Smart meters for firms in Ethiopia: Evidence from a randomised experiment

Policy brief Energy and Sustainable Growth

Ethiopian SMEs face frequent power outages and weak electricity systems, while low awareness and revenue losses hinder investment in a reliable supply. A large RCT will test smart meters as a solution to improve billing, reduce theft, and enhance productivity.

  • Ethiopian small and medium enterprises face severe electricity constraints: firms experience, on average, nearly 24 power outages per month, and 71% have no backup power source.
  • Electricity theft, billing errors, and delayed payments undermine the Ethiopian Electricity Utility’s (EEU) ability to invest in grid improvements – a cycle that smart meters could help break.
  • An ongoing randomised controlled trial (RCT) across 1,583 firms in Addis Ababa and Sheger City will provide the first rigorous causal evidence on smart meter effects for businesses in a low-income country.
  • Awareness of smart meters is low: only 33% of firms had heard of the technology at baseline, suggesting rollout should be paired with targeted information campaigns.
  • Findings will directly inform EEU’s planned installation of 125,000 smart meters for three-phase SME customers, with broader lessons for sub-Saharan Africa.

Reliable electricity is essential for firm productivity and economic growth. In Ethiopia, however, small and medium-sized enterprises (SMEs) face some of the most severe power supply constraints in sub-Saharan Africa. Poor billing systems, electricity theft, and limited grid monitoring prevent the Ethiopian Electricity Utility (EEU) from generating sufficient revenue to invest in infrastructure. The result is a vicious cycle: unreliable power forces firms to rely on costly diesel generators, while the utility lacks resources to improve service. Smart meters – which enable remote readings, automated billing, and remote disconnection – offer a promising solution. Yet rigorous evidence on their effects for firms in developing countries remains almost entirely absent (Demeke et al., 2026).