# Optimal solution for false positive effect on business strategy

I am facing a challenge to reach an optimal solution. I will try to explain with an example :

Suppose I have created an algorithm to predict if a customer will subscribe to bank deposit plan or not. The model has already reached optimal performance. The marketing for this deposit plan is done my telemarketing team (which has certain resource capacity).

My model has some false positive which means these false positives will result in increased cost on resources and effort. My target is to reach an optimal value of target that needs to be set for telemarketing team which they can achieve with existing resources and my profit is also optimized .

Can anybody help to understand the technicalities and how this problem be approached ?

You need to set the cutoff to minimize cost (or maximize profit).

Cutoff: By default, models will tend to assign the highest probability. If they predict a 55% chance of A, and a 45% chance of B, they will predict A.

Cost: By default, all errors are of equal value. You can think of the cost matrix as looking like this:

But in this case, your errors aren't of equal value, so you want to reflect the gain and loss from client contact (what's the gain from having them sign up; what's the loss from false positives) and the gain and loss from not contacting them (what are you missing when you have a false negative, for example).

This might lead to this matrix:

Different software will have different ways of entering the cost matrix, so I'm not going to discuss that. Since you have your original predictions, you can also play around with setting the cutoff at .4, .6, .7, etc. by hand instead of .5 and see whether that changes the confusion matrix in a way that is more profitable for the bank.

This problem is common; it's often called the precision-recall tradeoff.