Data scientist

A data scientist specialises in statistical analysis, programming and data modelling. They turn a business question into an analysis or model to inform a decision. Their work connects users’ needs, the quality of the available data and the interpretation of results.

A request may come from management or a department seeking to understand a phenomenon, improve quality or anticipate equipment failures. The data scientist starts by clarifying that request. They then identify the relevant data, prepare it and choose an appropriate approach. Building an algorithm is therefore only part of their work: understanding which decisions the result should inform remains essential.

Their scope varies from one organisation to another. They may work from data collection through to presenting findings, or focus on analysis within a multidisciplinary team. They work with business teams and technical specialists to identify sources, explain data collection requirements and prepare for the use of models.

Data scientist and data analyst: activities that can overlap

• A data scientist prepares data and develops models, including statistical and machine learning models, to address the problems assigned to them.
• A data analyst produces analyses, visualisations and dashboards that support users’ decisions.

These activities can overlap: the job title alone does not describe the work assigned to each role.

The first challenge is to connect technical work to a business decision. An analysis can be carried out correctly without answering the question asked. Recruitment must therefore take account of the ability to understand users, reframe their request and explain what the data actually allows them to conclude.

Data quality is another decisive factor. Incompatible formats, duplicates or unusable information require preparation work. Reducing the role to model creation risks leaving this responsibility poorly defined. Before hiring, identify who knows the sources and who can help explain their contents.

The choice of indicators also matters. To assess a model, you need to specify what its performance measure captures and what the result means for the intended use. A score in isolation does not explain the value of a solution. Ask for results to be accompanied by the limitations needed to interpret them and the conditions under which they can inform a decision.

Fictional example: an industrial team wants to use its data to anticipate equipment failures. Before requesting a model, it clarifies what information is available and which decisions the analysis should inform. The data scientist examines its quality, tests an approach and presents the results alongside their limitations. Management can then discuss the intended use, without confusing a technical trial with a solution that is already operational.

Finally, the level of autonomy must match the support available. A hiring decision becomes difficult to assess if the business expects one person to cover the entire project alone without defining the work they will actually be responsible for.

• Turn a business request into an analytical question and specify the expected results.
• Identify and extract the data relevant to the problem being studied.
• Clean and structure data, examining formats, duplicates and unusable information.
• Create useful features and training data for models.
• Create, test and adjust statistical or machine learning models.
• Evaluate results using indicators suited to the intended use.
• Present analyses to business teams through reports, visualisations or presentations.
• Contribute to making models production-ready and to checks before deployment, according to the allocation of responsibilities.

The education routes cited by Apec and Onisep include French master’s degrees in data science, statistics or applied mathematics, as well as diplômes d’ingénieur combining statistics and computing. Their relevance to recruitment lies in the foundations they help develop: reasoning with data, programming data processing tasks and understanding modelling methods.

The combination of statistics and computing remains essential: a mathematics-focused education can be built on through programming projects, while a computing education must also develop statistical reasoning and the ability to choose methods. The degree title alone does not justify assuming a gap in either area.

Nor is a qualification enough to establish someone’s level of autonomy. Education is not the same as full professional experience.

GetPro’s common assessment approach

GetPro structures assessment around a prioritised set of criteria. It distinguishes what can be verified from a candidate’s career history from what needs to be explored in an interview, and assigns an assessment method to each criterion. Interviews explore key skills through open questions and concrete examples. Reference checks help corroborate the responsibilities held and shed light on points requiring attention that arose during discussions.

Adapting this approach to hiring a data scientist

The suggestions below are advice for your recruitment process. The choice of technical criteria and exercise depends on the work assigned to the data scientist and the level expected.

1. Define the criteria before the interviews

Specify the work the person will carry out and the support available to them. Distinguish business problem definition, data preparation, modelling, programming and presenting findings. For each criterion, define the expected level of autonomy.

Choose the tools to assess according to your projects. Do not require a neural network library if the planned work does not justify it. Prepare a shared assessment grid so assessors can compare answers against the same criteria.

If your business has no data science expertise, involve a professional who can examine statistical reasoning and code. Ask business stakeholders to assess the candidate’s understanding of the need and presentation of findings.

2. Examine a past project

Ask the candidate to present a project, starting with the initial question. Have them specify the data used, the preparation required and the choices they personally made. Ask about the adjustments made after the initial results.

Look for an explanation that connects technical choices to the problem addressed. An answer that goes into detail about the tool but is vague about the request or the data calls for further questions. Distinguish the project’s results from the candidate’s own contribution.

To assess their rigour, ask how they identified unusable data and checked the transformations. To explore their curiosity, ask them to explain a hypothesis they chose to re-examine.

3. Use a case close to the expected work

Set a limited exercise that allows you to observe their approach, then discuss the trade-offs. Provide the necessary information about the intended use and adapt the difficulty to the level sought.

Fictional example: provide a simplified industrial dataset and a question about equipment failures. Ask which checks to perform before modelling, which features to examine and how to assess the results.

Observe whether the candidate identifies questions that need clarification before starting to process the data. Ask them to justify their choice of an indicator and explain its limitations. A confident conclusion without discussion of data quality deserves further exploration.

If the role includes development, also examine a script and its tests with the technical assessor. Ask how they would check that the script processes data correctly before others use the script.

4. Assess collaboration and communication

Ask for a short presentation of findings to a business stakeholder. Examine whether the candidate explains what the analysis contributes, what it does not allow them to conclude and what additional information would be useful.

For a management role, explore how they support less experienced colleagues and organise reviews of work. Ask for an example of a technical disagreement and the information used to reach a decision.

5. Corroborate responsibilities and reach a conclusion

During reference checks, ask for details of the responsibilities actually held, the level of autonomy and collaboration with others. Connect the feedback to the points explored during interviews, without expecting a reference to replace technical assessment.

Finally, compare the observations with the initial assessment grid. Distinguish a skill that is essential from the start from something that can be learnt with the planned support. Make the remaining uncertainties explicit before deciding.