Data quality from a life-cycle point of view

For at least two generations, information technologists have extolled the virtues of new technologies to automate work, to improve decision-making, and to transform businesses and entire societies. Actual progress has been slower and more uncertain. One reason, perhaps the biggest, is poor quality data. The old saying “garbage in, garbage out” has proven true. Today, with the ubiquitous nature of AI, the implications of utilizing poor quality data are even higher. Exacerbating this, there are few solutions to be found in the literature, despite significant research on data quality. With this background, we propose a new data life-cycle model building upon the data quality foundations developed by researchers at Bell Labs some thirty years ago, integrating key aspects of the data quality literature since then, and designed specifically to support the adaption of proven quality improvement methods, such as statistical process control, to data quality. We find these historic foundations hold up quite well in a modern era, with a major exception of the dramatic increase in unstructured data, and their subsequent use in AI systems. The life-cycle model developed at Bell Labs was focused on structured data, and without the thought of modern large language models or agentic AI. We argue that the proposed data life-cycle model is more amenable to AI and especially to supporting data quality improvement efforts.