Data-Driven Learning and Firm Performance: Evidence from Kenya
How to cite
Ashraf, Anik; Lyons, Elizabeth. 2025. "Data-Driven Learning and Firm Performance: Evidence from Kenya." AEA RCT Registry. AEARCTR-0016338. https://doi.org/10.1257/rct.16338-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Many entrepreneurs in developing countries under-invest in collecting and learning from information such as sales and inventory records. In this project, we propose a field experiment with small and medium enterprises (SMEs) in Kenya to understand if SME entrepreneurs in developing countries are limited in what they learn from their business records and, therefore, under-invest in keeping records. We focus on a key complementary input to the learning process – the ability to process raw information into relevant knowledge for improving business strategies. In particular, we introduce a large number of Kenyan SMEs to a mobile-phone-based Point of Sale app, where for a random subset of the SMEs, the app provides support with data analytics. We study subsequent app usage, business practices, and business performance.
Experimental design
This field experiment evaluates whether SMEs in Kenya underinvest in record-keeping because of challenges in turning raw data into actionable knowledge. We will test whether integrating analytics into a smartphone-based POS system (“mSpark”) encourages consistent record-keeping and improves business practices and outcomes. Participants comprises of retail-focused SMEs (e.g., small supermarkets, pharmacies, salons, and mom‑and‑pop stores) recruited via Technoserve’s network.
The key intervention is implemented in two stages. In the first stage, we promote mSpark app among 3,300 SMEs in our sample pool and encourage them to adopt the app. All participating businesses who decide to install the app receive the standard mSpark app, which enables sales and inventory tracking, low-stock alerts, and viewing transaction histories. In the second stage, one month after the initial onboarding, we roll out the main treatment to a random subset of the users of the app. To be specific, we introduce them to an advanced version of the app with analytical features.
At the stage of the randomization – which we intend to do in mid-July 2025 – we expect to have a sample of around 2,000 consistent users of the app. We will introduce the advanced mSpark app to random half of the consistent users we will have at this point. We will follow a clustered randomization approach, where we will use location of the businesses as randomization clusters.
Additionally, to be able to speak towards the effect of the basic mSpark app, we set aside a random subsample of our initial sample pool (a total of 300 SMEs) to serve as a pure control group. This randomization is done before the first stage of app promotion commences. These businesses do not receive information about the mSpark app but we conduct a complete baseline survey of their businesses.
We collect data through at least three rounds of surveys, including one listing, one baseline and at least one follow-up survey. If funding allows, we intend to conduct a final round of survey around 12 months after the treatment roll-out.
Design details
Cited by
No citation data available for this registration yet.