Module 2 - Machine Learning for Product Managers

 

Machine Learning for Product Managers — [2/6]

Subscribe to our Youtube Channel — Technomanagers

What will you learn in this course?

In this course, we will tell you how to turn the business need into a machine learning problem, and then make a prototype out of it. If that performs well then how to deploy and productionise it.

This course is divided into 6 modules.

  1. In the first module, we will see, see the introduction to machine learning
  2. In the second module, we will see when to say yes or no to machine learning. Machine learning is very mainstream nowadays so it is very important for a manager to know when to say yes and when to say no.
  3. In the third module, we will see how to use machine learning weapons.
  4. In the fourth module, we will see how to prepare the training data for machine learning, that’s the first step in developing any machine learning model.
  5. From there we will move on to our fifth module where we will learn how to build the machine learning model
  6. Finally, in our last module, we will see how to deploy the model.

Book a Product Mock interview Session with Me

Module 2— When to use Machine Learning and when not to?

Why use Machine Learning?

But before going into it, I want you to know the AI flywheel concept. AI flywheel concept states that when AI technologies are integrated into the product perfectly then it kind of creates a positive feedback loop which makes your product better and better.

For example, you have one product that is used by a number of users and now you have integrated the AI technology. The user will generate the data and this data will again feed back into the model which will make our model accurate and make our product better for the users and which intern will attract more users, more data will be generated, and so on.

Through this process, your product will become better and better and become a great product for users in future

When to use Machine Learning?

A key skill that every manager working with machine learning should develop is when to use machine learning and when not to.
Why it is important?
Because machine learning projects are expensive, complex, and have a high failure rate so it is important for managers to know where they want to put their resources to get the desired return.

So now let’s see when should you use machine learning. You should use machine learning when all these points are valid simultaneously, if any of the points are not valid do not use it.

  1. The first thing is data, You have to see if you have got enough data and if the quality of data is good. Because in machine learning everything starts with data and it would be the first step toward building any machine learning model.
  2. The second thing is the complexity of the problem, if the problem which you are trying to solve can’t be solved by traditional rule-based programming then you can think of using machine learning.
  3. The third thing is the scale of the problem, if the problem which you are trying to solve will be scaled to a larger audience in the future then you can think of machine learning.

All these things should be valid for example if the nature of the problem you are trying to solve is complex but isn’t affecting users at a very large scale then machine learning is not the right choice for you

Let’s see one example of when machine learning is the right choice to solve a problem.

Personalization: That's a complex problem you can’t write millions of rules for millions of users. So the scale and the complexity of the problem are high. If you get the correct data, then Machine learning is the right way to solve this problem.

When NOT to use Machine Learning?

We have seen when to use machine learning, now let’s see when not to use machine learning.

As we have already said, don’t use machine learning when

  1. Your problem can be solved by traditional rule-based programming.
  2. Additionally don’t use machine learning when you require 100% accuracy. Because machine learning is about probability and if the use case for which you are using machine learning has some regulatory or legal implications then machine learning is not for you.
    Let’s say if you are solving a problem, and if your output gets wrong anytime you will be sued by legal for 100s or 1000s dollars then it will outweigh the goodness which it brings incorrect output.
  3. The third situation when you should not use machine learning is when you have to explain the complete interpretation of the results. because at the end of the day it is getting patterns from the data and no fixed rules are there.
    For example, you are a bank and you have a machine learning model which predicts whether to give a loan to a user or not. In this case, if the user protests about not getting a loan, then to make the situation better, you should have to know why the model made that decision at the first place which will become difficult because the ML model will be a black box for you.
    Although Some machine learning models like linear regression or decision tree have interpretability to some extent complex models don’t
  4. The fourth situation when you should not use machine learning is when you don't have enough data or the data which you have is biased or sparse which means the quality of the data is not good. Or it may happen that the data has some privacy and legal concern then don’t use machine learning as a problem-solving option.

With this, we have come to the end of module 2.
Stay tuned for module 3

Comments

Popular posts from this blog

Guesstimates for Product Interviews, Tips to ask clarification Questions

Design Uber for Kids. [Product Management Interview Question]

System Design Concepts for Product Managers — Part 1