How do we know whether algorithmic systems are working as intended? A set of simple frameworks can help even nontechnical organizations check the functioning of their AI tools.