Predicting Revenue from Drinking Water Supply Infrastructure in Rural Africa
Briefing Note | March 2025
Duncan McNicholl, Rob Hope and Emil Aydinsoy
Sustainable financing for rural drinking water services depends on realistic assumptions about how much revenue water infrastructure can generate. Yet infrastructure investments are frequently based on optimistic expectations of water consumption and user payments, creating financial risks for long-term operation and maintenance.
Published in March 2025, Predicting revenue from drinking water supply infrastructure in rural Africa introduces Uptime Global's Revenue Predictive Model (RPM), a machine learning tool designed to estimate annual user payments from rural handpumps and piped water schemes.
Drawing on operational data from 11 African countries, the model helps governments, development banks and water service providers assess financial viability, identify potential subsidy requirements and improve investment decisions.
Building a Revenue Predictive Model
Since 2020, Uptime has collected detailed performance data from professional water service providers through results-based contracts.
The Revenue Predictive Model draws on a substantial dataset collected between 2021 and 2023, combining observed service performance with publicly available demographic, geographic and climate information.
<escape> </escape>
The model's underlying dataset includes:
More than 200,000 months of waterpoint records.
13,777 unique handpumps and piped water schemes.
11 African countries.
An estimated service population of 2.2 million people.
These records provide an empirical foundation for estimating the revenue that rural water infrastructure can realistically generate.
How the model works
The Revenue Predictive Model uses machine learning to estimate annual revenue for individual waterpoints and piped schemes.
It combines two categories of information:
Infrastructure characteristics: Waterpoint type, location, estimated population served, number of connections and water volume measurement systems.
Contextual factors: Local population density, nearby water infrastructure and rainfall patterns.
The model first classifies infrastructure into three annual revenue ranges: low (USD 0 to 5), medium (USD 5 to 135) and high (above USD 135). Additional machine learning models then generate more precise estimates within these ranges.
This two-stage approach provides both transparent revenue classifications and more detailed financial predictions.
Key Findings
1. Revenue predictions achieve approximately 80% accuracy
Model testing found approximately 80% accuracy when classifying annual waterpoint revenues into the three defined categories.
When predictions were incorrect, they were generally assigned to an adjacent revenue category. High-revenue waterpoints were incorrectly classified as low-revenue in only 3.1% of cases, while low-revenue waterpoints were classified as high-revenue in 2.1% of cases.
These findings demonstrate the potential for the model to challenge unrealistic financial assumptions and inform investment decisions.
2. Local data substantially improves predictions
The model performs best in contexts similar to those represented in its training dataset.
For new locations, incorporating historical revenue records can significantly improve accuracy. The paper finds that 12 consecutive months of data from approximately 150 waterpoints can improve predictions to around 80% accuracy.
This highlights the importance of local calibration before using predictions to inform major financing decisions.
3. Revenue forecasting can identify hidden subsidy requirements
The paper presents a hypothetical investment in 50 piped water schemes serving 20,000 people.
Initial financial assumptions project annual revenue of USD 146,000, substantially exceeding estimated operating costs of USD 65,000.
However, after calibrating the Revenue Predictive Model using historical data from nearby schemes, the maximum predicted annual revenue falls to USD 55,000.
This would imply an annual operating deficit of USD 10,000, demonstrating how more realistic revenue assumptions can identify potential financing gaps before investments are made.
Applications for Governments and Development Banks
The Revenue Predictive Model offers six potential applications:
Validating revenue forecasts for planned drinking water infrastructure investments.
Assessing the financial viability and subsidy requirements of existing water infrastructure.
Comparing alternative infrastructure rehabilitation and investment strategies.
Exploring how demographic and climate changes may affect future revenue.
Informing the design of results-based contracts and Payment for Results programmes.
Benchmarking the financial performance of rural water services.
These applications can help governments and development banks design more realistic financing arrangements, particularly where long-term subsidies may be required to sustain professional service delivery.
Model Limitations and Future Development
The paper recognises that revenue predictions remain subject to uncertainty.
The available dataset is concentrated among particular service providers, infrastructure types and geographic contexts. Predictions may therefore be less accurate when applied to locations that differ substantially from those represented in the training data.
Other factors, including alternative water sources, seasonal changes in demand, service reliability and water quality, can also affect actual user payments.
Uptime recommends calibrating the model with local historical data wherever possible and interpreting predicted revenue ranges alongside the uncertainty associated with individual estimates.
Expanding the dataset through collaboration with governments and development banks offers opportunities to improve the model's accuracy and applicability across sub-Saharan Africa.
Conclusion
Uptime's Revenue Predictive Model provides a data-driven approach to estimating user payments and assessing the financial sustainability of rural drinking water infrastructure.
By combining verified operational data with machine learning and publicly available contextual information, the model can help governments and development banks challenge optimistic revenue assumptions, identify potential subsidy requirements and design more sustainable financing strategies.
The research demonstrates how data collected through results-based contracts can generate wider value by informing infrastructure investment and national water service financing.
Uptime invites governments and development banks to collaborate on applying and improving the model, with priority given to large public financing initiatives.
For further information about the Revenue Predictive Model, contact data@uptimeglobal.org.
Related reading: The Data Dividend | Frequently Asked Questions