Batch as much as possible. For companies that absolutely must have actual ML the temptation to make everything real-time/on-demand is high but sometimes (often

Batch as much as possible. For companies that absolutely must have actual ML the temptation to make everything real-time/on-demand is high but sometimes (often) unnecessary. Take a serious, hard look at where that line is. It’s like companies that need simple rules or analytics reaching for ML: an unnecessary expense in terms of development time and hard cash.