##### Scenario Data scientists need a predictive model to estimate how the new 'Show similar products' button will influence buyer engagement. ##### Question Describe how you would build a model—including feature selection, algorithm choice, sample-size considerations, evaluation and decision threshold—to predict engagement uplift and support the launch decision. ##### Hints Discuss demographic and behavioral features, pick algorithm based on data size, cross-validate, and translate predicted uplift into a launch recommendation.

Meta machine learning prompt on building a causal uplift model for a "Show similar products" button, covering experiment data, pre-treatment features, uplift algorithms, evaluation, thresholds, and rollout decisions.

How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

What difficulty level is this interview question?

This is a medium difficulty Machine Learning question, commonly asked during Technical Screen rounds at Meta.

What role is this question designed for?

This question is commonly asked for Data Scientist candidates at Meta during technical interviews.

Build Predictive Model for Buyer Engagement Uplift

Predict Engagement Uplift for a New "Show Similar Products" Button

A new "Show similar products" button may change buyer engagement. You need to build a predictive model that estimates the causal uplift from showing the button and supports a launch decision.

Constraints & Assumptions

The goal is causal uplift, not just predicting engagement likelihood.
Use only pre-exposure features in the model.
Prefer randomized experiment data when available; if not, state the assumptions needed for observational data.
Connect the model output to a business decision: global launch, targeted rollout, further testing, or no launch.

Clarifying Questions to Ask

What is the primary outcome: click, product detail view, add-to-cart, purchase, or long-term engagement?
Is randomized treatment/control data available?
What unit is being scored: user, session, impression, product, or user-product pair?
Are there fairness, merchant-quality, or user-experience constraints?

What a Strong Answer Covers

Causal framing: ATE, CATE/uplift, treatment assignment, counterfactual outcomes, and assumptions.
Data setup with assignment, exposure, outcome, and pre-treatment features.
Feature sets: demographics, device, geography, history, product/category affinity, session context, price band, and inventory/recommendation quality.
Leakage controls, sample-size and power considerations, and sufficient data for heterogeneous effects.
Algorithms such as two-model uplift, T-learner, S-learner, causal forest, meta-learners, or regularized models depending on data size and complexity.
Evaluation using Qini/uplift curves, calibration by segment, policy value, A/B backtesting, cross-validation, and holdout performance.
Decision thresholds based on expected incremental value, user-experience cost, confidence intervals, and guardrails.

Follow-up Questions

Why is a normal engagement model insufficient for uplift?
How would you validate uplift estimates offline?
How would you prevent treatment leakage in features?
What if the model finds high uplift only in a small segment?

Predict Engagement Uplift for a New "Show Similar Products" Button

A new "Show similar products" button may change buyer engagement. You need to build a predictive model that estimates the causal uplift from showing the button and supports a launch decision.

Constraints & Assumptions

The goal is causal uplift, not just predicting engagement likelihood.

Use only pre-exposure features in the model.

Prefer randomized experiment data when available; if not, state the assumptions needed for observational data.

Connect the model output to a business decision: global launch, targeted rollout, further testing, or no launch.

Clarifying Questions to Ask

What is the primary outcome: click, product detail view, add-to-cart, purchase, or long-term engagement?

Is randomized treatment/control data available?

What unit is being scored: user, session, impression, product, or user-product pair?

Are there fairness, merchant-quality, or user-experience constraints?

What a Strong Answer Covers

Causal framing: ATE, CATE/uplift, treatment assignment, counterfactual outcomes, and assumptions.

Data setup with assignment, exposure, outcome, and pre-treatment features.

Feature sets: demographics, device, geography, history, product/category affinity, session context, price band, and inventory/recommendation quality.

Leakage controls, sample-size and power considerations, and sufficient data for heterogeneous effects.

Algorithms such as two-model uplift, T-learner, S-learner, causal forest, meta-learners, or regularized models depending on data size and complexity.

Evaluation using Qini/uplift curves, calibration by segment, policy value, A/B backtesting, cross-validation, and holdout performance.

Decision thresholds based on expected incremental value, user-experience cost, confidence intervals, and guardrails.

Follow-up Questions

Why is a normal engagement model insufficient for uplift?

How would you validate uplift estimates offline?

How would you prevent treatment leakage in features?

What if the model finds high uplift only in a small segment?

Build Predictive Model for Buyer Engagement Uplift

Quick Overview

Build Predictive Model for Buyer Engagement Uplift

Predict Engagement Uplift for a New "Show Similar Products" Button

Constraints & Assumptions

Clarifying Questions to Ask

What a Strong Answer Covers

Follow-up Questions

Write your answer

Build Predictive Model for Buyer Engagement Uplift

Quick Overview

Build Predictive Model for Buyer Engagement Uplift

Predict Engagement Uplift for a New "Show Similar Products" Button

Constraints & Assumptions

Clarifying Questions to Ask

What a Strong Answer Covers

Follow-up Questions

Write your answer