deal is not selling well and the consumer sentiments are too negative, then it may be necessary for them to make deep price cuts. If sentiments are not that negative, though, drastic price cuts might not be a good idea because that might cause the flash sale item to sell out too fast. In short, there is this interesting interplay between sentiments and pricing, with which practitioners should become more familiar.
What would you say is the key takeaway message of the research?
There is a way to make FSM (and similar business models) much more efficient and effective by leveraging the power of social media information. In our case, that would be information from consumer sentiments conveyed through discussion forum posts.