The purpose of this study was to obtain a poverty data cluster in Pagar Alam City. The data collection of beneficiaries of the Program Keluarga Harapan (PKH) is not correct, the provision of assistance only pays attention to the criteria for poverty in general, so there are still many poor people who feel more deserving of PKH assistance. To overcome the problem of PKH recipients, it is necessary to cluster the community into various levels, so that the government can know the level of poverty of the community and can provide PKH assistance appropriately. The methods used in this study are CRISP-DM and the K-Means clustering algorithm. The attributes used are Identity Number, Name, Family Family Card Number, Poverty Rate, Pregnant Women, Early Childhood, Elementary School, Junior High School, Senior High School, Elderly, and Family Hope Program Recipient Group. This clustering process produced three clusters, namely cluster_0 as many as 156 people, cluster_1 as many as 82 people, and cluster_2 as many as 233 people. Furthermore, it was developed into a system with the Rapid Application Development (RAD) system development method. Thus producing a K-Means algorithm system to classify the poor in Pagar Alam City. The system test method uses black box testing with the alpha method and obtained database test results with a value of 4, interfaces with a value of 4, functionality of 4.42, and algorithms with a value of 4. In the testing process with UAT, in the system aspect got 87% of users agreed, in the user aspect 86% agreed, and in the interaction aspect 87% of users agreed. So it can be concluded that this system is worth using.