Assesing the success of private labels online: differences across categories in the grocery industry

Abstract

This paper analyses online competition between private labels and national brands. Purchase data from a grocery retailer operating both on and offline are used to compute two measures of competition (intrinsic loyalty and conquesting power) for both the private label, and what this paper terms the “reference brand” (a compound of the different national brands within a category), in 36 product categories. The results show that the competitive position of the private label, relative to that of the reference brand, varies across categories and across channels. Using the framework devised by Steenkamp and Dekimpe (Long Range Plan 30(6):917–930, 1997. https://doi.org/10.1016/S0024-6301(97)00077-0) we combine the two computed measures of competition, and classify the private label as a miser, a giant, a fighter or an artisan in each channel and category. The results show: (1) that private labels significantly improve their competitive position online; and (2) that this improvement is not equal across all categories.

Appendices

Appendix 1: Details of our approach

The model in [15] is based on the construction of a switching matrix, which requires the consideration of only two observations per consumer. Given that we observe the purchase behaviour of our customers over a whole year, we apply the model in [15] to successive switching matrices to obtain the values of $\alpha_{i}$ and $\pi_{i}$ at purchase occasion level, rather than customer level. This means that we examine the repetitive or switching behaviour of every customer in our database between brands from one purchase occasion to the next.

Imagine a market with three brands, brand A, brand B and brand C. Consider that customer h purchases brand A on her first purchase occasion, brand A on the second, and brand C on the third. Further consider that customer k purchases brand B on his first purchase occasion, brand A on the second, brand C on the third, and brand C on the fourth. Our approach does not limit attention to two consecutive purchases of each customer (let us say the first two purchase occasions), but to every purchase occasion of every customer. For customer h, we consider the following switching matrices:

Purchase occasion 1	Purchase occasion 2			Purchase occasion 2	Purchase occasion 3
	Brand A	Brand B	Brand C		Brand A	Brand B	Brand C
Brand A	1	0	0	Brand A	0	0	1
Brand B	0	0	0	Brand B	0	0	0
Brand C	0	0	0	Brand C	0	0	0

For customer k, we consider the following switching matrices:

Purchase occasion 1	Purchase occasion 2			Purchase occasion 2	Purchase occasion 3			Purchase occasion 3	Purchase occasion 4
	Brand A	Brand B	Brand C		Brand A	Brand B	Brand C		Brand A	Brand B	Brand C
Brand A	0	0	0	Brand A	0	0	1	Brand A	0	0	0
Brand B	1	0	0	Brand B	0	0	0	Brand B	0	0	0
Brand C	0	0	0	Brand C	0	0	0	Brand C	0	0	1

Hence, we apply the methodology in [15] to the following switching matrix:

Purchase occasion t − 1	Purchase occasion t
	Brand A	Brand B	Brand C
Brand A	1	0	2
Brand B	1	0	0
Brand C	0	0	1

If we wished to consider only the first two purchase occasions of each customer, we would need to apply the model in [15] to the following switching matrix:

Purchase occasion t − 1	Purchase occasion t
	Brand A	Brand B	Brand C
Brand A	1	0	0
Brand B	1	0	0
Brand C	0	0	0

With our approach, we (1) take into consideration every purchase of every customer in a product category and therefore introduce the weight of each customer’s purchases in the total purchases per category, i.e., we give more weight to heavy than to light buyers within the category. Thus, we (2) are able to consider a larger number of observations in each category, and extend our investigation to a wide range of categories. Otherwise, the switching matrices for some categories would have presented too many zeros to enable estimation of the parameters.

A further aim of this investigation is to identify differences in brand power between the online and offline settings. Given that the construction of a switching matrix requires the evaluation of pairs of consecutive purchases, we need to define the terms “offline observation” and “online observation”. In our database, we find four different combinations for a pair of purchases: (1) both purchases are made offline, (2) both purchases are made online, (3) the first purchase is made offline and the second online, and (4) the reverse, the first purchase is made online and the second offline. By limiting our attention to the first two cases, where the differentiation between an offline observation and an online observation is clear, we would have disregarded many of the purchases registered on our database, since many customers switch between channels from one occasion to the next. Hence, we establish the following criterion to distinguish between offline and online observations: a consecutive pair of purchases is considered an offline observation when the second purchase is made offline, whereas it is considered an online observation when the second purchase is made online. In other words, it is the channel used for the second purchase that determines whether an observation is classed as offline or online. Thus, in determining current shopping behaviour, more importance is attached to how the customer is shopping currently than to how she has shopped previously.

We build a switching matrix per category and channel, such that for the estimation of the intrinsic loyalty and conquesting power parameters we use 72 switching matrices (36 product categories × 2 channels).

Appendix 2: The importance of the relative approach

The importance of the relative approach can be demonstrated in a specific category, say, the spaghetti category. Figure 5 shows the “absolute” market position of the different brands competing in the spaghetti category offline and Fig. 6 those competing online. Each brand is located on these “absolute” maps by its intrinsic loyalty and conquesting power estimates; while its size (circle diameter) indicates its market share.

Fig. 5

Offline brand position map for the spaghetti category

Fig. 6

Online brand position map for the spaghetti category

Direct observation of either of these maps clearly reveals the importance of the relative analysis. In the spaghetti category, the PL competes with four national brands: La Familia, El Pavo, Gallo and Barilla. For example: a simple examination of the online map (Fig. 6) indicates that the PL is doing well in terms of market share, conquesting power, where it takes first place; and intrinsic loyalty, where it takes second. However, it is worth examining the performance of the national brands in this last dimension, where we find that the leader in terms of intrinsic loyalty (Barilla) is placed second in terms of market share, whereas all the remaining brands in the market present lower values than the PL. So how is the PL actually positioned against national brands in terms of intrinsic loyalty in this market? This is not easy to answer, because there are several national brands, so it depends which one the PL is being compared with. This is precisely where the relative analysis plays its role. It can be seen in Table 5, “Appendix 3” that the RIL(PL, On) = IL(PL)/IL(RB) = 0.85/0.82 = 1.04, which means that, in global terms, the PL’s intrinsic loyalty online is 4% higher than that of the average national brand in the market. We can conclude that, although the PL ranks second for intrinsic loyalty in absolute terms, it ranks first in relative terms. To illustrate this visually, Fig. 7 depicts the relative position of the PL in the spaghetti category both online and offline.

Fig. 7

Relative position of the PL in the spaghetti category

Appendix 3: Tables

See Tables 4 and 5.

Table 4 General descriptives

Table 5 Intrinsic Loyalty and Conquesting Power Values for PL and RB