Meta DS Roles: What They Actually Hire For
Meta has three DS tracks and the difference matters more than at most companies:
-
Data Scientist (Product) — Works with product teams on experiments, feature evaluation, and product decisions. This is the most common DS role. You'll design A/B tests, analyze metrics, and influence product roadmap. Most DS roles at Meta are this type.
-
Research Scientist — Publishes papers, works on novel algorithms. PhD required. Works on problems like federated learning, self-supervised learning, or new recommendation architectures.
-
Applied Scientist — Builds and deploys ML models. Closer to ML engineering. Works on ranking, recommendation, ads optimization, content understanding.
The biggest difference from other companies: Meta's DS culture is experimentation-first. Every product decision goes through A/B testing. If your resume doesn't show deep experimentation skills, you signal you can't do the work.
What Meta DS Teams Actually Work On
| Team | What They Do | Resume Signal |
|---|---|---|
| Reels recommendation, creator tools, engagement optimization | Recommendation systems, engagement metrics, content understanding | |
| Feed ranking, meaningful social interactions, groups | Ranking, NLP, graph analytics | |
| User growth, messaging quality, business messaging | Growth metrics, funnel analysis, NLP | |
| Ads | Ad targeting, bid optimization, attribution | Causal inference, revenue metrics, auction theory |
| Reality Labs | AR/VR analytics, spatial computing, user behavior | Time series, sensor data, 3D understanding |
| ** Integrity** | Content moderation, hate speech detection, safety | Classification, NLP, trust & safety metrics |
Meta DS Resume Format
- Single column. Clean, minimal.
- 1 page for under 5 years. 2 pages max for senior/PhD.
- PDF format.
- Standard fonts. Helvetica, Arial, 10-12pt.
Example: Meta Data Scientist Resume (Product DS, 3 Years Experience)
Note: All names, emails, and links below are placeholders.
First Last [email protected] | linkedin.com/in/first-last | github.com/first-last
Summary
Data scientist with 3 years driving product decisions at scale through experimentation and causal inference. Designed experiment framework serving **2B+** users across Instagram and Facebook, enabling 25+ experiments per quarter. Built content personalization model increasing Reels engagement by **11%** for **500M+** DAU. Strong in Python, SQL, Spark, and experimental design. MS in Applied Mathematics from Columbia.
Technical Skills
- Languages: Python, SQL, R, Scala
- ML/Stats: scikit-learn, PyTorch, XGBoost, causal inference (DiD, synthetic controls, propensity), Bayesian methods
- Data: Presto, Hive, Spark, BigQuery, MySQL
- Experimentation: A/B testing, sequential testing, CUPED variance reduction, interference detection, power analysis
- Visualization: Looker, Matplotlib, Plotly
Experience
Data Scientist | Flipkart | Aug 2023 - Present
- Designed recommendation A/B testing framework handling 2B+ daily events with interference-aware design across connected surfaces (Instagram, Facebook, Messenger), enabling 25 experiments per quarter with 95% confidence intervals and proper power analysis
- Built content personalization model using collaborative filtering and deep learning, increasing Instagram Reels engagement by 11% for 500M+ daily active users. Shipped to production in 3 weeks (Move Fast)
- Analyzed creator monetization funnel across 50M+ creators, identifying drop-off points that led to feature changes increasing creator payouts by $15M annually and creator retention by 8%
- Developed real-time anomaly detection for Stories engagement using statistical process control, catching 5 pipeline bugs before they impacted 1B+ daily stories views
Data Science Intern | Meta (Instagram) | May 2022 - Aug 2022
- Analyzed 1B+ Facebook feed events to optimize ranking algorithm, increasing meaningful social interactions by 4% in A/B test across 50M users
- Built causal inference framework using synthetic controls for regional feature launches where A/B testing wasn't feasible, reducing experiment time from 4 weeks to 2 weeks and enabling 3x more launches
- Created automated reporting dashboard for Reels metrics in Looker, adopted by 3 product teams for weekly product reviews and experiment decisions
Why these bullets work for Meta:
- Experimentation depth: Interference-aware design, CUPED, synthetic controls — not just "ran A/B tests"
- Scale: 2B+ events, 500M+ DAU, 50M+ creators, 1B+ stories
- Product impact: Engagement lift, creator payouts, meaningful interactions
- Velocity: "Shipped in 3 weeks" signals Move Fast culture
Education
Columbia University | MS Applied Mathematics | 2021-2023 | GPA: 3.85/4.0 University of Michigan | BS Mathematics + CS | 2017-2021 | GPA: 3.9/4.0
Projects
- Experiment Analysis Toolkit — Python package for automated A/B test analysis with sequential testing, CUPED variance reduction, and interference detection. 400+ GitHub stars. Used by 5 data teams for daily experiment decisions.
- Social Network Graph Analytics — Built graph-based feature engineering pipeline using NetworkX and PyTorch Geometric for friend recommendation. Improved precision@10 by 8% and shipped to production.
Why projects matter for Meta: Meta values tools that speed up experimentation. The Experiment Analysis Toolkit shows you think about velocity — making the whole team faster, not just yourself.
Meta DS Resume Mistakes That Get You Rejected
Mistake 1: Showing correlation instead of experimentation
[object Object], "Users who used feature X had **20%** higher engagement"
Meta is an experimentation culture. Showing correlation without experimentation signals you don't understand how Meta makes product decisions. Even if the analysis was observational, frame it as a study design.
[object Object], "Designed A/B test for feature X with proper randomization and power analysis, measuring **12%** causal lift in engagement with **95%** confidence"
Mistake 2: No velocity signals
Wrong: "Developed comprehensive analytics framework over 6 months"
Meta values Move Fast. A 6-month timeline without velocity signals suggests you're slow. Show you ship quickly, even if the project was complex.
Right: "Built and shipped Reels engagement dashboard in 2 weeks, adopted by 3 product teams for weekly experiment reviews"
Mistake 3: Showing model metrics without product context
Wrong: "Improved recommendation model AUC from 0.72 to 0.81"
AUC means nothing to a product manager. Meta DS works with product teams. Your metrics need to be product metrics: engagement, retention, conversion, time spent.
[object Object], "Improved recommendation model, increasing Reels engagement by **11%** for **500M+** daily active users"
Mistake 4: No interference or network effects awareness
Wrong: "Designed A/B test for feed ranking"
Meta's products have network effects (what your friends see affects what you see). If your resume doesn't mention interference detection, network effects, or cluster randomization, you signal you don't understand Meta's unique experimentation challenges.
Right: "Designed interference-aware A/B test for feed ranking using cluster randomization across social graphs, properly accounting for network effects"
What Makes Meta DS Resumes Different
Meta DS resumes differ from Google and Amazon in 3 specific ways:
1. Experimentation is the core skill, not a nice-to-have At Google, experimentation is important. At Meta, it's the core of the job. Your resume should show you can design, analyze, and ship experiments — not just run them.
2. Velocity signals matter Meta's "Move Fast" culture means they want to see you ship quickly. Include timelines: "shipped in 3 weeks", "launched in 2 sprints", "reduced experiment cycle from 4 weeks to 1 week".
3. Product metrics over model metrics Meta is a product company. Engagement, retention, conversion, time spent — these are the metrics that matter. Model accuracy, AUC, F1 — these don't impress Meta recruiters.
Keywords Meta's ATS Scans For (DS Roles)
Must-have keywords:
- Python, SQL, Spark, Presto
- A/B testing, experiment design, causal inference
- Product metrics, engagement, conversion, retention
- Statistical modeling, machine learning
- Hive, MySQL, data warehousing
Nice-to-have keywords:
- CUPED, sequential testing, interference detection
- Synthetic controls, DiD, propensity scoring
- PyTorch, XGBoost, deep learning
- Graph analytics, network analysis
- Funnel analysis, cohort analysis, retention curves
Tailor Your Resume for Meta
Our AI tailors your resume to match Meta's specific data science requirements. Paste the job description and get a tailored version.
