Despite billions pouring into AI, research suggests most projects stall or fail because the underlying data isn’t ready.