The Consumer Expenditure Quarterly Interview Survey is an ongoing panel survey of U.S. households in which sample units typically receive the same survey protocol during each interview. Because of the high burden associated with the survey request, the U.S. Bureau of Labor Statistics is exploring alternative designs that, if implemented, would change many features of the data collection process. One such alternative is adaptive matrix sampling. In general, matrix sampling involves dividing a survey into subsets of questions and then based on some probabilistic mechanism administering each to subsamples of the main sample. To potentially compensate for the resulting loss of information, as not all questions are asked of all sample units, we propose an adaptive assignment of subsampling probabilities based on data from the first interview. We use historical data to explore potential efficiency gains incurred by the use of this form of adaptive matrix sampling, develop point estimators based on simple weighting adjustments for expenditures collected under this design, and evaluate their variance properties.