The demand for gas-turbine (GT) based flexible power generation and mechanical drive is increasing due to the growing penetration of renewables and due to the need to quickly adjust production and operate at part load respectively. As efficiency operability low emissions, small footprint, availability and maintainability are of paramount importance, engine designers are leaning towards aircraft engine architectures that, with appropriate modifications mostly to the combustion system and turbine, can meet market needs. To leverage the large experience from aircraft propulsion, aero-derivative engines maintain the same architecture, with a high-speed shaft core, and a low-speed shaft driven by a multi-stage low-pressure turbine. While in aircraft engines power is adjusted by changing fuel rate and shaft speed, that go hand in hand, mechanical drive engines have more stringent needs that require changing the delivered power by keeping the shaft speed under control to guarantee the operation of the driven equipment (an LNG compressor or an electric generator). Therefore, the power turbine may deliver exit flow profiles and angles that put the turbine exhaust diffuser under severe off-design conditions, with the onset of large scale separations, large kinetic losses, and ultimately a significant drop on cycle performance. This paper describes Baker Hughes, a GE company experience in the CFD assisted design and scale-down testing of aero-derivative exhaust diffusers. The design incorporates the requirements of hot-end mechanical drive in multiple the power turbine operating conditions to determine the best compromise between peak design performance and off-design operability. The test in similitude conditions considered four relevant operating points. The inlet conditions matched with the power turbine exit profiles by the concerted action of swirl vanes and perforated plates, the design of which was heavily CFD assisted. Predictions matched measurements in terms of pressure recovery, kinetic losses, and exhaust velocity profiles. Different data post-processing and averaging were considered to properly factor in the diffuser losses into the overall turbine performance.