Informed Merchants: A Randomized Field Experiment on Peer Data of SMEs

The intervention provides small and micro stores with weekly peer-performance information through the Shouqianba app. Peers are defined as stores in the same industry category, and within a 3-kilometer radius of the focal store. Treated stores receive an in-app message notifying them that a new peer-information feature has been launched.The main experiment randomly assigns 100,000 vendors to euqal sized treatment and control groups: 50,000 vendors to the control group and 50,000 vendors to the treatment group. Treated vendors receive weekly peer-information (revenue and relative standing) displays through the app. Vendors in the control group receive no change to the app interface and do not receive access to the new peer-information feature. Outcomes are measured using administrative data from the platform at the store-week level.

Control group:
Stores in the control group receive no change to the app interface and do not receive access to the new peer-information feature.

Treatment: Peer revenue information, relative standing, and top-performer benchmarks
Stores receive weekly information on: their own revenue in the previous week; the median revenue of peer stores in the previous week; their percentile rank within the peer group; and the revenue of a peer store in the top 10 percent of the peer revenue distribution. Information is updated every Monday using the previous week’s performance.

2026-06-29

2026-08-31

Weekly store revenue, measured using Shouqianba administrative transaction data.

Weekly revenue is the main measure of store performance. It captures whether access to peer benchmark information improves merchants’ business outcomes relative to the control group.

Weekly average transaction value
Weekly transaction count
Weekly share of revenue from returning customers
Opening hours
Adoption of value-added services
Other important outcomes

These outcomes capture the potential margins through which merchants may respond to peer information.

The intervention provides small and micro stores with weekly peer-performance information through the Shouqianba app. Peers are defined as stores in the same industry category, and within a 3-kilometer radius of the focal store. Treated stores receive an in-app message notifying them that a new peer-information feature has been launched.The main experiment randomly assigns 100,000 vendors to euqal sized treatment and control groups: 50,000 vendors to the control group and 50,000 vendors to the treatment group. Treated vendors receive weekly peer-information (revenue and relative standing) displays through the app. Vendors in the control group receive no change to the app interface and do not receive access to the new peer-information feature.Outcomes are measured using administrative data from the platform at the store-week level.

Control group:
Stores in the control group receive no change to the app interface and do not receive access to the new peer-information feature.

Treatment: Peer revenue information, relative standing, and top-performer benchmarks
Stores receive weekly information on: their own revenue in the previous week; the median revenue of peer stores in the previous week; their percentile rank within the peer group; and the revenue of a peer store in the top 10 percent of the peer revenue distribution. Information is updated every Monday using the previous week’s performance.

Method for generating market information:
Directly including treated merchants in the peer benchmark would create a contamination problem. If treated merchants respond to the intervention—for example, by increasing their revenue during the eight-week experimental period—the peer median and top-10-percent benchmark would also shift in response to our own treatment. As a result, a treated store that genuinely improves its performance may see its displayed rank remain unchanged, or even decline, simply because the benchmark itself has moved upward due to treatment-induced changes among its peers. Conceptually, this creates interference across units and is related to a potential violation of the Stable Unit Treatment Value Assumption, since one firm’s treatment status can affect the information environment faced by another firm.
To address this issue while still preserving the representativeness of the full market benchmark, we replace each treated merchant in the peer pool with a pseudo merchant. The peer pool for each focal merchant therefore consists of actual untreated merchants and pseudo versions of treated merchants. Each pseudo merchant is constructed to approximate the treated merchant’s counterfactual revenue trajectory in the absence of treatment. Specifically, starting from the treated merchant’s baseline revenue, we update the pseudo merchant’s revenue in each subsequent week using the aggregate revenue growth rate of untreated merchants in the same peer group.
Under this approach, the benchmark received by a treated store provides an estimate of the local market distribution that would have prevailed absent contamination from the intervention itself.

Eligible stores in the main experiment will be randomly assigned at the store level to the control or treatment groups using computer-Eligible stores are randomly assigned at the vendor level to either the control group or the treatment group using computer-generated random assignment. Assignment is fixed for the duration of the experiment.
Randomization is stratified by industry, city, and market-intensity level. These variables are central to the research design because the effect of peer information may depend on the competitive environment in which the information is provided. Vendors in different industries face different demand conditions, pricing structures, and customer-retention strategies. Vendors in different cities operate in different local markets, with different consumer bases and business environments. Market intensity captures the degree of local competition and the density of comparable peers, which may shape how vendors interpret and respond to peer benchmarks.
Stratifying on these three dimensions helps ensure balance between the treatment and control groups across key determinants of vendors’ future decisions. It also allows us to examine heterogeneity in treatment effects across different competitive environments.

Vendor

No

Not applicable. Treatment is assigned at the vendor level and is not clustered.

100,000 vendors.

Control group: 50,000 vendors.
Treatment group: 50,000 vendors.