Why Don’t Small Firms Adopt Analytics? Experimental Evidence from Rural U.S. Retailers

How to cite

Karim, Muhammad. 2026. "Why Don’t Small Firms Adopt Analytics? Experimental Evidence from Rural U.S. Retailers." AEA RCT Registry. AEARCTR-0019211. https://doi.org/10.1257/rct.19211-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Muhammad Karim*

* Primary investigator

Registry ID
AEARCTR-0019211
DOI
https://doi.org/10.1257/rct.19211-1.0
First registered
Last updated
Status
In development
Trial dates
2026-11-17 – 2028-01-01
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/19211

Abstract

A defining puzzle of advanced economies over the past two decades is the coexistence of cheap, widely available productivity-enhancing technology with slow, highly uneven adoption — a divergence between frontier firms and a long tail of laggards that weighs on aggregate productivity (Andrews, Criscuolo and Gal, 2016; Syverson, 2011). The laggards are disproportionately small: U.S. Census data show that advanced technology use is rare and skewed toward large, old firms (Zolas et al., 2020), and OECD evidence shows the small-firm adoption gap widening precisely where technologies become analytical rather than transactional (OECD, 2021). Yet nearly everything known about this gap in developed countries is observational. The diffusion literature documents who adopts; it cannot say which barriers bind, or what intervention would change adoption, because adopters differ from non-adopters in unobserved ways that also drive performance. The experimental literature that could answer such questions has been built almost entirely in developing countries, while the single prominent randomized study of analytics adoption in a rich country — eBay’s Seller Hub rollout (Bar-Gill, Brynjolfsson and Hak, 2024) — concerns online platform sellers for whom the tool arrives frictionlessly inside software they already use. No randomized evaluation exists for the modal laggard: the brick-and-mortar, owner-operated small firm. This proposal describes such a trial. Exploiting an established field platform of 200–250 independent retail locations across Tennessee, Arkansas, and Mississippi and a sampling frame of 300+ operators, the study (i) fields a diagnostic survey with enumerator-verified behavioral measures of data fluency; (ii) randomly assigns operators to a control group or to one of three arms that hold analytical content fixed while varying delivery format, workflow integration, and interpersonal support — the three mechanisms that theory and our own pilot evidence identify as candidate binding constraints; and (iii) measures effects on engagement, managerial behavior, and operational performance over twelve months using high-frequency point-of-sale administrative data. The design converts the developed-country adoption question from a correlational to a causal one, at the firm margin where the adoption gap is largest and least understood.

Experimental design

This study is a multi-site, four-arm randomized controlled trial involving approximately 300 independent retail owner-operators in Tennessee, Arkansas, and Mississippi. Following completion of the baseline survey, participants will be assigned in equal proportions to a control group or one of three treatment groups. All affiliated stores owned by the same operator will receive the same assignment.

The three treatment groups receive identical analytical content generated from their own point-of-sale data but through different delivery formats. Treatment 1 receives a simplified standalone weekly summary delivered in print and electronically. Treatment 2 receives the same content as brief, action-specific prompts timed to recurring decisions such as reordering, pricing, and staffing. Treatment 3 receives the workflow-embedded prompts plus standardized in-person onboarding and brief monthly check-ins. The control group will not receive the study analytics during the experimental period and will be offered the best-performing format after endline.

Randomization will be stratified by business type, revenue category, baseline data-use level, and single- versus multi-store ownership. The intervention will operate for twelve months. Outcomes will be measured using baseline, midline, and endline surveys; point-of-sale and inventory records; staffing records; and delivery-system engagement data. Midline and endline enumerators will be blinded to treatment assignment.

Design details

Randomization method
randomization done in office by a computer
Randomization unit
Firm
Clusters
300 firms
Observations
Approximately 300 owner-operators as the primary experimental observations. Repeated weekly administrative observations will also be collected for all participating stores during the pre-intervention and twelve-month intervention periods.
Arms
Approximately 75 owner-operators in the control group; 75 in Treatment 1 (standalone summary); 75 in Treatment 2 (workflow-embedded prompts); and 75 in Treatment 3 (workflow-embedded prompts plus standardized onboarding and monthly check-ins).
Primary outcomes
Sustained data use (primary index 1); Managerial behavior (primary index 2); Operational performance (primary index 3)

Cited by

No citation data available for this registration yet.

Keywords