High-temperature heat pumps can decarbonize industrial process heat, but assumptions of constant compressor efficiency introduce uncertainty into predicted performance and cost. This contribution presents a data-driven model combining an artificial neural network with a k-nearest-neighbors reference search to predict compressor isentropic efficiency from inputs available during early-stage design. Developed using a dataset of approximately 4,000 manufacturer operating points, it provides condition-dependent efficiency estimates together with comparable real compressors for engineering validation. Integration into the ASTRA optimization framework supports realistic selection and comparison of industrial heat-pump configurations, working fluids and compression stages, including applications above 100°C.