Is paid search overrated? When bricks-and-mortar-only retailers should not use paid search

Abstract

While bricks-and-mortar-only retailers do not offer online purchasing, they often take advantage of multi-channel management strategies to reach consumers in a pre-purchase phase. We investigate whether paid search can increase the sales of brick-and-mortar retailers who promote their offers via an informational website. Although a sizeable one third of all retailers still trade without an online-shop, previous work has been silent about the effects of paid search for them. We make use of a randomized field experiment and an end-to-end tracking mechanism to investigate the cross-channel behavior of individual consumers. Our empirical results suggest that, whilst paid search increases the number of potential customers through enhancing the reach of marketing initiatives, store sales are not increased. We conclude that customers who search online to buy offline primarily use paid search as a navigational shortcut to the retailer’s website. Consequently, bricks-and-mortar-only retailers seeking to increase store purchases should approach paid search with caution.

Appendices

Appendix 1: Top 100 Search Terms

In accordance with Rutz et al. (2011) we made use of generic search terms as well as brand-specific requests and placed ads for both categories in our experiment. Furthermore, we excluded search requests containing the retailer’s name to enhance the probability of new customers being exposed to the crafted ad. The list below provides the 100 search terms which received the most clicks in the course of the experiment. Please note that search terms were translated from German to English. Location names were replaced by “<location name>“. Naturally, furniture brand names (marked with the prefix “b:”) were not translated. A “+” indicates the applied modified broad match option of Google AdWords.

armchair	bench +online	corner sofa	furniture +online	shelf
b: bretz	bookshelf	corner wardrobe	glass tables	side table
b: hülsta	bowls	couch	guest bed	sideboard
b: jensen	box spring bed	cushion	henders hazel	single bed
b: kare	box spring sofa	cutlery	innerspring mattress	slatted frame
b: kare + < location name>	bunk bed	decoration	kitchen table	sliding door cabinet
b: loddenkemper	bureau	dining room table	leather sofa	sliding door cabinet +cheap
b: musterring	cabinet	dining set	library	sofa bed
b: rolf benz	canopy bed	dining table	lounge suite	solid wood cabinet
b: stressless	cantilever	dishes	loungers	table
bar table	CD shelf	double bed	mattress	television armchair
beanbag	coffee table	dressing table	mattress + < location name>	television furniture
bed	cold foam mattress	DVD shelf	metal bed	upholstered beds
bed + < location name>	commode	extending table	pans	upholstered furniture
bed +buy	console table	frame + < location name>	rattanbed	upholstered furniture + < location name>
bed +cheap	corner bench	frame +buy	recamiere	wardrobe
bed +online	corner bench group	functional bed	relax arm chair	wardrobe + < location name>
bedstead	corner cabinet	furniture	relax loungers	wardrobe +buy
bench	corner display case	furniture + < location name>	room divider	wing chair
bench + < location name>	corner shelf	furniture +cheap	round tables	wooden bench

Appendix 2: Experimental Design: Grouping Combinations

To allow for valid claims in regard to the offline impact of paid search, a matched-subject design is applied. Grouping was performed based on two key performance indicators, namely (1) advertising reach and (2) revenue per customer. In order to determine the most appropriate candidate group all possible grouping combinations were evaluated based on the performance indicators. The Table below depicts all possible grouping combinations. We chose the grouping (BDF-ACE) as this grouping yielded the lowest group differences in terms of advertising reach as well as revenue per customer. To prevent selection biases, we chose the group BDF-ACE instead of group ACE-BDF as this group yields a lower but statistically insignificant (p = 0.854) advertising impact (−1.2%) as well as a reduced but statistically insignificant (p = 0.455) revenue per customer (−0.3%).