This column is by Sthir Malkapur, Founder at Giving Privilege
Product Managers have enthusiastically adopted the data-driven approach to building products and have learnt not to rely solely on experience. Very few features today escape A/B testing. For some features it is a continuous process that helps the Build-Measure-Learn iteration. Intuition backed by data is a product manager’s most powerful weapon. If we have already made the shift towards data then why do we need Machine Learning, you ask? Why all the hype?
In this post, I am going to share why I believe every Product Manager should understand Machine Learning and where to start.
Why all the hype?
AI has been the talk of the year. Google announced it is an AI-first company and is building machine learning into all of its products; Gartner added Machine Learning to its hype cycle and all big tech firms are racing to buy AI startups, more than 10 major acquisitions in 2016 alone. Many products today already have Machine Learning baked into them. If you aren’t already working with a data scientist to build Artificial Intelligence into your product, you will soon.
Here are three reasons why a Product Manager should have technical knowledge of Machine Learning
1. Inconceivable Features
Today, a product’s features fall broadly into five categories – Basic features, Delighters, Customer Requests, Long-term Strategic features and Tech & Performance features. Machine Learning will add the category Inconceivable Features. These are features you have not thought of yet and never knew were possible but will add tremendous value to your product. Smart Reply in Gmail is a great example.
More user-generated data (data that is captured when a user interacts with your product) is created and captured today than ever before. This data is beautiful! If this data is one critical component, machine learning is the other. Now you can mix these two in various combinations to come up with mind-boggling features that previously didn’t seem plausible. Without understanding machine learning algorithms, it’s impossible to comprehend the possibilities.
2. Error Rate and Impact
The acceptable error rate of an algorithm depends on the product. The impact of these errors can either go unnoticed or cause major damage. Showing an unrelated ad to the user has little negative impact, however, classifying a human as an ape is unacceptable. Understanding accuracy and precision/recall trade-offs and associated impact on the product falls in the PM’s realm of duties. To gain this understanding, knowledge of machine learning is important.
3. Forecasting & Predictions
Machine Learning can be used to do many everyday tasks, such as forecasting product demand, determining the next city to launch in or analyzing the impact of price changes. Let’s take the example of product demand forecasting. Today, Product Managers work in excel, use factors such as seasonality, existing competitors, market conditions, previous quarter sales and so on, add weights and maybe even use the regression function in excel to come up with the best forecast for the following quarter/year.
Now imagine feeding all this data to a Least Square Regression Algorithm – the error rate is minimized, predictions are more accurate, new patterns are discovered all while the algorithm continuously learns and gets smarter. Many PM’s are working with data teams to use predictive analytics for important decisions. My point – for all predictions, and a PM does a lot of it, even simple Regression algorithms can be extremely useful.
How to get started?
Andrew Ng’s popular Machine Learning course on Coursera is the best course to gain a solid understanding of machine learning algorithms. This course that is considered an introductory course for Engineers is the best course for Product Managers for the following reasons –
The course covers the basics of many algorithms from Logistic Regression to Support Vector Machine to K-Means Algorithm. It also introduces advanced concepts such as neural networks. Collaborative Filtering, a technique used by recommender systems, Map Reduce and all aspects related to preparing data for machine learning such as feature scaling, feature engineering are also covered.
The course does flex your math muscles, but note that it is difficult to study machine learning thoroughly without understanding the math involved. Machine Learning is after all Statistics. The course covers basic linear algebra and calculus and builds the mathematical foundation that is necessary to understand the algorithms. By week 3 you will understand and be comfortable with equations that look like this –
Along with the math and intuition behind all algorithms, practical applications of each are explained.
High-level Programming Environment
The quizzes are essential to ensure you understood the material, however, given that a product manager doesn’t code, I’d say the programming assignments are optional. The course recommends using Octave or Matlab; both are high level programming environments that abstract out the implementation details and make them couple best programming environments for product managers, if one decides to complete the assignments. Based on the case I made above for Regression, I would highly recommend completing the Regression programming assignment (It took me few hours)
There are many other online machine learning courses. I found that some don’t cover the math behind the algorithms and point you to this course for in-depth understanding, and some place heavy emphasis on programming because they are geared towards Engineers.
All those folks who love math and are highly analytical, but were never really gung-ho about programming, you will love machine learning!
To build innovative futuristic products, tomorrow’s PM needs to continue to be everything she is today – the nexus between Customers, Engineering, Marketing and Sales, her product’s evangelist with relentless customer focus and a clear product vision, God – and also understand machine learning!
This is a curated post. The statements, opinions and data contained in these publications are solely those of the individual authors and contributors and not of iamwire or its editor(s). This article was originally published by the author here.