How Machine-Learning improved me as a Product Manager
Every year I set for myself the goal of learning a new technical skill. I don’t plan to become an expert on this new skill, at least not right away. Nevertheless, I love tech and the engineering side of things, and I love learning about the “tech-trends” that are out there for two main reasons. First, it keeps me updated and aware of what is happening in the world. Moreover, it opens my eyes to new possibilities and how I could be applying those in my day to day as a Product Manager. Second, it allows me to grasp, understand, and communicate better on technical matters. This second, at the same time, has two benefits for me. One, I can talk to the dev teams and engineers on the same level as them. And two, I can explain difficult technical concepts to business people who don’t quite grasp all this complexity.
This year I decided to learn machine learning. Again I don’t plan to become a Data scientist nor an ML-engineer. Yet, this is trending and in my opinion, it has a lot of opportunities in it that are yet to be discovered and implemented. In any industry that you can imagine machine learning is being implemented or is about to be in the upcoming years. Therefore my need of learning about it, and dive as deep as I can in it.
Despite its fancy name, machine learning concept and goal can be simplified in a few sentences. Imagine that you have a problem with tons of data. All this data can be related to the problem or mistakenly assumed to be related to the problem. Machine learning's overall outcome is to take all the data, see its correlation (or the lack of it), and based on it try to predict a result. Of course, this is an oversimplification of the whole concept. The ‘learning’ of machine learning, is that the system(better-called model) is able to ‘memorize’ this correlation, these patterns, and therefore make predictions.
By replicating this concept, machine learning can do multiple complex things. From enabling cars to drive autonomously, up to generating music by itself. It will all depend on how many patterns it ‘learns’ and how precise are these predictions.
Again, this is an over-simplification of machine learning, and there are tons of complex elements involved in creating an ml-model: getting the data, understanding the data, identifying and labeling the data, training the models, and generating an accurate outcome. Data scientists and ml-engineers are artisans.
How machine learning helped me become a better product manager?
Well, for starters, I needed a source of knowledge about machine learning and I got this amazing book written by Aurelién Géron:
With it, I learned of the steps that are needed in order to engage with a machine learning project. These steps for me were very similar to any kind of ‘cheat sheet’ you could create to prepare your self for developing a product. The steps are the following ones and I will explain how they are similar to product management:
Look at the big picture
The first step of looking at the big picture aims to make the person aware that the problem in front of them has a lot of things that are not immediately seen, or maybe hidden. It aims to try to find all the hidden connections and understand the real situation of a problem.
Product-wise: Not all problems that we are trying to solve are as simple or ‘obvious’ as it may seem. Sometimes we need to step back and realize that there is a bigger picture with hidden implications and impacts. That the problem may be expanding beyond our scope and that its roots may have an origin way out of our grasp. This will give us scope and understanding.
Get the data
Pretty straightforward and not much to add. For a machine learning project, you need the data, lots of it. And it needs to be the right data too. For a machine learning project, this could mean defining the data pipelines and sources for it.
Product-wise: It should be essential to have data when you are doing ‘product’. If you don’t have the data, get it. Data will provide a grounded base to start measuring the success of your product. Data will also be a compass guiding you towards where you should be heading next, as the next point explains.
Discover and visualize the data to gain insights
One of the main tasks before starting building machine learning models and jumping into the code is understanding the data that you have gathered. Data can come in multiple forms and is essential to understand it and how it relates within itself. Data can be pictures, videos, sounds, numbers, or text. The most important part of this is pre-identify the patterns and correlations that will help us guide the project towards the right path. It will show us what are the complex elements to identify and with this, where to focus next. It will be like having a rough map of the situation.
Product-wise: It is one of the most important tasks a product manager needs to do first. See and understand the data is essential to identify problems, correlations, and possible solutions. The insights a product manager gets from looking and understanding the data will lead to a good product manager in the right direction. It may expose the source of the problem and it may outline possible solutions from the very beginning. In both cases, this step will also allow you to see if you have all the needed data or if you need to take a step back and gather some additional elements. Visualizing the data will also give you direction and understanding of your following steps.
Prepare the data for machine learning algorithms
You have seen the big picture, gathered the data, and understand it. Now it is time to get your hands dirty. This step is pretty much similar to peeling all the fruits, and seasoning the meat before cooking a recipe. For machine learning, data needs to be in a specific ‘format’, so it can be read and understood by the machine. Also, those correlations that you identified first, need to be generated and translated into machine language so the system also gets to identify those correlations. This does not mean, that the system won’t find correlation on its own. You also need to split the data into training data, validation data, and testing data. These segments of the data will allow you to see if your models are accurate and realistic. It provides something to compare your results against.
Product-wise: Prepare your KPIs, define your dashboards, and set your test segments. Data, as mentioned before, will not only guide you but also will tell you if you are reaching your goal or not. By splitting the data, you as a product manager can prepare for testing a product, and measure its results against real uncontaminated data.
Select a model and train it
On machine learning, there are multiple predefined models that search for data correlations, make calculations regarding predictions or similarities, and provide an outcome. These models also managed multiple elements, like variation, loss, learning rates, distributions, and how in overall each one of those finds the most optimal outcome. In machine learning, it is important to understand how each of them works, where each one works bests and not, and what are the best approaches to use them. All of these will allow the person to choose the best model for the problem to be solved.
Product-wise: Select your best frameworks and your best solutions. Product management is not linear, therefore this step can happen in different moments of the product life cycle. Yet, the product manager needs to know and understand the product tools (just as data scientists need to know their models). It is important to know when to use what, and what will be the output of using each tool. Similar to this, when solutions have been defined (not yet implemented) the product manager needs to select one, and proceed to the implementation. Here is where data and testing will back up the decision and will help the product manager to select the best solution.
Fine-tune your model
The overall goal in machine learning is to obtain the most accurate model that allows you to re-use it and solve multiple times your problem. If you want to identify types of trees, then you want to have a model that identifies each type of tree as accurate as possible. This is achieved by fine-tuning the model. In other words, adding more specific data, or tweaking some parameters of the model to allow it to learn in a better way and predict more accurately.
Product-wise: Two words: testing and iterations. When you have selected a solution, or sometimes before selecting the ultimate solution you want to test. By testing your proposed solution, you get to fine-tune its outcomes while proving and testing if it is doing what you expected of it and if it is solving the problem in the ways it was expected. Is it reaching its expected goals? Is it generating new problems? Is it affecting other KPIs? Is it too complex? Does it need to be more user friendly? All these answers will be obtained by testing and all the outcomes can be improved by iterating on your solution. Remember that a product is never finished.
Launch monitor and maintain your system
This goes for both machine learning and product. Once everything has been properly tested and developed, it will go against its biggest trial: surviving when deployed. Launching a model or product could be exiting, finally, people will get to use it and see how it will behave out there ‘all by itself’. Yet it is only the first step of the real-life of the model/product. A machine learning model must probably need to evolve and adapt even after being deployed. New and different data may be available time after deployment, and the model will need to learn from it to improve its predictions. Likewise, a product will need to adapt to the user experience around it and the evolving necessities of the user. Here, data will also be essential to monitor the efficiency of your model/product and will guide you towards the right direction of improvement, maintenance, and scalability.
So… What ML really did for me?
Overall, machine learning helped me improve as a product manager by reinforcing, outlining, and reflecting on the main steps and the best approach towards understanding and solving a problem. From overseeing all the elements to understand the real reasons behind something, to testing and tweaking in search for the best outcome( 95% is good but small tweaks can take you up to 98%). Machine learning also made me more conscient of the dangers of false positives, and how important it is to have the right translation and understanding of data.
I have become more conscient of the meaning of the data, therefore I have made myself more data-driven than before.
So, you may be wondering, Do I need to learn machine learning to be a good product manager? The straight forward answer is “not really”. Unless, you are aiming to become a machine learning product manager, or your products surround ml, deep learning, or AI. There is no bounded reason why you “should” learn machine learning.
Nevertheless, learning about ML is fun and interesting and will open your mind to all the possible opportunities that there are in the near future by implementing it. You may find similarities between what you do and ML, just as I did with the product. And this, in the end, may trigger an improvement for your self.
Want to learn more about product and growth? Reach out and book a session in any of the following platforms:
