It seems so obvious to us that this is a short term win at the expense of the long term. I wonder if data driven companies like this are locked into local minimums via tons of short term optimizations to boost metrics.

In optimization algorithms like stochastic gradient descent, some form of entropy (crucially temporarily decreasing fitness) or exploration (having parallel evaluations of the fitness function) is necessary to not get stuck. What mechanism incentivizes this in a large corporation, where every initiative must aim to improve the business and there is the capability to actually measure it all?