A widespread example of this is linear regression analysis. Mathematically, it only works because we make assumptions about the data (and even these can be relaxed) & limiting the scope of the problems it can solve.
This also applies to delivering projects. A fully engineered & robust project could possibly require a great deal of human, physical resources and time. But if we start iterating over a list of problems your system solves (sorted by decreasing value), the project suddenly costs only a fraction of what it would otherwise.
This doesn't come for free though: your MUST understand the limitations & assumptions of your systems, otherwise you or someone will pay dearly. Look at Zillow & failure to understand its forecasting system. Look at so many other startups applying ML where it has no place.
Zillow's failure is the best case (though catastrophic): the principal paid the cost. But take something like Google, applying a fully automated system to make decisions about banning accounts. These are real world decisions that have a huge effect on people & businesses. In this case, Google saves money on providing human support & offsets the costs to third parties who often can't afford it.