Impact of AI-Assisted Financial Nudges on Micro-Enterprise Resilience: A Longitudinal Evidence from Aguascalientes, Mexico.

1. Control Group (C): AI Training Only Participants in this group receive a technical training program focused on "AI Survival Kits" for small businesses. This includes the use of generative AI tools for basic administrative tasks, customer service, and digital presence.

2. Treatment Group (T): AI Training + "Bridge Habit" Nudges In addition to the AI training received by the control group, this group receives a behavioral intervention via WhatsApp. This consists of:

Daily Digital Nudges: Automated messages sent at the end of the business day requesting partial cash-flow closings.

Bridge Habits: A structured routine designed to restore "Digital Monetary Salience" (DMS) by reducing the gap between the owner’s estimated balance and the actual registered cash balance.

Weekly Summaries: Feedback on income volatility and financial discipline to encourage precautionary savings.

The intervention is designed to test if these behavioral "Bridge Habits" reduce the cognitive load of financial management and improve resilience against exogenous shocks, such as the local economic surge caused by the San Marcos Fair

2026-02-01

2026-04-30

1. Financial Resilience (Cash-flow Volatility)Definition: The variation in daily cash flow, measured by the coefficient of variation (CV) of daily revenues reported in the business diary.
Measurement: Calculated as the ratio of the standard deviation to the mean of daily income over a weekly period4.Goal: To determine if the combined intervention (IA+Nudge) reduces weekly volatility compared to the AI-only group.

2. Precautionary SavingsDefinition: The net amount of money reserved in a dedicated emergency fund.
Measurement: Net monthly balance of the emergency fund as reported in the final evaluation.

3. Monetary Salience (Digital Monetary Salience - DMS)Definition: The percentage difference between the owner's estimated cash balance and the actual real balance recorded at the end of the week.
Measurement: {Estimated Balance - Real Balance} | [cite_start]/ {Real Balance}.

1. Revenue Volatility IndexThis outcome is constructed to measure financial instability. It is calculated as the Coefficient of Variation (CV) of daily revenues reported by the micro-enterprise owner over a seven-day period.
Formula: CV = sigma\mu, where sigma is the standard deviation of daily income and mu is the mean daily income for that week.}
Adjustments: To prevent bias from extreme values during the San Marcos Fair (April), the monetary variables used for this calculation will be winsorized at the 1% and 99% levels.

2. Digital Monetary Salience (DMS) Gap This variable measures the cognitive accuracy of the entrepreneur regarding their liquidity. It is a constructed percentage difference between perceived and actual financial states.
Construction:
DMS = {Estimated Balance - Real Balance} | {Real Balance}} * 100.

Data Sources: The "Estimated Balance" is the amount the owner predicts they have before checking records, while the "Real Balance" is the closing figure recorded at the end of the week.

3. Economic Resilience (Rebound) This is a temporal measure of recovery following an exogenous shock.
Construction: It is defined as the number of days elapsed for a micro-enterprise to return to its pre-shock average sales level after a self-reported shock or a known local event (e.g., the conclusion of the San Marcos Fair).

Cognitive Load (Financial Stress): Perceived financial stress level measured through a 5-point Likert Scale.

AI Adoption Rate: The frequency and intensity of using digital AI tools for business decision-making

Financial Discipline (Bridge Habits): The frequency of daily cash-flow closings and consistent record-keeping.

Perceived Administrative Structure: An index measuring how organized the entrepreneur feels regarding their business operations

Cognitive Load Index: This is a self-reported measure where entrepreneurs rate their level of mental effort and stress related to financial management. We will aggregate these responses to create a standardized index of perceived administrative burden.



AI Adoption Intensity: This outcome is constructed by tracking the number of business tasks (e.g., customer service, inventory, marketing) where AI tools are actively applied, as reported in the weekly follow-up surveys.


Bridge Habit Consistency: This is a behavioral metric calculated as the percentage of days per week the entrepreneur successfully completes the partial cash-flow closing prompted by the nudge.



Effect of Mediation: We will test if the "Frequency of Financial Records" acts as a mediator (Bridge Habit) between the intervention and the reduction in cash-flow volatility.

This research is designed as a Randomized Controlled Trial (RCT) with a longitudinal panel design spanning 4-5 months in Aguascalientes, Mexico. The study aims to measure the impact of combining Artificial Intelligence (AI) tools with behavioral "financial nudges" on the economic resilience of micro-enterprises (0-10 employees).

Sample and Stratification: The study targets an initial sample of 115 micro-enterprises. To ensure a balanced representation of the local economic landscape, the sample is stratified across three distinct geographic and commercial clusters:
Cluster A (Established): Formal businesses in the Historic Center.
Cluster B (Growth): Semi-fixed businesses in Growth Corridors.
Cluster C (Informal): Informal businesses in local street markets (Tianguis).

Randomization and Groups: Randomization is performed at the individual micro-enterprise level. Participants are assigned to one of two experimental arms:

Treatment Group (T): Receives technical training in Generative AI tools plus a behavioral intervention consisting of daily and weekly financial nudges delivered via WhatsApp.

Control Group (C): Receives the technical AI training only, without the behavioral reminders or nudges.

Data Collection Timeline: The study follows a 4-wave collection structure:
Wave 1 (Baseline): January 2026.
Wave 2 (Intervention): February 2026.
Wave 3 (Monitoring): March 2026.
Wave 4 (Final Evaluation): April 2026, capturing the exogenous shock of the San Marcos Fair.

Not available

Method: Stratified randomization performed in-office using a computer.


Randomization was conducted in-office by the research team using a computer-generated random assignment script. To ensure balanced groups across the diverse economic landscape of Aguascalientes, we employed a stratified randomization approach. Participants were first grouped into three strata based on their commercial location and formality:

Cluster A: Historic Center (Established/Formal)

Cluster B: Growth Corridors (Semi-fixed)

Cluster C: Tianguis (Informal)

Within each stratum, businesses were randomly assigned to either the Treatment (IA + Nudges) or Control (IA only) group in a 1:1 ratio. This method ensures that environmental and institutional factors specific to each cluster do not confound the treatment effect

The randomization unit is the individual micro-enterprise. Although the sample is stratified into three geographic clusters (Historic Center, Growth Corridors, and Street Markets) to ensure balance, the random assignment to treatment or control groups was performed at the firm level within each stratum. There is no multi-level or group-based randomization; each business owner is assigned independently to avoid spillover effects between different types of commercial activities.

No

0
The study uses an individual randomization design. While the sample is divided into 3 geographic strata (Historic Center, Growth Corridors, and Street Markets) to ensure balance, these are not clusters of treatment. Randomization occurs at the firm level.

Planned Number of Observations: 460 observations (115 micro-enterprises x 4 waves).

The total sample of 115 micro-enterprises is divided into two arms using a 1:1 randomization ratio within each geographic stratum.

Control Arm (C): 57 units receive the baseline technical training on Generative AI tools.

Treatment Arm (T1): 58 units receive the same AI training plus the behavioral intervention consisting of daily and weekly financial nudges via WhatsApp.

This initial allocation accounts for a projected 30% attrition rate over the 4-wave panel study, aiming for a final analytical sample of approximately 40 units per arm (80 total) to maintain statistical power.

Power Calculation: Minimum Detectable Effect (MDE)Minimum Detectable Effect (MDE): 0.40 Standard Deviations (Cohen's $d$).Statistical Power: 0.80.Significance Level ($\alpha$): 0.05.Explanation and Assumptions:The power calculation was performed for the primary outcome: Reduction in Cash-flow Volatility (Coefficient of Variation of Daily Revenue).Sample Design: The calculation assumes an individual randomization design with a total initial sample of 115 micro-enterprises (58 Treatment, 57 Control).Attrition Adjustment: We accounted for a conservative 30% attrition rate over the 4-month period. This results in a final analytical sample of 80 units (approx. 40 per arm). Effect Size in Percentage: Based on pilot data and similar behavioral interventions in micro-finance, an MDE of 0.40 SD corresponds to an approximately 12% to 15% reduction in the coefficient of variation of daily revenue.Clustering: Since the randomization is individual and not clustered, the intra-cluster correlation (ICC) is not applicable ($ICC = 0$), which maximizes the power for this specific sample size.Gain from Panel Data: By utilizing 4 waves of data per unit and a Fixed Effects (FE) model, we anticipate a reduction in the standard error of the treatment effect, effectively increasing our ability to detect the specified MDE even with the projected attrition.