Experimental Design
The empirical design leverages the staggered rollout of the sourcing tool across Northern Lima, a low-income sector of the city home to about 25,000 corner stores and one-fourth of the city’s population.
We divided the area into 119 similarly sized neighborhoods with about 200 corner stores, which constitute the study strata. Each neighborhood was then divided into three smaller clusters, excluding the areas around markets, which were left out of the study (businesses in and around markets are very different and the implementing firm preferred not to include them in the randomization). Two clusters per neighborhood were randomly assigned to the treatment group to receive weekly visits by sales agents. Stores in the remaining clusters form the control group and do not receive scheduled visits until after data collection (but they can download the app and make orders independently, for example if they hear about the service from friends). Clusters were drawn to ensure that treatment areas are contiguous within each stratum regardless of the randomization result, to reduce compliance issues.
Data collection:
Outcomes will be measured through a combination of the firm's administrative data (on client transactions) and a representative survey of corner stores in the study area. The survey began in March 2025, eight months after the initial rollout, and is scheduled to end in August. Neighborhoods are randomly assigned to monthly surveying rounds.
The survey covers a random sample of 10 corner stores per stratum (3 per treatment cluster and 4 in each control cluster). Sampling was based on a 2019 corner store census provided by the partner firm. Locations were randomly drawn from among the stores in the census, and surveyors were instructed to visit each point and locate the nearest open store. Replacements—such as in cases of refusal to participate—were allowed within three blocks of the original location.
Before enrollment, stores were screened based on the number of storage shelves: surveyors had to verify that each store had at least two before beginning the survey. The reason is that some corner stores are very small and open only occasionally, and thus are not covered in the firm's marketing strategy. Still, this criterion excluded fewer than one-third of the stores in the census.
The survey consists of two visits, two weeks apart. In the first visit, the team collects information on the store owner demographics, the store's history, its operations (employees, schedule, etc.), its financial performance over the last week (sales, profits, etc.), and retail prices for the most popular product in each category. Surveyors also document recent wholesale purchases by photographing all receipts provided by suppliers during the week prior to the survey. In the second visit, store owners are asked to provide updates on their financial performance and supplier-provided receipts for the two weeks between the visits. A separate team then encodes the data from the pictures, including supplier identity, order dates and total amounts, and product-level quantities and prices. The result is a granular dataset of all purchases made by each store over a three-week window.
Stores receive a small enrollment incentive (USD 1.50) and a USD 5 gift upon completing both survey visits. Additionally, stores are paid a small incentive for each receipt they show. This incentive was randomized between USD 0.15 and USD 0.30, as a test of whether stores withhold some receipts from surveyors.