The first thing that is important before thinking about becoming a good product manager and artificial intelligence is to learn how to be a good product manager. ML techniques are an add-on, something you develop over time, and you can bring in data skills and at the same time be customer obsessed and understand the skill set that makes a good product specialist.
Translating unmet customer needs into crisp technical requirements.
Understanding and empathizing with customer pain.
Balancing the use of data with product/market/customer intuition to drive decisions.
Running effective meetings
Influencing without authority.
“Companies that understand how to apply machine learning will be best positioned to scale and win their respective markets over the next decade.” Peter Skomoroch
Deep learning theory is used to obtain unstructured data, for example, as classifiers. The simplest use of deep learning is to classify images and text in real life.
Scope definition is where the team comes together to understand the challenges they face with regards to visioning AI products and try to predict the efforts that will be involved. Experimentation happens just like any other product development life cycle. The team will need to create a prototype, but instead of creating a simple functional product, the team will need to choose the right approach to use the available data or choose a mathematical model that shows the best results in training. The main part of the work at this stage is related to the formation of the data necessary to train the pretrained model.
We often take pre-trained models, shape data that fits into them, and then try to get the best possible results.
The model will be tested in production with real data coming into it.
And then either the model delivers the expected output with the level of confidence that the team saw during the training phase, or it fails due to differences between the actual data the model receives and the training data that is used. If not, the team needs to return to the drawing board.
A dataset is what makes up 90% to 95% of the work needed for any AI or Machine Learning project, and this is true for any data science project.
You cannot advance in ML without understanding the structure of CRISP-DM.
AI projects require feedback from both the product development process and the AI products themselves.
Because AI products are based on research, experimentation and iterative development are necessary. Unlike traditional software development in which inputs and outcomes are often deterministic, the development life cycle is probabilistic.
Understanding data is a huge part of the lifecycle of any product. Its completion is one of the strongest indicators of future success.
The product manager often has to balance the investment of resources against the risks of moving forward without fully understanding the data available to them. Obtaining data is also extremely difficult, especially in regulated industries. Once the data is in, the AI product manager's goal is to be able to guide data scientists, analysts, and domain experts to evaluate product-centric data and design a meaningful experiment. The goal is to make sure that there is a measurable signal about what data exists. A clear understanding of the relevance of this data, a clear vision of where to focus your efforts when developing features. One of the most complex and important phases of every AI project is data processing and feature engineering.
During a typical product development lifecycle, 80% of the time data scientists spend on feature development.
The project manager must ensure that the project plans take into account the time, effort, and people needed.
During the simulation phase, data is output and results are obtained.
Managers are usually responsible for ensuring that all components of a software product are properly compiled into binaries and organize the build script.
Monitoring. Forecast with which you train the model may not match the prediction you get from the real data. Comparing reality to training data and having biases in the system is essential for reliability teams or engineers to act on this model and make changes.
Problem Identification
Validate that you're solving the right problem
Design
Confirm understanding of data required
Specify model outputs, design and proposed UI
Data Collection
Collect data for and specify the requirements for training, test, holdout and validation datasets
Feature Engineering
Perform feature engineering, model training and offline validation
Model Deployment
Monitor the model in production, adjust for data drift, and iterate. 80% of the gains come from post launch data!
“But while ML grows more important, few Pms know how to integrate it in their own products.” Clemens Mewald
Where the prediction will be displayed, what the actual result is, how to make it actionable, and how to make sure the items in your shopping cart are more relevant to users, including new features. An example of learning about a new product is providing information and then recommending to users what they can change to improve the culture.
“Artificial intelligence and machine learning, as a dominant discipline within AL, is an amazing tool. In and of itself, it's not good or bad. It isn't the core of the problems in the world. ” Vivienne Ming
“Any damn fool can make something complex; it takes a genius to make something simple.” Pete Seeger
Chinmaya Madan, Microsoft Product Leader