What would you do if automated tests are frequently failing?