ML makes predictions while MIP makes decisions. When your problem involves complex tradeoffs between competing activities and allows for trillions of possible solutions, only MIP has the power to find the best or optimal one. MIP is often complementary to ML.
For example, instead of using just ML to decide which offer goes in front of which web customer, you can marry ML to MIP to choose a set of offers that drives the greatest profitability. Or consider predictive maintenance (e.g., elevator repair). ML can predict when certain types of failures are likely to occur, and MIP can then allocate and schedule the resources required to perform the needed maintenance at minimum cost.
Listen to this podcast to discover how machine learning and optimization can complement each other; the former making predictions about likely future business outcomes, and the latter suggesting appropriate actions to take in order to take advantage of these outcomes.