A Practical Guide to AI for Product Managers

11 min read2 days ago

often come across articles discussing the impact of Artificial Intelligence (AI) on the work and careers of product managers and product development teams.

However, few explore in a practical way how PMs and development teams can apply AI to fulfill their mission: solve customer problems with technology in a way that generates business results.

In this article, I provide a direct and practical guide to help product managers, designers, and engineers understand how to use AI in product development to deliver real solutions for customers and boost business outcomes.

AI is a tool

Before diving into the concepts, it’s important to clarify that AI is just a technology tool. Let’s remember:

Product management is about developing products that solve customer problems through technology in a way that generates business results.

AI is just another tool to help us solve customer problems while driving business results.

What kind of problems can we solve with AI?

AI allows us to solve various types of problems, such as:

Predicting or recommending future data points in a dataset. Examples: predicting real estate prices, recommending books, movies, or music, predicting the likelihood of a customer canceling a subscription (churn), and demand forecasting for inventory control.
Segmenting or classifying data. Examples: grouping customers with similar purchasing profiles, classifying emails as spam or not spam, medical diagnosis, such as distinguishing cancerous cells in exams, and automatic product categorization in an e-commerce platform.
Recognizing patterns in text, videos, and images. Examples: facial recognition for unlocking smartphones, automatic text translation (Google Translate), fraud detection in financial transactions, and surveillance systems that identify suspicious behavior.
Interacting intelligently and adaptively. Examples: virtual assistants that learn from conversations (ChatGPT, Siri, Alexa), customer support systems that tailor responses based on user history, and games that adjust difficulty based on player performance.
Automating and optimizing complex processes. Examples: AI in autonomous vehicles, interpreting traffic and making real-time decisions, robots in factories that adjust production based on demand, AI models for discovering new drugs.

Algorithms

An algorithm is a set of rules or logical steps used to solve a problem or perform a task. It can be followed by humans (like a recipe) or by machines (like a computer program).

Every digital product relies on algorithms, such as an authentication algorithm, which processes user credentials to grant access to a product.

There are several algorithms used in AI. Here is a list of the main ones:

Simple Linear Regression: predicts a numerical value based on a single variable.
Example: Predicting the price of a property based on its size.
Multiple Linear Regression: uses multiple independent variables to predict a numerical value.
Example: Predicting the price of a property considering size, number of rooms, and location.
Non-Linear Regression: models more complex relationships where data does not follow a straight line.
Example: Predicting user growth in an app, which may follow an exponential pattern.
Logistic Regression: classifies something into two categories (yes/no, approved/rejected).
Example: Determining whether a customer will renew a subscription or not.
K-Nearest Neighbors (KNN): finds the nearest neighbors to a new data point and uses them for classification.
Example: A recommendation app suggests restaurants similar to the ones you have visited.
Decision Trees: makes decisions based on “if… then…” rules.
Example: Chatbots that follow a structured flow in customer support.
Random Forest: combines multiple decision trees to provide more reliable predictions.
Example: Fraud detection systems in banking transactions.
Support Vector Machine (SVM): finds the best boundary to separate different data categories.
Example: Classifying emails as spam or not spam.
Artificial Neural Networks (ANNs): simulate the functioning of the human brain to recognize complex patterns.
Example: Facial recognition for unlocking smartphones.
Convolutional Neural Networks (CNNs): specialized in image processing.
Example: Diagnosing diseases from medical imaging.
Recurrent Neural Networks (RNNs): process sequential data, such as text or time series.
Example: Word suggestion when typing on a smartphone.
Transformers: process long texts and understand context more efficiently.
Example: ChatGPT and intelligent virtual assistants.
Genetic Algorithms: simulate natural evolution to find optimized solutions.
Example: Optimizing delivery routes to reduce costs.
K-Means (Clustering): groups similar data without predefined categories.
Example: Segmenting customers based on purchasing behavior.
Autoencoders: reduce data dimensionality and detect patterns without supervision.
Example: Noise reduction in images and anomaly detection in financial transactions.
Unsupervised Neural Networks: learn hidden patterns in data without labels, identifying clusters or generating new samples.
Example: Generating realistic images with GANs, detecting anomalies in financial data.
Generative Adversarial Networks (GANs): use two competing neural networks (generator and discriminator) to create realistic synthetic data.
Example: Generating realistic images, deepfakes, digital art creation, voice synthesis, and medical data simulation.
Q-Learning: classic reinforcement learning where an agent learns through trial and error to maximize rewards.
Example: Training agents to play simple games, like the snake game.
Deep Q Networks (DQN): an enhanced version of Q-Learning that uses neural networks to learn optimal actions in complex environments.
Example: Training AI to play Atari games or control autonomous vehicles.
Policy Gradient: reinforcement learning based on policy, where the agent learns the best action directly without needing to store rewards.
Example: Teaching robots to walk or developing strategies for complex games like AlphaGo.

How do algorithms become intelligent?

These are the main algorithms used in AI, but how do they actually learn? Through machine learning (ML), a field of Artificial Intelligence that teaches machines to learn patterns from data without being explicitly programmed for each situation. The algorithms listed above are all used within ML, but each in a specific way.

ML algorithms can be divided into three main types:

Supervised Learning: In this approach, the algorithm learns from labeled data, meaning data that already has a known answer.
Examples of algorithms: Simple Linear Regression, Multiple Linear Regression, Non-Linear Regression, Logistic Regression, K-Nearest Neighbors (KNN), Decision Trees, Random Forest, Support Vector Machine (SVM), Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Transformers.
Practical example: Predicting whether a customer will cancel a subscription based on purchase history.
Unsupervised Learning: The algorithm receives unlabeled data and must find patterns on its own.
Examples of algorithms: K-Means, Genetic Algorithms, Autoencoders, Unsupervised Neural Networks.
Practical example: Segmenting customers without knowing in advance which groups exist.
Reinforcement Learning: The algorithm learns through trial and error, receiving rewards or penalties for its actions.
Examples of algorithms: Q-Learning, Deep Q Networks (DQN), Policy Gradient.
Practical example: Training a robot to walk on its own or teaching a self-driving car how to navigate.

Where does Deep Learning fit in? Deep Learning is a subcategory of ML that uses deep neural networks to learn complex patterns. The most advanced algorithms, such as CNNs, RNNs, and Transformers, belong to Deep Learning.

If ML were a set of tools, Deep Learning would be like using a more sophisticated tool for more complex problems.

To use a simple analogy, algorithms are like recipes, and machine learning is the kitchen — the overall process of preparing food. Just as you can cook different dishes using different recipes, in ML, you can solve problems using different algorithms. You follow well-established recipes (algorithms like regression, decision trees, etc.), and you can adjust ingredients (hyperparameters), but the method is relatively straightforward.

Deep Learning, on the other hand, is like molecular gastronomy, a cooking style that uses scientific principles (physics and chemistry) to transform ingredients in innovative ways, creating surprising textures, flavors, and presentations. It applies advanced techniques (deep neural networks) to create something new. It requires more resources (high computational power, large amounts of data) and can produce impressive and sophisticated results (such as AI generating images or writing texts).

Are model and algorithm the same thing?

Now that we understand the main algorithms, let’s clarify a common question: are model and algorithm the same thing?

Good question! Model and algorithm are not exactly the same, but they are closely related.

An algorithm is a recipe or set of mathematical rules that define how data will be processed to learn a pattern. It serves as the foundation for building the model. For example, the Linear Regression algorithm defines a mathematical equation to find the best-fitting line for a dataset.

A model, on the other hand, is the result of training the algorithm with specific data. It is the “learned” version of the algorithm, adjusted to solve a specific problem. For example, if we train a Linear Regression model to predict apartment prices based on size, the final model will have specific coefficients and weights that were fine-tuned using the training data.

Going back to our cooking analogy, the algorithm is the bread recipe — it defines the general steps to make bread, while the model is the baked bread. It was made by following the recipe but can have variations depending on the ingredients used (data).

The algorithm is the technique that defines how data will be processed. The model is the final product — the trained version of the algorithm, ready to make predictions or classifications on new data.

Does AI exist without machine learning?

Yes! Not all Artificial Intelligence (AI) uses Machine Learning (ML). There are many examples of AI systems that follow predefined rules without learning from data. Rule-based systems are still used when the problem is well-defined and does not require continuous learning. These are known as deterministic algorithms. Some examples:

Rule-Based Chatbots (Without ML): Before modern chatbots that use Natural Language Processing (NLP), many companies used fixed decision flows for customer support.
Example: A support chatbot that follows a predefined script, responding based on keywords.
Limitation: It does not understand language variations and does not improve with new interactions.
Business Rules in Digital Products: Many products apply deterministic AI to define automatic actions without machine learning.
Example: An e-commerce platform that blocks a user after three failed login attempts.
Another example: A banking app that automatically declines transactions above a certain limit without ML verification.
Simple Recommendation Engines (Without ML): Many products implement fixed-rule recommendations, without learning from user behavior.
Example: A streaming app that recommends movies based on popular genres, without considering personal viewing history.
Limitation: It does not personalize recommendations for each user.
Simple Virtual Assistants: Some older voice assistants followed fixed command-based scripts.
Example: A phone system that responds to commands like “Press 1 for technical support,” without interpreting natural language.
Rule-Based Pricing Algorithms: Some products adjust prices based on fixed rules, without learning from purchase patterns.
Example: A ticketing website that automatically increases prices when fewer than 10 tickets remain.
Limitation: It does not dynamically adapt to changes in consumer behavior.

Deterministic AI is nothing more than the traditional algorithms we have always used, where rules and decisions are explicitly programmed (for example, a classic “if this, then that” logic). These algorithms do not learn from data; they simply follow predefined instructions.

AI becomes truly interesting when Machine Learning (ML) comes into play, as algorithms start learning patterns from data and evolving over time. Instead of being programmed for every specific situation, they adjust their behavior as more data is processed.

The more data we have, the better the model becomes (as long as the data is of good quality). ML models do not just execute fixed rules; they identify subtle relationships that might go unnoticed, continuously refining their predictions and recommendations.

Machine Learning Models vs. Deterministic Algorithms

Models only exist for algorithms that use Machine Learning (ML) or Deep Learning (DL) because, in these cases, the algorithm must learn patterns from data.

Deterministic algorithms do not generate models because they follow fixed rules and do not learn. Since there is no learning, there is no need for a trained model. All logic is already embedded in the code itself or in rules defined by the developer.

Now, let’s summarize all we’ve learned in a table.

And, in one image:

Artificial Intelligence, Machine Learning, and Deep Learning

Conclusion

Remember:

Product management is about developing products that solve customer problems through technology in a way that generates business results.

AI is just another technology to help us solve customer problems while driving business impact. Now that you have a better understanding of AI, Machine Learning, Deep Learning, algorithms, models, and learning types, here are some tips on how to use AI to create products:

Start with the problem, not the technology: do not implement AI just because it is trendy. First, understand the real problem that needs to be solved and determine whether AI is the best solution or if a simple deterministic algorithm is enough. Companies that integrate AI strategically — not just as an “extra feature” — create real advantages. Think about how AI can reduce friction, personalize experiences, or automate repetitive tasks to improve your product.
AI relies on good data: the quality of data matters more than the complexity of the model. Without organized, clean, and representative data, no model will be effective. Invest time in collecting and structuring data before thinking about the algorithm. Unlike traditional software, an AI-powered product may require large volumes of data to function well. Often, it makes sense to start with simple rules (deterministic algorithms) and evolve to AI as data accumulates.
Define success and manage expectations: AI is not magic and will always have limitations. Set clear metrics to measure its impact on customers and business, and monitor the model’s performance over time.
Prioritize efficiency and scalability: AI models can be heavy and slow. Evaluate response time, computational cost, and impact on user experience to ensure that the solution is practical and viable.
Evolve iteratively: do not try to launch a complex system from the start. Begin with simple rules and, as data and product maturity grow, implement AI where it truly makes a difference. The work does not end when the model is launched. Continuously monitor performance metrics, adjust hyperparameters, and apply iterative improvements to keep the model effective over time.

Now, you have the essential knowledge to assess whether AI is the right technology to solve customer problems while driving business results. Enjoy exploring!

Workshops, coaching, and advisory services

I’ve been helping companies and their leaders (CPOs, heads of product, CTOs, CEOs, tech founders, and heads of digital transformation) bridge the gap between business and technology through workshops, coaching, and advisory services on product management and digital transformation.

Digital Product Management Books

Do you work with digital products? Do you want to know more about managing a digital product to increase its chances of success, solve its user’s problems, and achieve the company objectives? Check out my Digital Product Management books, where I share what I learned during my 30+ years of experience in creating and managing digital products:

Digital transformation and product culture: How to put technology at the center of your company’s strategy
Leading Product Development: The art and science of managing product teams
Product Management: How to increase the chances of success of your digital product
Startup Guide: How startups and established companies can create profitable digital products