The Effect of Sales Assistance on Purchase Decisions: An analysis using retail video data

Abstract

We investigate the role of sales assistance in driving customer’s purchase decision using unique observational video data from a cosmetics retail chain. The data contain visual descriptors of customers and their in-store activities including the time they spent in interacting with salespersons (sales assistance), and are linked to their purchase decisions. Our empirical specification is based on the process of customer deciding to engage in sales assistance to acquire information meaningful toward purchase decision. Thus, we treat sales assistance as endogenous construct, and employ a control function approach to correct for this endogeneity using instruments pertaining to the supply of sales assistance. Our analysis quantifies the role of sales assistance in driving a customer’s purchase decision, and it shows that the effectiveness of sales assistance diminishes with its amount. In addition to highlighting the importance of sales assistance towards purchase decisions, our results also quantify how the retailer can influence sales assistance by increasing the availability of salespersons. We also examine the effect of sales assistance on in-store search carried out by the customer. Finally, we offer number of context specific insights into the heterogeneity of customers’ shopping and purchasing behavior.

Appendix

We offer a brief derivation of the model employed for purchase decisions in Eqs. 3 and 4. We begin by assuming an underlying random utility specification such that the indirect utility from shopping at the store is described by

$$ \bar{v}\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)=\left( \frac{\theta_{i}}{\rho-1}\right)\left( \frac{p}{\psi_{i}}\right)^{1-\rho}-\frac{e^{-\eta y_{i}}}{\eta}. $$

Correspondingly upon non-purchase the customer obtains

$$ v_{0}=\left( \frac{\theta_{i}}{\rho-1}\right)p_{0}^{1-\rho}-\frac{e^{-\eta y_{i}}}{\eta}. $$

In the above y represents income, ψ_i the customer’s valuation for the stores offerings, and p quality adjusted price. By assumption ρ < 0 to make the model coherent. The choice to purchase is then

$$ \delta\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)=\left\{ \begin{array}{l}1\\ 0 \end{array}\right.\begin{array}{c} \text{if }\bar{v}\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)\geq v_{0}\\{\text{otherwise}} \end{array} $$

with corresponding probability $\mathbb {E}\left [\delta \left (p,y_{i},\psi _{i},\varepsilon _{i}\right )\right ]=\pi \left (p,y_{i},\psi _{i}\right )$. As pointed in Hanemann (1984), under this specification the purchase choice boils down to comparing the quality adjusted prices of the inside and outside options and picking the one that is lower. We assume that the outside good has a unity quality adjusted price so that p₀ = 1. Then the purchase probability reduces to

$$ \Pr\left( \bar{v}\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)\geq v_{0}\right) = \Pr\left( \frac{p}{\psi_{i}}\leq 1\right) = \Pr\left( \ln\psi_{i}-\ln p\geq 0\right). $$

Consistent with Hanemann (1984), we assume that the perceived value index (ψ_i) is defined as $\ln \psi _{i}=\lambda _{i}+\varepsilon _{i}$, where the λ_i is a function of observed covariates and ε_i is a Logistic random deviate with scale parameter μ. This results in the familiar logit probability of purchase in Eq. 3. Further, by Roy’s identity, the demand (conditional on buying) is given by

$$ \bar{q}_{i}\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)=-\frac{\partial\bar{v}\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)/\partial p}{\partial\bar{v}\left( p,y_{i},\psi_{i},\varepsilon_{i}\right)/\partial y_{i}}, $$

which gives the following expenditure function $R_{i} = p\bar {q}_{i}\left (p,y_{i},\psi _{i},\varepsilon _{i}\right )=\theta _{i}e^{\eta y_{i}}e^{\left (\rho -1\right )\left (\lambda _{i}-\ln p+\varepsilon _{i}\right )}$. As we observe expenditures only upon purchase, the expenditure function can be obtained by taking conditional expectation of R_i specified in Eq. 4.