We study the extent to which collusion can explain the under-provision of clean sanitation technologies in developing countries. Using latrine desludging services in Dakar as a case-study, we document that prices are 40% lower in competitive areas than in areas where prices are coordinated by a trade association. We then develop an experimental just-in-time auction platform with random variation in several design features to formally test for collusive conduct and estimate the welfare costs of imperfect competition. Consistent with the collusion hypothesis, we find that bidders systematically avoid competition by placing round bids and refraining from undercutting rivals. We use a K-means clustering algorithm to classify bidders as competitive, collusive, or na¨ ıve and simulate counterfactuals in which non-competitive bidders are replaced with competitive bidders. This would significantly increase take-up of the improved sanitation technology, with back-of-the-envelope calculations suggesting improvements in health of a similar magnitude to those from building improved sewerage systems.