We study a dynamic problem of selling data without commitment to a budget-constrained receiver. The sender has access to a data-generating process, informative about a fundamental state, and can sell it either as granular observations (raw data) or summary statistics (information). Properly designed, such statistics ensure the residual uncertainty declines predictably along with the receiver's budget, supporting efficiency under gradual information sales. In contrast, selling raw data poses a risk that future observations increase residual uncertainty, exceeding the receiver's remaining budget. Consequently, selling data is inefficient if the fundamental is discrete and requires excess budget if the fundamental is non-Gaussian.