← Meta Interview Insights

Meta·Data Scientist·Onsite - Product Sense / Strategy·Senior

Senior
Jul 2026

Summary

This was a Meta DS onsite focused entirely on a single sprawling product case about Instagram Shopping. Three parts, all connected, and they expected you to hold the whole thing together across hypothesis validation, revenue sizing, and post-launch diagnosis. Brutal in scope but kind of interesting if you like this stuff.

Questions Asked (3)

Q1

Instagram is considering a new Shopping tab but can't directly observe off-platform purchases. What in-app signals would you use as proxies for off-app purchase intent, and how would you account for bias and build counterfactuals to validate the hypothesis that users discover products on Instagram but buy elsewhere?

A/B Testing & ExperimentationProduct Analytics & MetricsProduct Sense & Ideation
Author's notes

This part tripped me up more than I expected.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining clear proxy signals for off-platform purchase intent, such as saves, link clicks, and dwell time on product pages. Then, discuss methods to identify and mitigate biases in these proxies, and propose counterfactual analyses like holdout experiments or synthetic control to validate the discovery-to-purchase hypothesis.

Pro tip: Emphasize that proxies are noisy and biased, so triangulate multiple signals and use causal inference techniques to strengthen your argument. Show awareness of Instagram's business goals by linking proxies to potential metrics like incremental shopping tab engagement.

1. Identify Proxy Signals

List in-app behaviors that indicate purchase intent, such as saves, shares, link clicks, product page views, and time spent on shopping-related content. Prioritize signals with high correlation to off-platform purchases.

2. Assess and Mitigate Bias

Evaluate selection bias (e.g., only engaged users click links) and measurement bias (e.g., clicks may not lead to purchases). Propose techniques like propensity score weighting, stratification, or using negative controls to adjust for bias.

3. Design Counterfactuals

Suggest experiments like holdout groups (users who don't see the Shopping tab) or switchback tests to create counterfactuals. Alternatively, use observational methods like difference-in-differences or synthetic control if randomization is infeasible.

4. Validate Hypothesis

Define success metrics (e.g., lift in proxy signals, survey-based recall) and analyze whether the Shopping tab increases off-platform purchase intent. Use causal models to estimate incremental impact.

5. Iterate and Refine

Recommend continuous monitoring and refinement of proxies based on new data, and suggest combining multiple proxies into a composite score for robustness.

Key Points to Mention

  • Proxy signals: saves, shares, link clicks, dwell time, add-to-cart (if available), and search queries.
  • Bias types: selection bias, measurement bias, and confounding variables like user intent.
  • Counterfactual methods: randomized controlled trials (holdouts), difference-in-differences, synthetic control, and instrumental variables.
  • Validation techniques: cross-validation with external data (e.g., surveys), A/B tests on proxy metrics, and sensitivity analysis.
  • Business impact: linking proxies to downstream metrics like revenue or app engagement.
  • Ethical considerations: privacy and data usage policies when tracking off-platform behavior.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

Using the given assumptions (50M DAU, 10% weekly intent, 0.2 purchases per intentor at $45 AOV, 5% take rate, plus ad impression fees), estimate the weekly gross revenue from transaction fees vs. ads, then identify which three assumptions most dominate the estimate and describe what sensitivity analyses you'd run.

Pricing & MonetizationProduct Analytics & MetricsProduct Strategy
Author's notes

The math itself is straightforward once you set it up: transaction side is 50M times 0.1 times 0.2 times 45 times 0.05, ads side is 50M times 0.1 times 6 times 0.002.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly separating the two revenue streams: transaction fees from purchases and ad impression fees. Compute each using the given assumptions, then perform a sensitivity analysis on the most impactful assumptions to show how the estimate changes. Finally, prioritize the assumptions that drive the most variance and suggest specific analyses to test them.

Pro tip: Always state your assumptions explicitly and round numbers for mental math, but note that in a real scenario you'd validate these with data. Also, consider that ad revenue might depend on the same user activity, so avoid double-counting.

1. Calculate transaction revenue

Compute the number of weekly intents (50M DAU * 10% = 5M), then purchases (5M * 0.2 = 1M), then gross merchandise value (1M * $45 = $45M), and finally transaction revenue (45M * 5% = $2.25M).

2. Calculate ad revenue

Assume an ad impression fee per user per week (e.g., $0.10) and multiply by DAU to get weekly ad revenue (50M * $0.10 = $5M). If not given, state a reasonable assumption and note it as a key driver.

3. Compare and identify dominant assumptions

Compare the two revenue streams (transaction: $2.25M, ads: $5M). Identify which assumptions most affect the total: likely ad impression fee, take rate, and purchases per intentor (or AOV).

4. Design sensitivity analyses

For each dominant assumption, vary it over a plausible range (e.g., ad fee from $0.05 to $0.20, take rate from 3% to 7%) and compute the resulting revenue to show the range of outcomes.

5. Summarize and recommend next steps

Conclude with which revenue stream is larger and more sensitive, and suggest data collection or experiments to reduce uncertainty in the top assumptions.

Key Points to Mention

  • Clearly define the funnel: DAU -> weekly intent -> purchases -> GMV -> transaction revenue.
  • State and justify the ad impression fee assumption; if not provided, use a benchmark (e.g., $0.10 per DAU per week).
  • Perform sensitivity analysis on the top 3 assumptions: ad impression fee, take rate, and purchases per intentor (or AOV).
  • Consider that ad revenue might be influenced by the same user activity, so avoid double-counting or note potential correlation.
  • Use ranges (e.g., low, base, high) for each assumption to show the uncertainty in the estimate.
  • Prioritize assumptions by their impact on the total revenue and suggest A/B tests or data analysis to validate them.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

After launching the Shopping tab, you see high click-through into merchant flows but purchases are way below forecast. Build a metric tree and investigation plan that localizes funnel drop-offs, evaluates cannibalization vs. incrementality, separates UX friction from supply/quality issues, and rules out data instrumentation problems. Also specify what would trigger a rollback.

Root Cause AnalysisProduct Analytics & MetricsA/B Testing & Experimentation
Author's notes

This was my favorite part and also where I probably talked too long.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining a metric tree that decomposes the purchase funnel into stages (impression → click → merchant visit → add-to-cart → checkout → purchase) and segment by user, merchant, and surface. Then systematically investigate each stage using data validation, funnel analysis, and experimentation to isolate the root cause, while evaluating cannibalization and incrementality. Finally, set clear rollback criteria based on the severity and nature of the issue.

Pro tip: Always validate data instrumentation first—many 'funnel drops' are actually logging errors. Also, use holdout groups or geo experiments to measure incrementality, as observational data can be misleading due to selection bias.

1. Build a Metric Tree and Validate Data

Decompose the purchase funnel into stages and define key metrics (CTR, conversion rate, AOV). Immediately check data quality: ensure events are logged correctly, no missing data, and metrics are computed consistently across platforms.

2. Localize Funnel Drop-offs

Analyze conversion rates at each stage, segmented by dimensions like device, user demographics, merchant category, and entry point. Identify where the largest drop occurs compared to forecast or baseline.

3. Evaluate Cannibalization vs. Incrementality

Determine if Shopping tab purchases are incremental or cannibalizing other surfaces (e.g., main feed, search). Use holdout experiments, geo tests, or difference-in-differences to measure the net impact.

4. Separate UX Friction from Supply/Quality Issues

Investigate user experience factors (load time, navigation, checkout flow) and supply-side factors (merchant quality, product availability, pricing). Use session recordings, surveys, and merchant data to diagnose.

5. Define Rollback Triggers

Establish thresholds for rollback: e.g., if purchase conversion is >X% below forecast and root cause is not fixable quickly, or if cannibalization outweighs incrementality, or if data issues cannot be resolved.

Key Points to Mention

  • Metric tree decomposition: impressions, clicks, merchant visits, add-to-cart, checkout, purchase, with conversion rates at each stage.
  • Data instrumentation checks: event logging, deduplication, bot filtering, and cross-platform consistency.
  • Segmentation: by device, user cohort, merchant category, and traffic source to localize drop-offs.
  • Cannibalization vs. incrementality: use holdout groups, geo experiments, or switchback tests to measure causal impact.
  • UX vs. supply: analyze page load times, error rates, checkout friction, and merchant quality metrics (ratings, return rates).
  • Rollback criteria: define specific thresholds (e.g., conversion drop >20%, negative incrementality, data integrity issues) and a decision framework.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.