Amazon DS Roles: What They Actually Hire For
Amazon has three DS tracks and they're more different than at most companies:
-
Data Scientist — Analytics, dashboards, A/B testing, business insights. Works closely with product managers and VPs. This is the most common entry point.
-
Applied Scientist — Builds and deploys ML models. Closer to ML engineering. Works on demand forecasting, recommendation, search ranking, fraud detection. Requires production ML experience.
-
Research Scientist (AWS AI Labs) — Publishes papers, works on novel algorithms. PhD required. Works on problems like time series foundation models, graph neural networks, or autonomous systems.
The biggest difference from other companies: Amazon evaluates Leadership Principles (LPs) from your resume, not just behavioral rounds. Your bullets need to demonstrate LPs implicitly through impact.
What Amazon DS Teams Actually Work On
| Team | What They Do | Resume Signal |
|---|---|---|
| Retail (Demand) | Demand forecasting, inventory optimization, pricing | Time series, supply chain, operational metrics |
| Retail (Search) | Product search ranking, query understanding, conversion | NLP, ranking, A/B testing, click-through modeling |
| AWS | Anomaly detection, capacity planning, cost optimization | Time series, monitoring, enterprise metrics |
| Advertising | Bid optimization, ad targeting, attribution | Causal inference, revenue metrics, auction theory |
| Alexa | NLU, conversation analytics, user engagement | NLP, speech, engagement metrics |
| Logistics | Route optimization, delivery prediction, warehouse ops | Operations research, optimization, real-time prediction |
Amazon DS Resume Format
- Single column. Clean, no graphics.
- 1 page for under 5 years experience. 2 pages max for senior.
- PDF format.
- Standard fonts. Arial or Calibri, 10-12pt.
- STAR bullets. Situation, Task, Action, Result. Amazon invented STAR — use it.
Example: Amazon Data Scientist Resume (Applied Science, 4 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 4 years building and deploying ML systems at Amazon scale. Led demand forecasting pipeline end-to-end, reducing stock-outs by **18%** across 500+ fulfillment centers (**$8.5M** annual impact). Built real-time feature pipeline cutting inference latency from 2 hours to 5 minutes. Strong in Python, SQL, SageMaker, and time series forecasting. MS in Computer Science from Carnegie Mellon.
Technical Skills
- Languages: Python, SQL, R, Java
- ML/Stats: SageMaker, scikit-learn, XGBoost, Prophet, DeepAR, causal inference, survival analysis
- Data: Redshift, S3, Athena, Spark, Airflow, DynamoDB, Kinesis
- Visualization: QuickSight, Tableau, Matplotlib
- Deployment: SageMaker endpoints, Lambda, Step Functions, model monitoring
Experience
Data Scientist | Amazon (Retail - Demand Forecasting) | Jul 2023 - Present
- Led demand forecasting pipeline end-to-end, owning model development, deployment, and monitoring across 500+ fulfillment centers. Reduced stock-outs by 18% and saved $8.5M annually in lost sales (Ownership + Deliver Results)
- Redesigned feature engineering pipeline from batch to real-time using SageMaker endpoints and Kinesis, cutting prediction latency from 2 hours to 5 minutes for 10M+ daily SKU predictions. Shipped ahead of Q4 peak season (Bias for Action)
- Built automated model monitoring system tracking prediction drift across 500+ markets using CloudWatch and custom metrics. Caught 3 model degradations before they impacted customer experience, preventing estimated $2M in lost sales (Customer Obsession)
- Designed A/B testing framework for new forecasting algorithms with proper power analysis and interference detection across fulfillment centers, enabling 12 experiments per quarter with 95% confidence (Invent and Simplify)
Data Science Intern | Amazon (AWS - Cost Optimization) | May 2022 - Aug 2022
- Built customer lifetime value model using survival analysis, identifying top 20% of customers driving 60% of revenue. Informed $3M annual retention budget allocation for AWS enterprise accounts (Customer Obsession)
- Analyzed 2B+ clickstream events to optimize AWS console search ranking, improving task completion rate by 9% and reducing support ticket volume by 15% (Bias for Action)
- Deployed CLV model to SageMaker serving 50M+ daily predictions with p99 latency under 100ms. Built automated retraining pipeline triggered by drift detection (Ownership)
Why these bullets work for Amazon:
- LP signals: Each bullet maps to 1-2 LPs — not labeled, but obvious to any Amazon recruiter
- Deployment focus: SageMaker, Kinesis, CloudWatch, automated retraining — not just model building
- Operational metrics: Latency, throughput, uptime — Amazon cares about operational excellence
- Scale: 500+ fulfillment centers, 10M+ predictions, 2B+ events
Education
Carnegie Mellon University | MS Computer Science (ML track) | 2021-2023 | GPA: 3.8/4.0 Indian Institute of Technology, Madras | B.Tech Data Science | 2017-2021 | GPA: 9.0/10
Projects
- Demand Forecasting Pipeline — End-to-end time series forecasting system using Prophet, DeepAR, and gradient boosting. Handles 100K+ SKUs with daily retraining and automated model selection. Open-sourced with 300+ GitHub stars.
- Real-time Feature Store — Built feature store using DynamoDB and Lambda for low-latency feature serving. Reduced inference latency from 200ms to 15ms for 10M+ daily predictions. Deployed across 3 Amazon teams.
[object Object], Amazon values tools that ship. A pipeline handling **100K+** SKUs shows you think about production, not just notebooks. The feature store shows you understand the infrastructure that makes ML work at scale.
Amazon DS Resume Mistakes That Get You Rejected
Mistake 1: Listing responsibilities instead of ownership
Wrong: "Worked on demand forecasting models"
This says you were present, not that you drove anything. Amazon evaluates ownership from your first resume scan. "Worked on" is the weakest possible opener.
[object Object], "Owned demand forecasting pipeline end-to-end, reducing stock-outs by **18%** across 500+ fulfillment centers"
"Owned" signals you drove the project, not just contributed.
Mistake 2: No deployment signals
Wrong: "Built churn prediction model using XGBoost"
Amazon is an engineering company. Models that stay in notebooks don't matter. If your resume doesn't mention SageMaker, deployment, latency, or monitoring, you signal you can't do the work.
[object Object], "Deployed churn prediction model to SageMaker serving **10M+** daily predictions with automated retraining and drift monitoring"
Mistake 3: Ignoring operational metrics
Wrong: "Improved model accuracy by 5%"
Amazon thinks in operational metrics: latency, throughput, uptime, error rates. Model accuracy without operational context doesn't tell Amazon if you can ship.
[object Object], "Improved demand forecast accuracy by 5%, reducing stock-out rate from **12%** to **10%** while maintaining p99 latency under 50ms"
Mistake 4: Not showing customer impact
Wrong: "Built recommendation system"
Amazon is customer-obsessed. Every metric should tie back to customer experience. Even backend work affects customers indirectly.
[object Object], "Built product recommendation system increasing conversion rate by 7% and reducing customer search abandonment by **12%**"
What Makes Amazon DS Resumes Different
Amazon DS resumes differ from Google and Meta in 3 specific ways:
1. Ownership language over contribution language Google and Meta value collaboration. Amazon values ownership. Use "led", "owned", "drove", "launched" — not "worked on", "contributed to", "assisted with".
2. Operational metrics alongside business metrics Google cares about engagement. Meta cares about growth. Amazon cares about operations: latency, throughput, uptime, cost per prediction. Include both business and operational metrics.
3. Deployment depth Amazon expects you to deploy models, not just build them. Show SageMaker, Lambda, Step Functions, CloudWatch, or equivalent production ML infrastructure.
Keywords Amazon's ATS Scans For (DS Roles)
Must-have keywords:
- Python, SQL, SageMaker, Redshift
- Time series forecasting, demand planning
- A/B testing, experimentation
- Deployment, production, monitoring
- S3, Athena, Spark, Airflow
Nice-to-have keywords:
- DeepAR, Prophet, XGBoost
- Causal inference, survival analysis
- Kinesis, DynamoDB, Lambda
- QuickSight, data warehousing
- Customer segmentation, LTV, churn
Tailor Your Resume for Amazon
Our AI tailors your resume to match Amazon's specific data science requirements. Paste the job description and get a tailored version.
