Multilateral organizations often impose conditions on procurement and construction procedures when financing public goods provision by low- and middle-income country governments. What do these do in practice? While this question has been fiercely debated by aid-receiving governments, multilateral organizations, and academics, it is difficult to answer causally due to the endogeneity of project choice and the relatively small sample size of projects funded via multilateral financing.

To provide causal micro-evidence on this topic, we leverage an unusual feature of Kenya’s nationwide electrification program: the quasi-random allocation of multilateral funding sources across nearby villages, with African Development Bank funded sites following turnkey contracting and World Bank sites following segregated contracting procedures and strengthened inspections.

We collect detailed on-the-ground engineering assessments of conductors and poles, minute-by-minute household-level outage and voltage data, and household surveys on connection quality and usage, and analyze a rich set of procurement contracts and inspection reports. We find that segregated contracting delayed construction completion at the average site by 9.6 months relative to turnkey contracting, but these procedures improved on-the-ground construction quality by 0.6 standard deviations, indicating a trade-off between the different approaches.

To disentangle the roles of two key dimensions of donor conditionality—contracting versus audits—we implement a randomized audits scheme mimicking the latter, and find that this improves household connectivity and electricity usage. In this context, streamlining contracting procedures that generate delays—such as contract segregation—while strengthening those that improve quality with minimal cost—such as ex post inspections—could improve project outcomes. Given the current regime, the net impacts of short-term delays and long-term grid resilience could reasonably be argued to favor either segregated or turnkey contracting procedures, depending on time preferences and technical assumptions