Encouraging Adoption of Digital Merchant Payments and Implications for Tax Administration: Causal Evidence from Rwanda

MoMoPay. Our intervention aims at incentivising adoption of digital merchant payments among Rwandan businesses. The product we aim to offer is MTN’s MoMo-Pay, the merchant-specific digital payment service. While adoption of mobile money through personal accounts reached almost saturation in Rwanda, that is not the case of MoMo-Pay. As also documented in WB (2024) a puzzling gap remains when comparing business’ adoption of mobile money personal accounts and merchant accounts. MoMo-Pay, mirroring Safaricom service Lipa Na M-Pesa in Kenya, offers many more benefits to merchants than simply using personal accounts. From a retail sales perspective, MoMo-Pay allows safe and quick reconciliation of payments for merchants, and are more attractive to a wider range of clients – as merchants are now offering an alternative mode of payments at no cost, which is arguably a good marketing choice - and offer more enhanced competitiveness in the sector, as well as additional benefits in terms of record-keeping and accounting, as transactions statements are retrievable from MTN. Also, the restrictions on the amount of money one can store on personal accounts are inexistent for MoMo-Pay. When it comes to fees, MoMo-Pay does not levy any fees on the remitter (customer), but instead on the merchant that is making the sale, and thus receiving the payment. This is in contrast to payments to personal accounts that levy fees on the remitter (customer) instead. As of January 2025, transaction costs levied on merchants are 0.5% for all transactions above 4,001 RWF, while transactions below this threshold are free. Importantly, transactions across merchant payments accounts, that is B2B through MoMoPay, as well as payments from MoMoPay to bank accounts, are free, in contrast to the standard fees applied to transactions from personal accounts.

In our 2023 survey, only 45% of respondents made any use of MoMo-Pay. Among the main barriers to DMP usage are charges and fees, difficulty in using, limited knowledge and mistrust (Figure A3), while the key drivers refer mostly to reduction in transaction time, cost and risk (Figure A4). Our encouragement design intends to alleviate such barriers, which largely refer to misperceptions around transaction costs, information costs and mistrust for the digital solution.

In order for merchants to register with MoMo-Pay, they need to provide the following documents:

1) RDB/RGB Certificate, where their TIN is also indicated (RDB covers incorporated business entities, RGB unincorporated ones)
2) A letter requesting the service, signed and stamped by a director or representative of the business entity
3) The ID of the company representative if not a director

Encouragement design. A key challenge of measuring impacts of DMP is that one cannot randomise its adoption, deciding which merchants will adopt and which will be impeded to do so. An immediate solution to this challenge, as offered by the applied economics literature, consists in running a randomised encouragement design (RED), as one type of a broader set of Randomized Controlled Trials (RCTs). Field experiments, or Randomized Controlled Trials (RCTs), are widely used to estimate the causal effects of programs or technologies on economic outcomes (Duflo et al., 2007; Khandker et al., 2010). This method involves randomly assigning units, such as SMEs, into treatment and control groups. With sufficiently large samples, randomization ensures the groups are similar in observable (e.g., sector, size, location) and unobservable traits (e.g., technological proficiency). As a result, differences in average outcomes between the groups reflect the treatment’s impact, with the control group serving as a counterfactual (Rubin, 1974).

REDs are a variation of RCTs used to evaluate programs or technologies already available but not universally adopted in a study area (Bradlow, 1998). An early example is Holland (1988), who examined how extra preparation for the GRE test influenced scores. While all participants were free to prepare, a random subset was encouraged to put in additional effort. Because the groups were initially similar, differences in scores could only result from the increased effort. Duflo et al. (2006) provide another example, investigating the adoption of agricultural technology and the role of social dynamics. Fertilizer demonstrations were held on randomly selected farms, open to all farmers. For some plots, the farmer’s friends were explicitly invited to attend. Differences in adoption rates were thus attributed to varying social interactions. Finally, Devoto et al. (2012) implemented an RED to assess the welfare effects of piped water access in Morocco. Although all households could purchase a connection, only a subset was encouraged through information and application assistance. Differences in welfare outcomes between the groups reflected connection uptake by the encouraged households.

2025-10-30

2026-03-31

We plan to measure impacts across different sets of outcomes:

Primary outcomes:

1. Digital payment outcomes - first stage
a. whether the business opened a MoMoPay account
i. self-reported measure
ii. objective measure of MoMoPay code displayed in the business premises
b. whether the business used MoMoPay for business purposes, i.e. received any payment from customers, in the past 30 days
c. amount of sales paid through MoMoPay
d. share of sales paid across different payment methods (as in Bernad et al 2023)
2. Tax perception outcomes
a. ease of compliance with tax obligations
b. ease of keeping books of accounts
c. time spent monthly on tax-related activities (hours)
d. perceived transparency of tax systems (liability amounts, deadlines, etc)
e. perceived fairness of the tax systems
f. perceived willingness to share data with Government and financial sector
g. perceived probability of a tax audit
h. level of justification of tax evasion
3. Tax behaviour outcomes
a. whether the business filed and paid taxes
b. amount of tax liability declared
c. amount of tax liability paid
d. timeliness of filing and payment

Secondary outcomes:
1. Digital payment outcomes - usage network effects
a. whether a C group business opened a MoMoPay account/used MoMoPay based on number of T group businesses within a 500-meter radius (test for geographical proximity)
b. whether the business used MoMoPay for business purposes, i.e. made any payment to their suppliers with MoMoPay, in the past 30 days
c. number of suppliers the business transacted with MoMoPay
d. whether the business used MoMoPay for business purposes, i.e. paid their employees with MoMoPay, in the past 30 days
e. whether the business tried to make another trading partner enroll on MoMoPay
f. whether the business tried to make a customer use MoMoPay
g. whether the business also adopted R-switch for payment integration
h. whether the business accessed financial statement data from MoMoPay
i. whether the business keeps business sales and payments records via MoMoPay
2. Digital payment outcomes - perceptions
a. whether the respondent believes that MoMoPay is useful for business
b. whether respondent believes that MoMoPay works well
c. whether the respondent had technical issues using MoMoPay
d. whether the business improved their knowledge on MoMoPay and mobile money
e. perceived safety in doing business
3. Business growth
a. hours per week of operation
b. days per month of operation
c. asset value
d. monthly expenses value
e. monthly sales value
f. monthly profit
g. inventory value and purchases in the last 30 days
h. access to financial services index
i. whether the business has a bank account
ii. whether the business has a formal bank loan
iii. amount of formal bank loans
i. number of full-time employees
j. number of part-time employees
k. number of customers in a day
l. number of new customers in last 30 days
4. Tax behaviour outcomes
a. accuracy index
i. discrepancy between amount filed and paid
ii. discrepancy between turnover in income tax and VAT
b. amount of turnover declared
c. amount of expenses declared
d. amount of deductions and exemptions declared
e. amount of profits declared
f. number of employees
g. amount of social contributions
5. Tax behaviour outcomes - network effects from EBM data
a. number of trading partners
b. trading partners’ tax liability declared
c. trading partners’ tax paid

Our intervention. In line with the literature, and especially referring to Dalton et al (2023), the first RCT of a payment technology, the encouragement intervention will be implemented immediately following the baseline survey (see section c). It will aim to address three potential barriers to the adoption of MoMo-Pay. First, we will seek to reduce the information barrier by providing detailed insights into the advantages and disadvantages of MoMo-Pay compared to other payment methods, such as cash or personal accounts. While most merchants are aware of MoMo-Pay's existence, they will likely lack comprehensive and accurate knowledge about its features, including costs, usage, and potential benefits (Bernad et al 2023). This component of the intervention will aim to supplement—and, where necessary, correct—the merchants' existing understanding of the technology.

Information will be shared through a leaflet and a video. The leaflet will highlight key benefits of using MoMo-Pay, emphasizing advantages for customers, such as the absence of transaction fees, and for merchants. Merchant-specific benefits will include (i) increased customer appeal, (ii) reduced risk of theft due to less cash on hand (with no limit on funds stored in a MoMo-Pay account), and (iii) simplified bookkeeping through automatic transaction recording, with instructions on how to retrieve transaction statements. The video will complement the leaflet by featuring an interview with a business owner from the same sector who has successfully adopted MoMo-Pay. This approach will leverage evidence from the literature suggesting that successful peers serve as effective role models, particularly in low-income settings in developing countries (Bernard et al., 2014; La Ferrara, 2016: Dalton et al 2024).

Second, to reduce the transaction costs associated with opening a MoMo-Pay account, we will offer comprehensive support to handle the application process. Although opening an account will be free, even small administrative barriers can deter individuals from pursuing efficient investment opportunities (Bertrand et al., 2004). To mitigate this, a trained enumerator will assist merchants by (i) collecting copies of the required documents, (ii) completing the necessary paperwork with MTN, and (iii) retrieving and delivering the account materials to the merchant once the application is approved.

Finally, we will address the technological know-how barrier by helping merchants set up and familiarize themselves with the MoMo-Pay system. When delivering the account materials, the enumerator will assist the merchant in installing the account, verifying its functionality, and learning to use it. This process will include guiding the merchant through a test transaction, in which they will charge our standard MTN account 100 RWF and complete the transaction.

After presenting this encouragement package to merchants in the treatment group, the enumerator will ask whether they would like a MoMo-Pay account to be opened on their behalf. The response will serve as a measure of the merchant’s willingness to adopt MoMo-Pay. For merchants expressing interest, the enumerator will proceed with the application process through MTN.

As a last note, our intervention will set up such that it not only encouraged some firms to adopt LPN but also, did not discourage adoption by those treatment firms that would have decided to adopt LPN had they been assigned to the control group (Bradlow 1998). Our encouragement design is unlikely to result in a violation of this so-called monotonicity assumption.

Not available

Through survey software SurveyCTO.

Individual merchant

No

0

1000 merchants

500 merchants in control vs 500 merchants in treatment group

We assume a comparison of equally sized groups for our encouragement treatment. While we allow for potential drop out between the start of the survey and the intervention, since this will occur before treatment assignment we will exclude these respondents from the study sample. We will conduct surveys until 1,000 respondents have completed the baseline survey. We assume a standard significance level (alpha) of 0.05 and a standard power of 0.8 and calculate all results using two-sided tests. We look at several variables representative of our primary outcomes: ease of compliance with tax obligations, time spent on tax-related activities, perceived probability of a tax audit, turnover and final VAT liability declared in VAT returns. The data for the mean and standard deviation of these outcomes comes from the sample of small enterprise owners in Bernad et al (2023), and was collected in 2022. These were the closest outcomes to our primary and secondary outcomes that were available in this dataset. The first row for each outcome focuses on our current approach and is what we would expect to detect under a standard intent-to-treat approach. We expect to be able to detect roughly .17 standard deviations. In other words, in an intent-to-treat approach, we are well powered to detect even small changes in outcomes. We also explore what our ability to detect effects would look like under much more conservative scenarios for compliance with the encouragement intervention. We calculate power calculations varying the take-up rate, from 100% (the same as the first scenario) to a minimum of 30%. While appendix figures show the increase in MDE for the interval between 100% and 30%, in the table we report our most conservative scenario for simplicity. Under our most conservative scenario of 30% take-up, we remain able to detect roughly .21 standard deviations, which remain credible and realistic.