Skeletons in the closet: Why humanitarian energy projects fail and how we can learn from them

Access to sustainable energy is critical for dignity and resilience, yet humanitarian energy (HE) initiatives face systemic failures despite escalating global displacement and funding shortfalls. HE remains peripheral within Sustainable Development Goal 7, and humanitarian coordination structures perpetuate reliance on unsustainable free-distribution models. This study investigates recurring causes of HE failures and proposes pathways to transform these shortcomings into actionable learning for sustainable energy provision. We conducted a systematic review of 42 academic and grey literature, and nearly 20 semi-structured interviews with HE stakeholders, such as donors, NGOs, private sector actors, and UN agencies. We found that failures cluster into five interdependent domains: Technological (e.g., context-inappropriate designs), Economic (e.g., unaffordable services, unsustainable financing), Social (e.g., exclusionary planning), Implementation (e.g., poor coordination, short timelines), and Governance (e.g., policy barriers). Economic misalignment, particularly users' inability or unwillingness to pay, was the most prevalent reason for failure. Critically, implementation failures dominated interviews but were underrepresented in the literature, indicating systemic reporting bias. Free distribution models undermine market viability and community agency, while siloed operations and rigid funding cycles hinder adaptive learning. Transforming HE requires dismantling epistemic and operational barriers, including prioritising context-driven co-design, longer-term funding cycles, and candid documentation. Market-based approaches offer potential for impact but demand cultural shifts to reconcile humanitarian principles with commercial viability. The sector must confront a pivotal question: will the HE sector perpetuate cycles of duplication by locking failure away, or will it harness these experiences as catalysts for innovation through transparent, actionable learning?