The provided guidelines can be used by retailers to improve their wider demand forecasting process. The DF technique evaluation results in chapter 8 showed how large the impact of using the right forecasting techniques can be and this chapter enables retailers to take action on those results by guiding them not only in the DF technique selection process, but also in improving the wider forecasting process.
Future research could also conduct a survey to gain insight into the relative weights grocers give to the different DF technique selection criteria in different situ- ations. For example in case of daily forecast supporting replenishment decisions, ac- curacy might be slightly less important and speed more important than for a weekly forecast, since any errors can be corrected the next day already. Future research could also include guidelines on how to asses the return on investment of adjusting the forecasting process (such as implementing new forecasting techniques).
Future research could validate the DF improvement process in practice, for ex- ample by applying it during various case studies. Future research could also examine to what extent the existing forecasting evaluation and audit frameworks are still ap- plicable now, whether they can be applied across industries and whether extensions are required to take into account recent advances in forecasting practices.
Part II
Dynamic Pricing of Perishable
Food Products
6
Dynamic Pricing
Fundamentals
As Warren Buffett once said: “Price is what you pay, value is what you get”. This quote illustrates that it is important to always consider customer value when pric- ing products. Since customer value decreases when perishable (grocery) products deteriorate, it makes sense to adjust prices dynamically over time. This chapter in- troduces the reader to the fundamentals of dynamic pricing. Section 6.1 describes different pricing strategies and covers how DF and DP are interrelated. Section 6.2 describes the different DP problem dimensions. Section 6.3 covers related work on dynamic pricing for perishable products.
6.1
Pricing Strategies
Sellers can adopt a wide variety of pricing strategies. On a high level, these strategies can be categorized into two main types [20]:
• posted-price: price is set by the seller and is a take-it-or-leave-it price
• price-discovery: price determined through a bidding process (e.g. auction) Within the posted-price mechanism, prices can either be static (fixed over time) or dynamic (changing over time). The field of dynamic pricing (DP) focuses on finding the optimal product prices over time to maximize profit for the seller.
The height of the price is one of the factors that influence customers’ demand for a product: generally the price-response function is downward sloping. Price elasticityeis defined as the fraction of demandQchange over priceP change.
e= dQ/Q
dP/P
For example, when a 2% price increase results in a 1% demand decrease, the price elasticity is -0.5. So if demand for a product changes relatively heavily in response to price changes, that product is said to be relatively elastic. Examples of relatively inelastic products are necessity products like water and bread. The elasticity of a product can differ on the short-term versus on the long-run. For example short- term elasticity for airline travel is low because a travel need exists then, but it is
CHAPTER 6. DYNAMIC PRICING FUNDAMENTALS 57
External Factors Sales History
Demand
Model ForecastsDemand DF Pricing Factors Price Constraints Pricing
Model SuggestionsPrice DP
Figure 6.1: Visualization of DF, DP and their link
higher in the long-term. Setting a certain price now does not only affect current demand, but it might also influence future demand through customers’ reference price, which is their perception of what they have paid for that product in the past [15]. But customers generally don’t have a reference price for all items they buy, so retailers can use that to their advantage. Key value items (KVIs) are the items that mainly drive customer value perception and for these items competitive positioning is especially important and these items should be sharply priced [40].
Another pricing strategy that differs from dynamic pricing is personalized pricing. With dynamic pricing, prices change over time, but all customers will pay an equal price at the same moment in time. However, with personalized pricing, retailers use their knowledge about customers to differentiate pricing on an individual level. This differentiation can for example be based on loyalty card data, but also on channel or browsing history. So dynamic pricing and personalized pricing are two different concepts, but they can also be applied in combination.
Dynamic pricing (DP) is closely related to demand forecasting (DF) and this relation is depicted in figure 6.1. Using the DF techniques discussed in section 2.3 a demand model is derived from the sales history and external factors, which can then be used to produce a forecast for future periods. The demand model is also the relationship between DF and DP, since the pricing model that is used for DP optimization depends strongly on the demand model, which also includes a price elasticity component. In addition, the pricing model takes into account price constraints (see section 6.2) and pricing factors such as remaining inventory levels and supply uncertainty. The pricing model is then used to provide optimized suggestions for product prices over time. Dynamic pricing is a form of prescriptive analytics, since it not only predicts what will happen in terms of demand, but suggests (or even automatically performs) an action which in this case is a price adjustment.
CHAPTER 6. DYNAMIC PRICING FUNDAMENTALS 58
For example in the case of e-commerce, a/b testing could be employed, to randomly serve part of the customers with the optimized price and the other part with the regular price. For a retailer with brick-and-mortar stores, a few pilot stores could be selected that operate with the optimized prices. However, testing out a pricing strategy in practice does come with quite some risks, so an interesting alternative is to do simulations. One downside of simulations is that the results depend on the validity of the underlying simulation model.