Finding better machine learning model for default prediction I have two logistic regression models, model 1 and model 2. I want to find out which model is better. I'm predicting the default rate. I have compared the two models using Gini, plotted gini on month to month basis.
But I want a business metric which can help me show a monetary benefit that model 2 can give over model 1. A good metric to fascinate stakeholders and make them realize that yes model 2 will make them more money.
 A: 
I want a business metric which can help me show a monetary benefit that model 2 can give over model 1. A good metric to fascinate stakeholders and make them realize that yes model 2 will make them more money.

You are doing it wrong. Your aim is not to find a metric that would make your model look good so you could sell it, but to have a metric to validate the model. You should meet with the business and find out what metric that they care about and apply it to the problem. It can either prove that the model does work for the problem, or that it doesn't. It doesn't necessarily need to give results consistent with your machine learning metric. In fact, in real-life scenarios, it is not always the case that the best-performing model is the deployed one (e.g. sometimes you may prefer a model that is more easily interpretable, easier to maintain, or has faster inference time).
A: A ML or statistical model in isolation will make nobody any money. It's the actions you take based on the model predictions that can make money (or not). If model 2 leads to better actions than model 1, then this is a monetary benefit. If both models lead to the same actions, then a better performance by one model is worthless in business terms.
So you will need to understand how your models are actually used, and what actions are taken based on the predictions. Then you need to understand the monetary benefit of a given action under a given outcome. What do you do if your model yields a low or a high probability of default, and what are the monetary costs and benefits of each action under default or no default? Note that these costs and benefits are often themselves very hard to pin down - how much does it cost in terms of goodwill lost to turn away a potential customer that would not have defaulted (which we may not even ever find out, because we turned them away in the first place!)?
Once you have reasonable assumptions or even hard numbers for these relationships, you can simulate your business processes under predictions from both models. You may find out that one model yields a monetary benefit over another one - or not.
Yes, this is a lot of work, and you need to leverage your domain knowledge. There is no shortcut that will give you reasonable and reliable results.
