How to find a data scientist.

Data Scientist is the least informative title in technology, and that is saying something. The same two words describe a product analyst running experiments at a large company, an engineer shipping recommendation models at a startup, and a PhD publishing papers at a research lab. All three are correctly titled. None of them can do each other's jobs well.

Company size inverts the meaning further: at a large technology firm data scientists often do analytics while ML engineers handle production, and at a small one the same title covers the entire pipeline. The fix is not a better string. It is establishing at intake what the person will actually produce, and only then deciding where to look.

Job titles worth searching

Grouped by what the person actually does, because searching all47 at once produces a result set you cannot triage. Decide which group you need first — that decision does more for the search than any string below.

Where data scientists actually are

This role leaves a strong published trail, and it is the most reliable way to read someone's actual depth. arXiv and Google Scholar carry author affiliations and show what a person genuinely works on. For applied modelling ability, Kaggle rank is a real signal — with the caveat that it measures modelling in isolation and says nothing about production engineering or stakeholder communication.

GitHub separates the notebook population from the shipping population, which is the distinction most job descriptions fail to test. A data scientist whose repositories touch MLflow, Airflow, dbt, or containerisation has worked beyond exploratory analysis. That single filter resolves a large share of applied-role mismatches before the first conversation.

The most underused pipeline is academic. PhDs and postdocs leaving research are technically strong and comfortable with ambiguous problems, but their CVs are written for academic audiences and consistently undersell them against industry keyword screening. They are also far less contested than candidates already working in technology.

Boolean search strings

Written to be pasted as-is. Each one is built around an intent rather than a platform, since the useful question is what you are trying to find, not which site you happen to be on.

(site:arxiv.org OR site:scholar.google.com) "{topic}" ("{city}" OR "{company}") -jobs

site:linkedin.com/in/ ("data scientist" OR "applied scientist" OR "research scientist") ("{technique}" OR "PhD") "{city}"

site:kaggle.com ("competitions" OR "notebooks") ("master" OR "grandmaster" OR "expert") "{technique}"

site:github.com ("mlflow" OR "airflow" OR "dbt" OR "docker") ("data scientist" OR "machine learning") -tutorial -awesome

("A/B testing" OR "causal inference" OR "experimentation" OR "difference-in-differences") ("data scientist" OR "product analyst") -jobs -course

("speaker" OR "talk" OR "workshop") ("PyData" OR "NeurIPS" OR "ODSC" OR "Strata" OR "useR") 2024..2026 -jobs

(site:towardsdatascience.com OR site:medium.com OR site:substack.com) ("how we" OR "lessons" OR "at scale") "{technique}" -course -tutorial

("biostatistician" OR "clinical data scientist") ("SAS" OR "CDISC" OR "clinical trial" OR "FDA submission") -jobs

("PhD" OR "postdoc" OR "doctoral") ("machine learning" OR "statistics" OR "computational") ("industry" OR "transitioning" OR "leaving academia") -jobs

Mistakes that cost the most time

1. Not establishing which of the three jobs it is

   Data Scientist covers product analytics, applied machine learning, and research science. These need different people, and a candidate excellent at one may be weak at another. The single most useful question at intake is what the person will produce — an experiment readout, a deployed model, or a paper. Everything else follows from the answer.

2. Reading the title without reading the company

   Company size inverts the meaning. At a large technology company a Data Scientist often does product analytics while ML work sits with dedicated ML engineers; at a small company the same title covers everything from pipeline to production model. A candidate's title tells you almost nothing until you know where they held it.

3. Requiring a PhD by default

   A PhD is genuinely necessary for research science roles and largely irrelevant for product analytics and most applied ML work. Requiring one by habit removes a large share of the strongest applied practitioners while attracting researchers who will find the role unsatisfying. Ask whether the work involves novel methods; if not, drop it.

4. Screening on tool lists instead of problems solved

   Tool requirements in data science job descriptions are frequently arbitrary — a strong practitioner moves between pandas, R, and Spark without difficulty. Screening on named libraries filters for CV keyword matching rather than ability. What does discriminate is whether the person has solved a comparable class of problem.

5. Confusing data science with data engineering

   A great many advertised data science roles are actually data engineering roles: the organisation has no reliable pipelines, and the first year will be spent building them. Data scientists hired into these roles leave. If the data infrastructure does not exist yet, the honest advert is for a data engineer.

6. Ignoring the academic pipeline

   PhDs and postdocs leaving academia are technically strong, accustomed to ambiguous problems, and dramatically less recruited than candidates already in industry. Their CVs are written for academic audiences and undersell them against industry norms, which means keyword-based screening rejects them systematically.

Common questions

What job titles should I search for when hiring a data scientist?

The title itself is unreliable, so search by what the person actually does. For experimentation and product work, look for Product Data Scientist, Decision Scientist, Quantitative Analyst, and Experimentation Scientist. For model building, search Machine Learning Scientist, Applied Scientist, and specialisation terms like NLP Scientist or Computer Vision Scientist. For research roles, Research Scientist and Member of Technical Staff are common. In regulated domains the domain term is more discriminating than the data term — Biostatistician, Quantitative Researcher, or Clinical Data Scientist.

What is the difference between a data scientist, data engineer, and ML engineer?

A data engineer builds and maintains the pipelines and storage that make data usable. A data scientist analyses that data to answer questions or build models. A machine learning engineer takes models and makes them run reliably in production. The distinction matters because many advertised data science roles are actually data engineering roles in disguise — if an organisation has no reliable pipelines, whoever is hired will spend the first year building them, and a data scientist hired into that situation will leave.

Does a data scientist need a PhD?

For genuine research science roles, usually yes — the work involves novel methods and the ability to read and extend current literature. For product analytics and most applied machine learning, no, and requiring one by default is actively counterproductive. It removes many of the strongest applied practitioners, who came through engineering or analytics routes, while attracting researchers who will find the role unsatisfying and leave. The useful test is whether the work requires developing new methods or applying established ones well.

Where can I find data scientists outside LinkedIn?

The published work trail is strong for this role. arXiv and Google Scholar carry author affiliations and show precisely what someone works on, which matters for research and applied science roles. Kaggle rank is a real signal of modelling ability, though it says nothing about production engineering. GitHub reveals whether someone has worked beyond notebooks — repositories touching MLflow, Airflow, or dbt indicate production experience. Conference speaker lists from PyData, NeurIPS, ODSC, and similar events identify senior practitioners who can explain their work.

Why do data science hires fail so often?

Usually because of a mismatch established before the candidate was ever sourced. The most common pattern is hiring a modeller into an organisation with no usable data infrastructure, where the actual first-year work is data engineering. The second is title mismatch: hiring someone whose experience is product analytics into a role that expects production model deployment, or the reverse. Both are intake failures rather than sourcing failures, which is why establishing what the person will produce — an experiment readout, a deployed model, or a paper — is the highest-value part of the process.