An optimization model cuts water pump downtime by up to 62% across rural Africa

When does it make sense to fix a water pump before it breaks? And when should you wait for the call that it has already failed? For NGOs maintaining thousands of handpumps across rural sub-Saharan Africa, getting this balance wrong means communities lose access to their only source of clean water. Getting it right means the difference between a pump that works and a walk to an unprotected water source.

More than 184 million people in rural sub-Saharan Africa depend on shared handpumps. At any given time, more than 50,000 of those pumps are broken. The organizations responsible for keeping them running operate with thin budgets, limited staff, and incomplete information about which pumps are failing and when.

Three countries, three maintenance strategies

Chengcheng Zhai, an assistant professor at Notre Dame's Mendoza College of Business, and her colleagues studied how NGOs in Ethiopia, Malawi, and the Central African Republic approach this problem. Each country uses a fundamentally different model.

In the Central African Republic, the NGO visits each pump once per year on a pre-scheduled rotation. In Ethiopia, an incoming call center waits for communities to report breakdowns. In Malawi, an outgoing call center proactively contacts communities to ask about pump condition, then dispatches mechanics to fix problems.

Each approach has obvious trade-offs. Scheduled visits may arrive when nothing is wrong, wasting a trip. Waiting for breakdown reports means pumps sit broken until someone calls. Outgoing calls gather better information but cost more to operate.

A model that adapts to what each NGO already does

Rather than proposing a single best approach, Zhai's team developed a Markov Decision Process, a dynamic optimization model that adapts to each NGO's existing maintenance strategy. The model observes how many pumps are reported broken in each geographic cluster and how long it has been since the last mechanic visit, then recommends which cluster to visit next.

The key insight is that during each visit, mechanics perform both preventive maintenance and any needed repairs. This bundling reduces the number of total trips while catching problems before they cause extended downtime.

The results, forthcoming in the journal Manufacturing and Service Operations Management, show substantial improvements across all three countries. Enhanced scheduling reduced maintenance downtime by 47% to 62% in Ethiopia, up to 53% in Malawi, and 42% to 55% in the Central African Republic. Cost impacts varied: Malawi and Ethiopia saw logistics savings, while the Central African Republic experienced a 15% to 19% cost increase, reflecting the fact that more frequent visits do sometimes cost more.

The counterintuitive finding on costs

NGOs often assume that adding preventive maintenance to their operations will raise expenses. The Notre Dame analysis found this is not always true. When preventive maintenance is scheduled optimally, the reduction in emergency trips and repeat visits can generate net savings. Even where costs do rise, the major improvements in pump uptime may justify the added expense for organizations whose core mission is reliable water access.

The research has already influenced operations. The ongoing collaboration has helped provide more consistent and safe water access for more than a million people across the three countries studied. MBA students at Notre Dame's Meyer Business on the Frontlines program are building on the research to advise NGOs and social enterprises on water management strategy.

Limitations and what they mean

The study depends on data from three specific NGOs operating in three countries with different infrastructure, geography, and institutional capacity. Whether the model's recommendations transfer directly to other contexts, such as different pump types, different climates, or regions with even less communication infrastructure, has not been tested. The model also assumes that NGOs can act on its recommendations, which may not hold when security concerns, seasonal road conditions, or sudden budget cuts intervene.

The data on pump functionality is itself incomplete, a constraint the model acknowledges. In settings where communities underreport problems or where call centers have limited reach, the model's inputs are noisier and its recommendations less precise.

Still, the core finding is practical and clear: well-timed preventive maintenance, even under imperfect information, substantially reduces the time communities spend without clean water. For organizations working with tight budgets and incomplete data, that is a result worth acting on.