Predicting which open items will clear on their own and which will need intervention, so effort goes to the ones that will not resolve themselves.
Binary classification over item attributes — amount, category, age, account, source — trained on historical outcomes, or a time-to-clear survival model
scikit-learn classifier; rules baseline for comparison