A wrong model could be useful if the action (based on the module) can compensate the models’ error effectively. Usually, you need to know some properties of the model’s error.
Even a wrong model could be very useful. For example, the earth is flat. That wrong model setup the question correctly and so that people could start thinking the shape of the earth.