Data Science for Product Managers: AMA with Gaurav Bubna, PM at Grab

Data science is no longer a niche skill reserved for specialized teams. It is now central to the product manager’s toolkit — especially in India’s fast-evolving tech ecosystem. The actual job is to translate data and AI capabilities into user value that can be measured and iterated on.

This lesson distills insights from Gaurav Bubna, a product manager at Grab and AI leader, who explains how PMs can master data science fundamentals without becoming data scientists. You will see how AI, analytics, and experimentation shape product decisions and how you can build a data-driven product mindset.

Data science is the foundation for modern product management

Product managers must ask: What problem am I solving? How do I measure success? Which data guides my decisions?

Gaurav highlights that data science for PMs is about framing problems quantitatively and designing experiments to validate hypotheses. It starts with defining clear objectives and success metrics. Without this clarity, data is noise.

For example, consider a signup funnel where new users take too long to engage with the product. You need to measure the time between key events — say login and first key action — using tools like Mixpanel or Segment. Then, design interventions (like an intro video), and compare conversion funnels before and after to test impact.

This is hypothesis testing in action. The PM’s role is to formulate hypotheses, design experiments, and interpret results — not to build machine learning models.

The essential data science concepts every PM must grasp

You do not need to become a data scientist, but you must understand:

• Metrics and KPIs: Learn to distinguish between vanity metrics and actionable KPIs that indicate progress toward business goals.
• Hypothesis testing and p-values: Know how to interpret A/B test results statistically to avoid false conclusions.
• Data sources and querying: Understand where data comes from — relational databases (MySQL), NoSQL (MongoDB), and how to query using SQL or basic Python/R if possible.
• Analytics tools: Familiarize yourself with Google Analytics, Mixpanel, Tableau, QlikSense — tools that help visualize and analyze data.
• Experimentation platforms: Use Optimizely, Google Optimize, or internal tools to run controlled tests.

Gaurav emphasizes that Google Sheets and Excel are often enough at the start, but scaling requires more sophisticated tools and querying skills.

AI and Machine Learning are reshaping product management

Gaurav leads multiple AI teams at Google, spanning NLP and computer vision. He explains that AI is not magic — it is a set of technologies that generate insights and automate predictions based on data.

For PMs, AI offers:

• Predictive analytics: For example, predicting user churn to proactively engage at-risk customers.
• Personalization: Tailoring product experiences using user behavior data.
• Automation: Reducing manual tasks with intelligent agents or recommendation systems.

However, the PM’s job is to translate model outputs into user impact. For instance, a model with 92% accuracy on a test set might still produce one wrong suggestion in every 12 interactions. The question is: Does that error rate erode user trust? If yes, the product needs a fallback or better UX, not just a better model.

Case study: Indian startups using AI for product innovation

An Indian startup shared how it leveraged AI to customize user experiences, which led to increased engagement and revenue. The key was integrating AI outputs seamlessly into workflows, not building AI for its own sake.

This example underscores what Gaurav calls the AI product mindset: AI is a means to an end — improving user outcomes — not the end itself.

The PM’s role in data-driven product management

Your actual job includes:

• Defining clear objectives: What user problem are you solving? How do you know when success is achieved?
• Choosing the right metrics: Focus on metrics that reflect user behavior and business impact.
• Designing and interpreting experiments: Use A/B testing to validate product changes.
• Collaborating with data and ML teams: Translate technical metrics into user-centric goals.
• Managing expectations around AI: AI is probabilistic and will err. Set realistic goals with stakeholders.
• Owning unit economics: Understand the cost implications of AI inference and data processing on your product’s margins.

Tools every PM should be familiar with

Gaurav recommends starting with:

• Data Querying: Basic SQL to extract data from relational databases.
• Analytics: Google Analytics, Mixpanel for user behavior tracking.
• Experimentation: Optimizely, Google Optimize for A/B testing.
• Visualization: Tableau or QlikSense to create dashboards.
• Scripting: Optional Python or R for advanced analysis if you want to be self-sufficient.

How to build a data-driven product culture

Building a data-driven culture requires:

• Cross-functional alignment: Engineering, design, data science, and PMs must share a common understanding of what data means.
• Continuous learning: PMs must stay current with data science concepts and tools.
• Experimentation mindset: Treat every product change as a hypothesis to be tested.
• User empathy with data: Combine quantitative data with qualitative insights.

Gaurav stresses that the best PMs balance data with intuition and customer empathy.

Demo: Building a simple predictive model with AWS SageMaker

To illustrate, a demo walks through creating a model predicting user engagement based on historical data using AWS SageMaker. The demo covers:

• Preparing data sets
• Training a model
• Deploying the model for inference

This exercise shows the technical workflow behind AI-powered features, helping PMs understand what’s involved and how to collaborate with ML engineers.

Common pitfalls PMs face with data science and AI

• Overemphasizing model metrics: Fixating on accuracy or F1 scores instead of user experience.
• Ignoring data quality: Poor data leads to poor models and wrong decisions.
• Treating AI as a feature, not a solution: Adding AI for marketing hype rather than user value.
• Lacking cost awareness: AI inference can be expensive; PMs must manage unit economics.

Slack Chat: Translating model accuracy into user impact

Field Exercise: Analyze your product’s data maturity (15 min)

1. Identify one key problem your product currently faces (e.g., low user engagement, high churn).
2. Define a clear success metric for this problem.
3. List the data sources you have access to that can help measure this metric.
4. Describe what experiments or analyses you could run to validate hypotheses about this problem.
5. Reflect on your current skills and tools — what do you need to learn or improve to execute this plan effectively?

Test yourself: Data-driven decision at a Series B Indian fintech

Video: AMA with Gaurav Bubna, PM at Grab

Where to go next

• If you want to master user research methods that complement data: User Research Methods
• If you want to build your skills in product metrics and analytics: Metrics and KPIs
• If you want to understand AI product strategy in depth: AI Product Strategy
• If you want to learn how to run effective experiments: Experimentation in Product Management
• If you want to improve your SQL and data querying skills: SQL for PMs