Choose a machine learning engineer when the main need is building and operating an ML capability in a product. Choose a data scientist when the main need centers on analytical questions, experimental reasoning, or developing and evaluating models.
These titles are not standardized. Describe the next outcome and the responsibilities it requires.
Map the lifecycle
| Work | Ownership question |
|---|---|
| Problem formulation | Who decides whether ML addresses the user or business need? |
| Data and labels | Who defines, prepares, and checks the inputs? |
| Experiments | Who compares approaches and interprets results? |
| Integration | Who connects the model to the application? |
| Operation | Who monitors behavior, updates, and failures? |
| Decision support | Who explains findings and limitations to stakeholders? |
A candidate may cover several areas. Identify the essential depth rather than assuming every ML-related title implies all of them.
Find the next missing outcome
If the team has a promising model but cannot operate it reliably, prioritize production engineering experience.
If the team cannot formulate a useful target, assess whether the problem is suitable for modeling, or interpret experiments, prioritize the relevant analytical and scientific judgment.
If both are missing, consider the support needed for a broad hire. Do not ask one person to resolve every uncertainty without clear priorities.
Include data ownership
ML systems can fail quietly when data becomes stale or feature behavior changes. Google’s ML guidance highlights freshness, silent failures, and documented feature ownership. Google Rules of Machine Learning.
For hiring, ask who the candidate believes should notice and act when inputs change. A model-quality discussion should connect to the system that produces and consumes those inputs.
Assess the intended responsibility
Use an experiment critique for model-development judgment, a fictional integration scenario for production judgment, and a data investigation for input-quality reasoning.
Ask candidates to distinguish what an offline result establishes from what remains untested in the product.
Make boundaries visible
State who owns data engineering, application code, infrastructure, and specialist research. If the role evolves, define what would trigger a change in emphasis.
Use AI infrastructure versus application engineering and the data quality interview. Discuss an ML hiring mandate with Refery.