Mobile Money Agent Cash and E-Float Forecasting to Prevent Turned-Away Customers
Mobile money agents in Kenya, Pakistan, Bangladesh and Nigeria lose customers when they run out of cash or electronic float, especially on salary days and before holidays. Build forecasting models for each agent's cash and e-float needs from simulated transaction histories with realistic calendars, and a simple app that tells the agent and the aggregator when and how much to rebalance. Measure how many turned-away transactions the forecasts would prevent.
Intermediate
A small team, or one strong student willing to learn something new.1 semester · 2 students
7 to ship
4 optional extrasSuggested stack
What you should ship
- A transaction simulator for 200 or more agents with cash-in, cash-out, salary days, remittance days, holidays and local events
- Daily cash and e-float balance tracking per agent, including when an agent could not serve a customer
- Forecasting models for next-day and next-week net cash flow per agent, with baselines
- Rebalancing recommendations: how much cash or e-float to move and when
- An agent view and an aggregator view of agents at risk
- A simulation comparing turned-away customers with and without forecasts
- A documented list of simulator assumptions
If you have time left
- Routing for cash runners visiting several agents
- Clustering agents by behaviour to improve forecasts for new agents
- SMS summaries for agents with basic phones
- Testing on a public mobile money dataset if one fits
The problem: no cash, no customer
Mobile money relies on agents: small shops where people turn cash into digital money and back. An agent needs both physical cash and electronic float. On salary days everyone wants to cash out; before Eid or Christmas, demand changes again. When an agent runs out, the customer walks away, often to an informal option, and trust in the service drops. Agents and aggregators manage this by experience and phone calls.
What already exists and the gap
- M-Pesa in Kenya, Easypaisa and JazzCash in Pakistan, bKash in Bangladesh and OPay and others in Nigeria run large agent networks.
- Aggregators and super-agents move cash and float between agents, mostly on fixed routines or calls.
- CGAP, the financial inclusion organisation hosted by the World Bank, has published research on agent liquidity management as a key challenge.
The gap is forecasting at the level of a single agent, tied to concrete rebalancing advice and tested by how many customers it would have served.
Inspiration
CGAP's work on agent networks identifies liquidity as one of the main reasons agents fail and customers are turned away. Forecasting cash demand is a well-studied problem for bank ATMs; your project applies those methods to small agents with patchy histories and strong calendar effects.
What you build
- Agents get a morning message: expected cash and float needs today, and whether to request a top-up.
- Aggregators see which agents are likely to run short and plan visits.
- Analysts get a reusable simulator and benchmark.
The technically interesting part
- Calendar effects. Salary days, government payments, Ramadan and Eid, school fees and harvests drive demand. Encode them as features, including moving holidays.
- Censored demand. When an agent runs out, you never see the transactions that did not happen. Model turned-away customers so you do not learn that demand is low when it was just unmet.
- Many short series. Hundreds of agents with short histories. Compare per-agent models with one global model using agent features.
- From forecast to action. Rebalancing has a cost. Recommend top-ups that balance the risk of running out against the cost of a trip.
How to evaluate it
- Forecast error against naive baselines, overall and on peak days.
- Turned-away customers in the simulator with and without recommendations.
- Rebalancing trips needed per week.
- Robustness when calendar patterns shift.
Suggested plan
- Weeks 1 to 3: research agent operations and design the simulator.
- Weeks 4 to 6: simulator and balance tracking.
- Weeks 7 to 10: forecasting models.
- Weeks 11 to 13: recommendations, views and impact simulation.
- Weeks 14 to 15: report and demo.
Path to a product
Mobile money operators, banks with agent networks and aggregators all lose money when agents run dry. A forecasting service integrated with their systems is a clear business case. The project is strong evidence for data science and fintech analytics roles.
Scope warning
Use simulated data unless a company gives you anonymised data under an agreement. Keep the scope to forecasting and recommendations, not moving real money. Document the simulator's assumptions so readers know its limits.
Learn what you need on Sefism
- Data science roadmap: forecasting and simulation.
- Machine learning roadmap: gradient boosted models across many series.
- Data analytics roadmap: dashboards for agents and aggregators.
- statsmodels documentation: time series baselines.
- Streamlit documentation: the views.
Get help with this idea
Ignoring censored demand will make every model look better than it is. Tauseef can help you design the simulator so the evaluation is honest. If you are new here, create a free account first.
Ideas and guidance, not finished projects
These are project ideas and scoping guidance, published free for students to use as a starting point. I do not build, write, or sell final-year projects, and I do not complete coursework for anyone. Take an idea, make it yours, and build it.
Not sure this one fits you?
I help students pick an idea that matches their skills and their deadline, then scope it down until it is finishable. Guidance only, never done for you.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.