Senior Data Scientist for Copilot Evals

Microsoft•Redmond, WA

8d•$119,800 - $261,000

About The Position

CADET (Customer and Analytics Driven Evals Team) is building a customer-grounded quality system for Copilot. Its mission is to identify important customer scenarios, represent them faithfully in evaluation and learning assets, run continuous quality gates, and turn failures into reusable product and model improvements. The team combines DSAT and other product signals with in-depth customer engagements to create representative evaluation sets and connect customer-grounded quality issues with quality teams. The Senior Data Scientist will build the measurement and decision system that determines where CADET invests and whether Copilot is improving for the customers and intents that matter most. The role combines DSAT, usage, customer engagement, product feedback, evaluation, and other quality signals to create a representation- and coverage-aware view of customer quality. The scientist will define taxonomies, metrics, prioritization models, analyses, and reporting mechanisms; synthesize quality opportunities; investigate loss patterns; shape evaluation portfolios; and ensure recurring quality gates reflect real customer experiences rather than static benchmarks. Microsoft describes its mission as empowering every person and organization to achieve more. Its employees work with a growth mindset, innovate to empower others, and collaborate toward shared goals, guided by respect, integrity, and accountability to foster inclusion.

Requirements

Doctorate in Data Science, Mathematics, Statistics, Econometrics, Economics, Operations Research, Computer Science, or a related field and at least 1 year of data-science experience, such as managing structured and unstructured data, applying statistical techniques, and reporting results; or
Master’s degree in one of the listed or related fields and at least 3 years of data-science experience; or
Bachelor’s degree in one of the listed or related fields and at least 5 years of data-science experience; or equivalent experience.

Nice To Haves

Solid experience using data to shape product strategy and decisions in a complex, high-scale product or platform environment.
Expertise in SQL and at least one analytical programming language such as Python or R.
Solid foundation in statistical analysis, experimentation, sampling, segmentation, measurement, and data visualization.
Experience integrating noisy quantitative and qualitative signals into actionable prioritization or measurement frameworks.
Ability to define durable metrics and taxonomies, explain their limitations, and prevent misleading interpretation.
Demonstrated ability to communicate complex analysis clearly to technical, product, and executive audiences.
Experience with AI product quality, LLM or agent evaluation, DSAT or customer feedback analysis, experimentation, or model telemetry.
Experience designing representative datasets, coverage models, quality scorecards, or regression portfolios.
Experience with clustering, text analytics, embeddings, classification, anomaly detection, or other methods for mining unstructured feedback.
Experience connecting offline evaluation results with online product and customer outcomes.
Experience working directly with enterprise customers, researchers, product managers, and engineers.
Familiarity with responsible AI, privacy-preserving analysis, data governance, and customer-data handling.

Responsibilities

Build and operate a data-driven prioritization framework for CADET quality forums and partner teams, combining DSAT, customer impact, usage, severity, strategic importance, representation, and current evaluation coverage to guide quality investments.
Define and maintain intent, sub-intent, customer scenario, coverage, and loss-pattern taxonomies for consistent use across signal intake, triage, evaluation, and reporting.
Measure how well evaluation portfolios represent production traffic, customer segments, workflow complexity, locales, grounding paths, and material failure modes.
Identify underrepresented customers, intents, scenarios, and loss patterns, and translate gaps into evaluation and data-collection priorities.
Design quality gates for top intents and customers, including success thresholds, segmentation, run cadence, escalation criteria, and reporting.
Build recurring scorecards connecting offline evaluation movement with online measures such as DSAT, task completion, retries, abandonment, and escalation; detect meaningful quality changes and alert accountable owners when action is required.
Analyze offline-online agreement, evaluation freshness, regression coverage, grader reliability, and quality movement over time.
Develop sampling, weighting, deduplication, clustering, and trend-detection approaches for noisy customer and product signals using resource- and performance-optimized data-analysis solutions that effectively use CPU, GPU, and platform capacity.
Use causal and experimental methods where appropriate to distinguish correlation, attribution, and treatment impact, and translate findings into product, model, data, and evaluation investment decisions.
Partner to translate customer evidence into valid task distributions, datasets, metrics, and reward signals, and encode metrics, taxonomies, data-quality checks, and reporting into automated pipelines.
Use signals from customer engagements to understand workflows, business impact, expected outcomes, and gaps hidden by aggregate metrics; produce clear investment recommendations and communicate them to senior leaders.
Establish sound practices for data provenance, privacy, responsible use, reproducibility, and metric governance.