Meta’s Data Scientist’s Framework for Navigating Product Strategy as Data Leaders

Author: Chelsea Park

Introduction

One question that I often get is what makes Product Data Scientist special at Meta. My answer has always been “You are by default a product leader, navigating product directions with data”. This is true across all levels, from new grads to directors. Data scientists at Meta don’t just analyze data — they transform business questions into data-driven product visions that help building better human connections.

The challenge? Product strategy development exists across a spectrum of conditions. Here I’ll explore how data scientists at Meta can drive product strategies across four distinct scenarios defined by data availability (low to high) and problem clarity (broad to concrete).

The Four Quadrants of Data-Driven Product Strategy

Before diving in, let me introduce our framework. Great product strategy begins with a clear problem statement, identifies unique value, and outlines executable plans. Data scientists have a unique opportunity to shape this journey, but our approach must adapt to the terrain.

Press enter or click to view image in full size

Consider these four scenarios data scientists frequently find themselves in:

1. The Pioneer: Low data availability + Broad problem space
2. The Craftsperson: Low data availability + Concrete problem space
3. The Explorer: High data availability + Broad problem space
4. The Optimizer: High data availability + Concrete problem space

Here is how I would suggest approaching each scenario to drive product strategies as a Data Scientist.

Quadrant 1: The Pioneer (Low Data, Broad Problem)

This is perhaps the most challenging scenario, often happening in an early phase of the product strategy development in an uncharted area — you have minimal data and an ambiguous problem space. Many product leaders get uncomfortable here, but this is where data scientists can shine through first-principles thinking.

Strategic Approach

When faced with a broad problem like “improve advertiser value” with limited historical data, your role is to:

1. Identify north star metrics that mapped to the high level problems we’re solving. it’s okay if we only have a conceptual version — the version that can only be described by human language but not by math formula yet.
2. Design metrics framework that leads to north star metrics by conducting quantitative exploratory analysis, leveraging qualitative insights (survey, interviews etc). Your metrics framework should have math formulas available in some sub-areas by now. Each framework component should map to an area of problems to solve.
3. Generate initial data insights through lightweight experiments or observational studies, identify key problems to solve and key proxy metrics
4. Use the proxy metrics and early results to guide early decisions

When we were in an early phase of exploring Ads Passback Attribution Strategy to maximize advertiser value based on the incrementality attribution they planned to pass back to us, we had practically no data from advertiser yet and a vague need to “improve advertiser true value”. We created a simple but effective approach:

1. Set advertiser true value (conceptually) as northstar, designed metrics framework that ladder up to it including known and unknown metrics (we had 1 known metrics from exploratory analysis, 6+ unknown metrics from advertiser interviews when we first sketched it)

2. Designed minimum viable analytics (we defined the minimum data that we need to populate our formula and work with 3 advertisers to obtain them)

3. Conducted a quasi-experiment (2-week cycle) and that validated one key hypothesis. This helped us to narrow down to one precise question to solve

4. Established leading indicator framework and created product strategies to drive measurable improvements

Collaboration is important. Partner with PM and XFN team to define northstar metrics, translate business questions into testable hypotheses, and use analytics to yield insights.

A good strategy decides which problems we’re prioritizing in solving as well as those we choose not to solve. In low-data, broad-problem spaces, DS’s biggest contribution is narrowing the problem space through structured discovery.

Quadrant 2: The Craftsperson (Low Data, Concrete Problem)

Here, you have a well-defined problem but limited data to inform your approach. This scenario often occurs when launching new features or entering new markets.

Strategic Approach

Your primary value comes from:

1. Designing targeted data collection aligned to the specific problem
2. Developing creative measurement frameworks that work with sparse data
3. Leveraging analogous data from similar contexts

When we were developing our Privacy Preserving Machine Learning strategy, we were facing the challenge that the majority of the data used for model labels would be taken away. Our goal was to achieve the same level of prediction accuracy even when the data was missing, as if it were still available. Our approach:

1. We identified a similar context product where we stopped using similar type data and extracted performance patterns
2. Created a lightweight analytical framework that could work with minimal initial data based on the performance patterns extracted
3. Designed progressive data collection that would refine our understanding in stages

In this quadrant, focus on setting clear learning milestones rather than promising specific outcomes. The goal is to systematically reduce uncertainty around a concrete problem with iterative data learnings to update our beliefs.

In this quadrant, execution means designing the right measurement system that can validate hypotheses with minimal data.

Quadrant 3: The Explorer (High Data, Broad Problem)

This scenario presents a distinct challenge — you have abundant data but a not yet defined problem space. Many large, established product areas find themselves here.

Strategic Approach

Your primary contribution is:

1. Pattern recognition at scale to identify unrecognized opportunities
2. Segmentation and clustering to create structure in an ambiguous space
3. Insight translation that transforms data patterns into business narratives

When we were establishing a first ever eCommerce ads product strategy, eCommerce ads was already a $20B business through horizontal ads investment, however we weren’t really sure if we have done everything right for eCommerce advertisers and if we were satisfying their unique needs. We had all sorts of data but only a vague mission to “drive eCommerce advertiser value.” Our approach:

1. Developed an advertiser segmentations and Job To Be Done model that identified 3 x 3 distinct advertiser problems to solve
2. Created a gap analysis framework highlighting underserved advertisers and jobs to be done
3. Built an opportunity sizing model that quantified potential impact of addressing each gap

The key here is to structure the problem space through data, allowing the product team to move from broad exploration to targeted opportunities.

When collaborating in this quadrant, data scientists should lead with insights, not just analysis. Your role is to transform overwhelming data into clear strategic choices for your product partners.

Quadrant 4: The Optimizer (High Data, Concrete Problem)

This is where data science traditionally shines — you have both abundant data and a well-defined problem. The team knows exactly what metric they need to move and by roughly how much. But even here, there are strategic approaches that separate elite practitioners.

Strategic Approach

Your focus should be on:

1. Metric deep-dive and monitoring
2. Analytics modeling that uncovers non-obvious optimization opportunities
3. Continuous learning systems that adapt as conditions change

When we launched a new ads format that we want it to drive conversion improvement without trading-off organic engagement , we:

1. Deployed a multi-armed bandit system that continuously tested format variations
2. Built an analytics framework that isolated the impact of ads size from the format effectiveness
3. Created a feedback loop that automatically refine our understandings and feature launches based on metrics and counter metrics movements

When working with product teams here, elevate the conversation from short-term metrics to long-term learning. The danger in this quadrant is over-optimizing for immediate gains at the expense of long term sustainable growth. Data scientists should ensure optimization efforts align with the strategic trajectory.

Read more ….