How do I use the output generated by ML and AI?
Comments
-
Machine learning models generate a guess of what they think the true value of the "target" or "label" column is. This is either a number in the case of a regressor, or a text string in the case of a classifier (or True/False).
When training a model it will look at data where we already know what happened and try to learn patterns that allow it to guess the correct outcome by combining the information we have about each observation.
This is useful from a business perspective when it is then fed new data that we do not know the answers to. And the predictions the model makes are used to make decisions.
For example, let's say we run a sales company and have a cold calling section. We have a section that generated leads for us. Thankfully they make a lot of leads, but we do not have enough agents to follow up on all the leads in time.
So what we want is way to rank the lead based on how likely it is to convert to a sale if called.
To do this we will have a number of leads and collect data about them in a table, then attach if they were called, and if they subsequently converted into a sale (it is very important here to understand the sale window, for some industries that means right there on the call, but sometimes that can be a 30 day window when it's more about upselling for example).
We take only the data where we did call the lead and then add a classifier tool to have it learn a model. We can check out the performance of the model (there is a lot to talk about here, please look at the underlying data section of the model).
Once we are happy with the model, we can add an export tool after the model and run an export on new leads that we have not called yet.
The model will append a prediction column that says if it thinks it will convert or not. It will also add a column of probabilities which tells you how certain it feels that the prediction is correct. So a prediction of True 99% is more likely (in the model's eye) to convert than a prediction of True 80%. And, inversely, a prediction of False 99%, is less likely to convert than a prediction of False 80%.
With this information we can now reorder how we call our clients and, hopefully, get more conversions by focusing our efforts on the best leads.
I would highly recommend whenever you use a model that you set up controls to measure how effective the strategy you are employing based on the model is and whether or not it is changing consumer behaviour. After all, no one can buy if we don't call them to offer the product, so there can be self biassing effects, but this leads to a larger conversation about implementing business strategies based on these predictions.
