The notion of technical debt attracts significant attention, especially in the context of reconciling architecture and agile development. However, most work on technical debt is still largely informal and if it provides a formalization it is often ad-hoc. In this paper, we provide a detailed, formal analysis of decision making on technical debt in development.Using this formalization, we show that optimal decision making is not effectively computable in real-world situations and provide several well-defined approximations that allow to handle the problem nevertheless in practical situations. Combining these approximations in a single method leads to a light-weight approach that can be effectively applied in iterative software development, including agile approaches.