Applied ML Scientist

Zendesk

Lisbon, Portugal

Role Overview

As an Applied ML Scientist, you will be the person who makes that happen. You will train models, design experiments, and uncover the patterns that connect customer behavior to revenue outcomes.Then ship those insights as production systems that the business relies on daily. You own problems end-to-end: from formulating the right question, to training and validating models, to deploying them and measuring whether they actually moved the needle, in a closed feedback loop.

You will work alongside a team of AI/ML Engineers, Data Engineers, and Analysts as part of the Enterprise Data & Analytics department. Your focus is the science: understanding what drives outcomes, building models that learn from data, and turning that understanding into systems that work.

Key Responsibilities

Train, evaluate, and deploy models that predict and explain revenue-related outcomes (churn, expansion, conversion, engagement)
Design and run experiments to establish relationships between customer signals and business results
Build production ML pipelines that learn continuously: ingest new data, retrain, validate, and serve predictions at scale
Work with large volumes of both structured data (usage metrics, revenue events, account attributes) and unstructured data (support tickets, conversations, product feedback)
Use LLMs and deep learning where they're the right tool (embeddings, fine-tuning, text feature extraction)
Define and track model performance in production: monitor drift, measure business impact, and iterate when models degrade
Own your models end-to-end from prototype through production deployment, monitoring, and maintenance
Partner with product and engineering to embed intelligence where users make decisions.
Translate model outputs into actionable insights for sales, success, and product teams: make the intelligence layer useful, not just accurate
Collaborate with stakeholders to identify high-leverage questions and prioritize modeling work based on expected business impact

This Role Is For You If…

You get more satisfaction from a simple model that ships and moves a metric than a complex one that scores well on a holdout set
You've trained models on real-world messy data and dealt with the unglamorous parts: label noise, class imbalance, feature leakage, data drift
You write production Python — tests, type hints, clean abstractions — not just notebooks with df_final_v3
You think about problems in terms of "what will someone do differently because of this?" rather than "what's the most sophisticated technique?"
You care that users actually change behavior because of your work: you're not done when the model is accurate, you're done when someone acts differently
You form opinions about what to build next based on data and user understanding, not just what's assigned to you

What We Are Looking For

Education & Experience

3–5 years' experience in applied machine learning, data science, or a related field
BA/BS in Computer Science, Statistics, Mathematics, or related quantitative discipline
Advanced degrees welcome but not required

Technical Expertise

Strong foundations in statistical modeling and machine learning: regression, classification, survival analysis, causal inference, uplift modeling, or similar
Experience training models on real-world datasets: feature engineering, validation strategy, handling messy data at scale
Comfort working with unstructured data (text, conversations) using embeddings, fine-tuning, or learned representations: you understand LLMs as modeling tools, not just API endpoints
Strong Python programming skills: you write production-grade code with tests, not just scripts
Strong SQL skills and experience with cloud data warehouses (Snowflake preferred)
Experience deploying and monitoring models in production (batch or real-time)
Nice-to-have: Experience with experiment design, A/B testing, and causal inference
Nice-to-have: Experience with orchestration tools (Airflow, dbt, or similar)
Nice-to-have: Experience with AI-assisted development workflows (Claude Code, Cursor, Copilot, or similar)