How to Find Your Most Valuable Outlets? Measuring Influence in Service and Retail Networks

Abstract

Problem Definition: Consider a network of stores operating under the same brand, for example, a
chain of coffee shops or banks. Increasing sales at one store may have a different impact on the sales of another store: from a negative impact as a result of potential cannibalization to a positive impact from increased brand awareness and customer engagement. We study how to causally identify the spatially heterogeneous network effects between store pairs, which can be used to measure store influence.

Academic/Practical Relevance: Our work provides a scalable methodology that can be used to causally identify network effects on a large scale within service/retail networks. We demonstrate that the economic impact of network effects is substantial.

Methodology: We develop an extension of a spatial econometrics model that allows for
identification of both positive and negative peer effects between stores. This semi-parametric
methodology allows us to handle a large-scale network and provides causal estimates for the network effects between all store pairs. We use three distinct sets of instrumental variables that are
based on location-specific (weather and nearby sports events) and store-specific (deviations in
daily staffing) attributes.

Results: By applying our method to a large dataset from a major national restaurant chain in the
US, we show that the total influence of each store on the network sales is consequential: a $1
sales increase in one store can generate a total of up to $3.87 in additional sales across all
other peer stores in the network, while it can also reduce the sales in the network by a total of
$3.12, with the average influence of $0.55.

Managerial Implications: Our influence estimates provide valuable insights for the design and
management of networks; for example, for prioritizing stores with the highest influences for
ownership or improvement to optimize return on investment.