Does Information Help Agents Perform Better?: A Mobile Money

Does Information Help Agents Perform Better?: A
Mobile Money Field Experiment in Tanzania
February 24, 2017
1.
Introduction
Mobile money refers to a technology platform that allows users keep track of – and virtually move
– their deposited money by using their non-smart ‘feature phones.’ Here is an example of how
it works, which is also the most popular application and the use-case that helped mobile money
spread so fast. Imagine a worker in Kenya who lives in an urban area and whose family lives far
away in a rural area. The worker earns cash in the city that he wants to send back to his family.
Traditional banking infrastructure in Kenya is not developed enough to allow most people to easily
transfer money without using physical cash. Thus, before the advent of mobile money, he would
have to physically carry the money to his family during a visit or he would have to trust someone
else to carry the physical cash to his family. With mobile money, however, he can take the cash to
a local ‘agent.’ When he hands the money to the agent, his account is credited. The worker can
then transfer this virtual money to his family far away. His wife can then go to a local agent near
her and make a withdrawal (less a transaction fee) to get the physical cash. Besides transferring
money, mobile money can be used to save, to take out loans, to pay bills, and more.
The impact that mobile money can have on people in developing countries can be significant.
Suri and Jack (2016) show that just having the mobile money platform in Kenya lifted 194,000
households – 2% of all Kenyan households – out of poverty.
One thing that is impeding the mobile money system currently is stockout rates at agents, which
are estimated to be 8% (Balasubramanian et al. 2015). An agent is a person who is authorized by a
wireless provider (wireless providers are the backbone of the system: they enable the technology and
hold the money reserves that back up the virtual cash, or e-float) to make deposits and withdrawals
on behalf of customers. Often, agents can be found in markets, on street corners, in stores, in
booths at bus stops, and so on. They are the interface between customers who want to deposit
and withdraw cash and the electronic mobile money system. The agent is not an employee of the
wireless provider. Instead, for every transaction with a customer, the agent earns a commission
paid by the wireless provider. The agent has a budget (which she decides) which is split between
physical cash and virtual cash (e-float). Every time a customer deposits hard cash, the agent’s cash
reserves go up and her e-float reserves go down, while the sum of her cash plus e-float reserves
remains constant. Every time a customer makes a cash withdrawal from an agent, the agent’s cash
1
Authors’ names blinded for peer review
2
reserves decrease and her e-float reserves increase, while the sum of her cash plus e-float remains
constant. Thus, it is up to the agent to make two decisions: 1) how much capital should the agent
invest into this business and 2) how should that capital be divided at the beginning of the day
between physical and virtual cash. (At the beginning of the day, the agent can ‘rebalance’ her
budget at a bank to attain a good allocation of cash and e-float.) Balasubramanian et al. (2015)
propose heuristics which address these two questions which aim to maximize the profits of the
agent, given that cost of capital is the main cost and commissions from the wireless provider are
the sources of revenues.
A stockout occurs when a customer wants to make a deposit and the agent does not have enough
e-float or the customer wants to make a withdrawal and the agent does not have enough physical
cash. These are lose-lose-lose-lost situations. First, the customer cannot perform the transaction
he wanted to make. Second, the agent loses out on commission. Third, the wireless provider loses
out on transaction fees and disappoints a customer. Fourth, faith in the system as a whole erodes,
which makes is more difficult to convince users that mobile money is a viable alternative to a pure
physical cash (or traditional banking).
2.
Experiment details
The agent’s problem is a operations one: what are the best inventory levels to have? Essentially,
we want to understand how agent behavior affects their operational decision making, and what
contributes to optimal and sub-optimal choices. While many laboratory experiments have been
performed regarding behavioral operations (Schweitzer and Cachon 2000), we believe that there
is an opportunity to learn about behavior outside of the laboratory, where agents have their own
money on the line. Thus, in this field experiment, we aim to answer the following questions:
1. Do agents adhere to explicit recommendations as to cash/e-float levels?
2. Do recommendations help agents perform better (fewer stockouts, higher profit)?
3. Do some agents who have information not available to us (for whom the recommendations
may be inaccurate) perform better with information instead of recommendations?
4. Does providing information along with recommendations increase trust and adherence?
5. Does in-person training help agents’ adherence to recommendations or their performance?
We partner with a wireless provider in Tanzania to perform a field experiment which tests
the impact of SMS message recommendations and in-person training on agent performance. Our
treatable population (almost 5000 agents) is comprised of agents in Dar es Salaam who fulfill
certain criteria, such as a minimum time with the wireless provider and minimum activity levels.
Of this population, we treat 1200, with the treatments being the following:
1. Recommendation: we provide an SMS message on a daily basis with a recommendation as to
how much cash and e-float to carry.
First 30 days
Authors’ names blinded for peer review
Manufacturing & Service Operations Management; manuscript no.
3
Treatment 1:
Recommendation
Treatments 1 & 2:
Recommendation
and information
250
250
250
Treatment 3:
Warm body training
offered
150
150
150
Treatment 4:
Voice notification
Control
(~3600)
Treatment 2:
Information
Second 30 days
Treatments 1 & 2: Recommendation and information
Figure 1
Control
(~3600)
750
Treatment 3:
Warm body training
offered
450
Treatment 4:
Voice notification
The breakdown of number of agents in each treatment. After 30 days, we give all agents recommendation and information.
2. Information: we provide an SMS message on a daily basis with information as to the agent’s
median and 90th percentile deposit and withdrawal volumes.
3. Training: agents are trained at several locations in Dar es Salaam by native Swahili speakers
as to the motivation of the project and ideas behind the heuristics.
4. Notification: agents are invited to call a pre-recorded message which explains that the program
can help them, but does not go into any kind of depth as is found in the in-person training.
Given these treatments, our experimental design can be summarized in figure 1.
3.
Analysis
We will analyze the data from this experiment and present the results at the MSOM conference.
In addition to the transactional data of agents, we also have 400 questionnaires filled out by agents
who attended training. We are currently transcribing these questionnaires into a database. Once
this data can be merged, we will be able to answer the questions we posed above.
References
Balasubramanian, Karthik, David Drake, Douglas Fearing. 2015. Inventory Policies for Mobile Money:
Improving Agent Profitability in East Africa. Working paper, Harvard Business School, Cambridge,
MA.
Schweitzer, Maurice E., Grard P. Cachon. 2000. Decision Bias in the Newsvendor Problem with a Known
Demand Distribution: Experimental Evidence. Management Science 46(3) 404–420.
Suri, Tavneet, William Jack. 2016. The long-run poverty and gender impacts of mobile money. Science
354(6317) 1288–1292.