Data · Job description template
Data Scientist job description template
Reports explain what already happened. A Data Scientist builds models that predict what will happen next and puts them to work inside the business, such as flagging customers likely to leave or forecasting demand. They are accountable for models that run in production, beat a simple baseline and keep working over time.
First-year outcomes
What should be visibly different a year from now. Results, not tasks.
- At least one predictive model, such as churn prediction, demand forecasting or lead scoring, runs in production and is used in a live business process.
- Every production model has a documented baseline, and its measured improvement over that baseline is reported to the business owner each quarter.
- Model performance is monitored automatically, with an alert when accuracy falls below an agreed threshold.
- At least one controlled experiment, such as an A/B test, is designed, run and analysed, ending in a clear go or no-go decision.
- All model code is in version control and reproducible, so another person can retrain any model from the repository.
The work
Recurring work
- Prepare and explore data to frame modelling problems.
- Build, validate and improve machine learning models in Python.
- Monitor models in production and retrain them when needed.
- Design and analyse experiments with product and marketing teams.
Projects
- Ship the first production model with the engineering team.
- Set up model monitoring and alerting.
- Create an experiment design template and a guide to reading results.
Decisions they own
- Which modelling approach and features to use for a problem.
- Whether a model is good enough to release compared with the baseline.
- Sample size and run length for an experiment.
People they work with
- Business owners, such as the head of sales or customer success, who use model outputs.
- Software engineers, to deploy models and connect them to products and tools.
- Data analysts, to share data definitions and reuse clean tables.
- Leadership, to agree which problems are worth modelling.
Requirements
Essential on day one
- Strong Python, including pandas and scikit-learn or similar libraries.
- Solid statistics: validation, overfitting, bias and experiment design.
- Experience taking at least one model from prototype to real use.
- SQL to pull and prepare your own data.
Helpful, but can be learned
- The cloud platform and deployment pipeline the company uses.
- Specific MLOps tools, such as MLflow.
- The business domain and what a prediction is worth to each team.
How to assess it
One piece of evidence for each first-year outcome.
| Outcome | Evidence to look for |
|---|---|
| A predictive model in production and in use | Ask them to walk through a model they shipped. Good names the business process it changed and who acted on its output. |
| Documented baseline and measured lift | Take-home or live exercise: build a simple churn model on a sample data set and compare it with a naive baseline. Good reports lift honestly on held-out data. |
| Automated monitoring with alerts | Ask how they found out a model had degraded. Good describes tracking input drift and accuracy, with alerts, rather than hearing from users. |
| One experiment run to a clear decision | Give a short A/B test result with a small sample. Good checks significance and power and does not overclaim. |
| Version-controlled, reproducible models | Ask to see a code sample or repository. Good has clear structure, pinned dependencies and a documented way to retrain. |
Opening for the job ad
We want to move from reporting on the past to predicting what comes next, and put those predictions to work. In your first year you will ship our first production model, prove it beats a simple baseline and set up the monitoring and experiments that keep it honest.
Turning this into an ad? See how to write a job ad that gets applicants.
Next: interview for these outcomes
Eight structured questions for this role, each tied to an outcome above.
Data Scientist interview questionsSimilar roles
Your next hire, start to finish, in one place.
Free workspace. No card. Full scoring free for 7 days.