The pressroom_tom point about customer-driven purchases is real. Half the servo presses I see installed weren't justified on traditional ROI — they were justified on "we lose the contract if we can't provide force curve data for every part."
That said, for anyone trying to build a traditional ROI case, the one line item that always gets left out is quality cost avoidance. Not just scrap reduction — the cost of sorting, rework, customer complaints, and 8D reports. On our 300T line, we tracked quality-related costs for 12 months before and after the servo conversion. Before: ~$45K/year in sort labor, rework, and one customer chargeback. After: under $8K. That $37K/year savings never showed up in the original ROI calculation because nobody thought to include it.
My rule of thumb for ROI estimates: whatever number you calculate, the actual payback will be 30% faster because of savings you didn't think to include. But also 20% slower because of integration costs you didn't budget for (training, tooling modifications, PLC reprogramming). Net effect: your estimate is probably close enough.