In real-life scenarios, it is mostly the case that you have to keep communicating with the business intermittently. You might have to build several models before concluding on a final production-ready model and communicate the results to the business.
An implementable model does not always depend on accuracy; you might have to bring in other measures such as sensitivity, specificity, or an ROC curve, and also represent your results through visuals such as a Gain/Lift chart or an output of a K-S test with statistical significance. Note that these techniques require business users' input. This input often guides the way you build the models or set thresholds. Let us look at a few examples to better understand how it works:
If a regressor predicts the probability of an event occurring, then blindly setting the threshold to 0.5 and assuming anything above 0.5 is 1 and less than 0.5 is 0 may not be the best way! You may use an ROC curve and take a rather...