Booz Allen Hamilton Interview Question

Why would the same model generate different success rates within the same dataset?