Starting and learning
Machine learning for product managers
Product teams use machine learning to rank and recommend content, predict churn or purchase intent, route support tickets and power assistants that answer questions from documents. A product manager defines the user action it supports, the data it needs and how quality is measured before and after launch.
How product managers use machine learning in products
Ranking search results
A ranking model orders results by how likely each is to satisfy the user, learning from clicks and bookings. The product manager defines what a good result is and how ranking changes are tested.
Real exampleApplying Deep Learning to Airbnb SearchAirbnb describes moving search ranking from a gradient boosted decision tree model to neural networks, and the lessons from doing so in a real product.
Recommending content
Recommenders choose what to show on a home page or feed from a very large catalogue. Product choices, such as what counts as success and how to avoid repetition, shape the model as much as the algorithm.
Real exampleDeep Neural Networks for YouTube RecommendationsGoogle researchers describe YouTube's recommender as a deep candidate generation model followed by a separate deep ranking model, with lessons from running it at scale.
Predicting arrival and wait times
Regression models predict how long something will take, such as a trip or a delivery, and the estimate is shown directly to users. Accuracy and how errors feel to users both matter.
Real exampleDeeprETA at UberUber describes a deep residual network that refines route-planning arrival time estimates, which Uber uses for fares, pickup times and matching riders to drivers.
Routing support tickets
Text classifiers predict the topic of a support request and suggest replies, so agents spend less time sorting. Product managers decide when to automate and when a person reviews.
Real exampleCOTA at UberUber researchers describe COTA, a system that classifies customer support tickets and selects suggested answers for support representatives using machine learning and natural language processing.
Personalising difficulty and content
Models estimate what each user knows or wants and adapt what comes next. Success is measured with experiments on engagement and outcomes, not only model metrics.
Real exampleDuolingo BirdbrainDuolingo describes Birdbrain, a machine learning model that estimates how much each learner knows and how difficult each exercise is, to set lessons at the right level.
Can product managers learn machine learning?
Yes. Product managers need to frame problems as predictions, understand data and labels, read evaluation metrics and plan for model errors; they do not need to train models themselves. Machine learning fundamentals plus one small hands-on project are enough to work confidently with an ML team.
- Machine learning fundamentalsKinds of problems, training and test data, baselines, and how a model's errors show up for users.Learn it in Machine Learning
- Metrics and experimentsPrecision and recall, offline evaluation and A/B tests, and how to choose a metric that matches the user outcome.Learn it in Statistics
- Language models and assistantsHow language models work, prompting, retrieval and evaluation, so you can scope assistant features realistically.Learn it in Generative AI
- Enough Python to look at dataBasic Python and pandas so you can inspect a sample of data and a model's mistakes yourself.Learn it in Python
Moving from product management into AI product roles
If you already work in the field
Product managers already know users, metrics and trade-offs. Add how models are trained and evaluated, then volunteer for a feature powered by machine learning and own its success metric, error handling and launch plan.
If you come from data science or ML
Data scientists moving into product management need user research, prioritisation and communication with design and engineering. Start by writing requirements for a model you know well, and explain its errors in terms of user impact.
Roles that do this workAI product managerTechnical product managerProduct manager, machine learningProduct manager, search and recommendations
Tools you will use
| Tool | What you use it for |
|---|---|
| Google Colab | Runs Python notebooks in the browser so you can explore data and prototype a model without setup. |
| scikit-learn | Fits quick baseline models and computes precision, recall and threshold curves. |
| Evidently | Generates reports on data drift and model quality for monitoring a feature after launch. |
| promptfoo | Tests prompts and language-model features against a set of cases before and after changes. |
Public data to practise on
- Online Shoppers Purchasing Intention Dataset: Website session behaviour such as pages visited and time spent, labelled with whether the visit ended in a purchase.
- Online Retail II: Online store transactions for analysing repeat customers, basket contents and demand.
Your first project
Download the Online Shoppers Purchasing Intention dataset, predict whether a session ends in a purchase with logistic regression, and compare it with a majority-class baseline using precision and recall at two thresholds.
Common questions
Do product managers need to know machine learning?
Product managers on AI features need to understand what a model can predict, what data it needs and how it fails. They do not need to code models, but they must define the metric, the fallback and how quality is checked.
What does an AI product manager do?
Defines the user problem a model solves, the data and labels it needs, the quality bar for launch and how errors are handled in the product. They work with data scientists, engineers and designers through experimentation and launch.
How do you write requirements for a machine learning feature?
State the user action it supports, the prediction it makes, the metric and threshold for launch, the baseline it must beat and what happens when it is wrong. Include data sources, privacy constraints and how you will monitor it.
How is a machine learning feature evaluated?
First offline, on held-out data with metrics such as precision, recall or ranking quality, then online in an A/B test on the user outcome. Monitoring continues after launch because data and behaviour change.
Courses for these skills
- Practical Machine Learning in PythonChoose an anomaly method and interpret its scores.51 published lessons · more lessons in production
- Data Tools: NumPy, pandas, Matplotlib and scikit-learnPreserve document identifiers, revisions and access metadata.23 published lessons · all planned lessons open
- Statistics and Probability for Machine LearningUnderstand tails, robust summaries and base-rate effects.41 published lessons · all planned lessons open
- Generative AI: LLMs, RAG and AgentsStudy retrieval-augmented generation and evaluating grounded outputs.Syllabus available · lessons open as produced
New to programming or maths? Start with Python, or follow the machine learning path in order. Published lessons are free to read; compare Free and Pro.