>it might be 80 or 90% of what they need
Yeah. Not good enough. For the life of me, I cannot remember the name, but I remember we used another IBM tool to translate cobol to Java, and what it generated was horrific dogshit Java that also was 80-90% of the way there. Thing is that, unless you have very comprehensive testing laid out (lol. Find me a mainframe shop with comprehensive testing and I’ll smash my testicles in a hydraulic press), it’s impossible to know what it missed, and thus the 80-90% of the way there was effectively 0.
We never did push forward with this tooling, but I strongly suspect that had we, the ROI would have been vastly net negative after horrendous business impacts.