While a lot of companies are excited about the development of machine learning, its application in business. There are a few things that business folks who intends to use these technologies should be mindful of.
Supervised machine learning models actually generate predictions with probability as the main product.
For instance, a Customer Lifetime Value model generates a prediction on how much the consumer is going to spend, with certain confidence. Or a campaign offer model generates a probability on whether a client will accept the offer. This means that the prediction is just a prediction, that may not be true in some instances. This means there will be false positive or false negatives unless you got a 100% accurate model which...my suggestion is to double check all your model training and data collection processes if you get such a model because it is too good to be true.
Thus it is important as a next step, after implementing a machine learning model into the tech stack, to think about how to deal with false positive and false negative.
Here are a few considerations:
1) What are the associated costs for the circumstances and company, if there is a false negative and similarly for false positive?
2) Is there a way to minimise the cost of false negative and false positive?
3) Can we be aware of false positive and false negative and how fast can we realise them?
4) Can we reduce or manage the negative impact of false positive and false negative?
While it is great that we adopt "new" technology to improve on business process and decision making we need to understand that all tools are useful and useless depending on the circumstances. In this case, while machine learning provide better accuracy and handling of use cases, but it brings with it the challenge of handling false positive and false negative. And in the business context, false positive and/or false negative can bring challenges and issues to business.