The answer does not depend on how big your plant is. It depends on what one specific machine stopping costs you, and how much warning you would need to avoid it.
Cost the failure honestly
Take one critical machine and total what an unplanned failure costs: lost production for the outage duration, the premium on expedited parts and freight, overtime labour, and any collateral damage a hard failure causes to couplings, shafts or product in process. Most plants find the figure is several times the repair invoice, because the repair invoice is the only part anyone ever writes down.
Compare it to the cost of knowing
Route-based vibration monitoring on a handful of critical assets is a modest recurring cost — a portable analyser, a defined route and an analyst's time each month. Set that annual figure against the failure cost above. For a machine whose stoppage halts a production line, monitoring usually pays back on the first avoided failure, and often on a fraction of one.
Understand what warning actually buys
A developing bearing fault typically announces itself weeks to months ahead. That window is what converts a 2am emergency into a planned Tuesday morning job: the part is ordered at normal freight, the labour is scheduled, the outage coincides with a stop you were taking anyway. The saving is not in the repair, it is in the circumstances of the repair.
Monitor by consequence, not by asset count
Monitoring everything is how programmes collapse under their own data. Rank assets by what their failure does to output, then monitor the top of that list properly rather than the whole list badly. A short, well-analysed route beats a comprehensive one nobody reads.
Data only pays when it changes a decision
The failure mode of most monitoring programmes is not technical — it is that reports arrive, get filed, and nothing is scheduled. A report should say what is degrading, how urgent it is, and what to do by when. If it does not change the maintenance plan, the measurement was an expense rather than an investment.