Data Scientist Job Description Template 2026
Last updated: April 26, 2026
A 2026 data scientist earns $130,000 to $185,000 base for mid-level and $180,000 to $260,000 for senior roles in the U.S., with total comp at top tech companies running $300,000 plus once equity is layered in. The job description below is built around the four data scientist profiles that are now genuinely separate hires, and the salary table sources five independent benchmarks so you can pressure-test the band you’re posting.
Robert Ardell, co-founder at KORE1. The data scientist title is the most overloaded job description in tech right now, and that is not a clever line. It is the reason most of the data scientist searches that come to us were already running for ninety days somewhere else. Companies post for one role and describe four. The candidate pool fragments. Nobody fits on paper, and the strongest applicants quietly close the tab.
The framework below is what we use at intake when a hiring manager calls and asks for a data scientist. The first question is never “what’s the comp band.” It is “which of these four jobs do you actually need.” Get that right and the rest of the search is mechanical. Get it wrong and the rest of the search is fourteen weeks.
One disclosure. KORE1 places data scientists through our data scientist and data engineer staffing practice, and we charge a fee when you hire through us. The template and the playbook work the same whether you call us or not.
What Is a Data Scientist?
A data scientist turns data into decisions. The job is statistical analysis, predictive modeling, experimentation, and increasingly the design and evaluation of machine learning and AI systems that drive product or business outcomes. Building dashboards is not the job. Running queries is not the job. The job is producing the model, the experiment, or the analysis that changes what the company does next.
The thing most JDs miss is that the title now covers four roles that used to be one. They share statistical fundamentals and Python fluency. After that, the work diverges sharply enough that hiring across the wrong profile is how companies end up paying $170,000 for a senior who cannot do half of what the team needed.
The four:
- The analytics-leaning data scientist. SQL-heavy. Lives in dbt, Snowflake or BigQuery, and a BI tool. Owns measurement frameworks, cohort analysis, retention modeling, and the experimentation platform when there is one. Closer to a senior analytics engineer with a statistics background than to a researcher. The biggest pool. The fastest to hire.
- The classical / experimental data scientist. Causal inference, A/B test design at scale, observational study methods, propensity scoring. Strong roots in statistics or econometrics. Lives at companies that actually run hundreds of experiments a quarter and have to defend results in front of a product leadership team that pushes back on confidence intervals. Smaller pool. Slower to hire. Worth it for the right team.
- The ML-forward data scientist. Builds and ships production models. Fluent with feature stores, model registries, MLflow or SageMaker, the basics of MLOps. Owns model performance after deployment, not just the offline accuracy number. Adjacent to a machine learning engineer, but the JD usually wants someone who can do the modeling work end-to-end rather than partner with an ML engineer to ship it.
- The applied / GenAI data scientist. The newest profile and the one most JDs are now confused about. Evaluates LLMs and agents, designs RAG pipelines, runs eval suites, and quantifies hallucination rates. Comfortable saying when a use case does not need a model and a prompt is enough, and when the prompt is the model and the eval harness is the entire job. Smallest pool. Highest comp variance. Most in demand.
Most JDs blur all four. The opening paragraph reads like the analytics-leaning role. The “what you’ll do” bullets read like the ML-forward role. The “preferred” section adds “experience with LLMs and generative AI” as a single line. The candidate who fits all four does not exist for less than $300,000.
Pick the profile. Write the JD for that person.
Data Scientist vs. Adjacent Roles in 2026
The lines next to a data scientist are blurrier than they used to be. Worth getting straight before the JD goes live.
| Role | What They Actually Own | Common Mis-hire Pattern |
|---|---|---|
| Data analyst | Reporting, dashboards, ad-hoc analysis. Descriptive, not predictive. | Hired as a “junior data scientist” and given modeling work they were not trained for. |
| Analytics engineer | dbt models, the semantic layer, data quality, metric definitions. | Confused with the analytics-leaning data scientist. The analytics engineer ships clean tables; the data scientist consumes them. |
| Data engineer | Pipelines, infrastructure, warehouse architecture, streaming. | Asked to also do modeling. Different job. See data engineer interview questions if that is the actual hire. |
| ML engineer | Production model serving, training infrastructure, latency, scale. | Posted as “senior data scientist” when the work is 80 percent systems engineering. Wrong pool entirely. |
| Applied / AI engineer | LLM application development, agents, prompt and eval pipelines. | The fastest-growing source of mis-titled JDs. See AI/ML engineer staffing. |
If two of these descriptions feel like the role you are trying to fill, the JD is going to fail. Decide which one before you post.
Data Scientist Job Description Template
This template is structured for a mid to senior generalist data scientist with a lean toward ML-forward work. Adjust the responsibilities and tooling to match the profile you actually need. If the role is analytics-leaning, drop the model deployment language. If it is GenAI applied, swap the modeling bullets for LLM evaluation and RAG specifics.
Job Title: Data Scientist
Location: [City, State / Remote / Hybrid]
Employment Type: [Full-time / Contract / Contract-to-Hire]
Department: Data Science / Machine Learning / Analytics
Reports To: Director of Data Science / Head of Analytics / VP of Data
About the Role
We are hiring a Data Scientist to partner with product and engineering teams on the modeling, experimentation, and analysis work that shapes how the company invests in its product. You will own problems end-to-end, from framing the question to shipping the model or analysis that changes a decision. The work spans predictive modeling, experimentation, and the evaluation of ML and AI systems we put in front of customers.
What You’ll Do
- Frame ambiguous business questions as quantifiable problems, then design the analysis or model that actually answers the question being asked
- Build, validate, and deploy machine learning models in partnership with ML engineering, including feature engineering, training, and post-deployment monitoring
- Design and run controlled experiments, including A/B and multivariate tests, with attention to statistical power, novelty effects, and the parts of an experiment that real product teams ignore
- Communicate results to product, design, and executive stakeholders in language that drives a decision rather than a dashboard refresh
- Evaluate generative AI and LLM-based features through structured eval suites, including hallucination measurement, faithfulness scoring, and human-in-the-loop quality review
- Partner with data engineering on the inputs your work depends on, including feature pipelines, data quality monitoring, and the parts of the warehouse you actually touch
- Mentor junior data scientists and analysts on statistical rigor, model selection, and the difference between “the model works” and “the model is correct”
- Maintain the documentation, lineage, and model cards that let someone else pick the work up six months later without starting from scratch
What We’re Looking For
- 4 or more years of professional data science experience, with at least 2 years shipping models or analyses that changed a real business decision
- Strong Python and SQL. PySpark or a comparable distributed framework if the data volume warrants it
- Working command of the modeling toolkit: regression, tree-based methods, gradient boosting, and at least passing fluency with deep learning frameworks (PyTorch or TensorFlow)
- Production experience with at least one cloud ML platform (SageMaker, Vertex AI, Azure ML, or Databricks)
- Real exposure to experimentation: power calculations, sequential testing, the failure modes of A/B testing in product environments where users are not independent
- Comfortable communicating uncertainty to non-technical stakeholders without burying it or overselling it
- Practical familiarity with at least one LLM application pattern, whether RAG, evaluation, or fine-tuning, in a context that was not a tutorial
Preferred
- Master’s or PhD in statistics, computer science, applied mathematics, economics, or a related quantitative field. Strong industry experience can substitute. We do not gate the role on the degree alone.
- Experience with feature stores (Tecton, Feast, or built in-house) and model registries (MLflow, SageMaker Model Registry)
- Causal inference background, particularly observational methods like propensity scoring, difference-in-differences, or instrumental variables
- Familiarity with model monitoring tools (Arize, WhyLabs, Fiddler) or equivalent in-house observability for production models
- Experience working in a regulated environment (healthcare, finance, insurance) where model explainability and fairness are not optional
Compensation
$140,000 to $190,000 base, plus equity and bonus. [Adjust for your market, seniority target, and total comp model. See salary breakdown below.]
Core Responsibilities in Depth
Bullet points are the cover letter. Here is what the work actually looks like, because the interview process surfaces the gap fast.
Problem framing is the part most candidates skip past in interviews and the part hiring managers most consistently undervalue in the JD. The strong data scientists do not start with the model. They start by reframing the question. A product team comes in and asks for a churn prediction model. The strong scientist asks who the prediction is for, what action it triggers, what the cost of a false positive is in this specific business, and whether the model needs to be live or whether a quarterly cohort analysis would actually move the same lever for one tenth of the engineering cost. The weak scientists open a notebook. The interview question is not “have you built a churn model.” It is “tell me about a problem your stakeholder asked you to solve where you ended up solving something different, and why.”
Modeling work has shifted in 2026. Five years ago a senior data scientist spent significant time on feature engineering and tedious model selection between handfuls of algorithms that all performed within two percentage points of each other. Today, gradient-boosted trees on tabular data still win most enterprise problems, and the real time has moved upstream into data quality work and downstream into evaluation, monitoring, and drift detection. The model itself is rarely the bottleneck. The candidate who can describe a model that performed well in offline evaluation and then degraded in production, what they noticed first, and what they fixed, is the candidate who has actually shipped something. The one who walks through hyperparameter tuning has not.
Experimentation is where the analytics-leaning and classical profiles overlap, and where the JD usually says less than it should. Production experimentation is not a textbook A/B test, and the gap between what graduate-level coursework teaches and what a real product experimentation platform actually requires is wide enough to swallow a junior hire’s first year. Users are not independent. Network effects break SUTVA. Cohort overlap creates leakage. Sequential testing, novelty effects, the cold-start problem on a new feature, and the question of how long to run an experiment before stopping it all matter, and the candidates who have actually owned an experimentation platform for a year can talk through these tradeoffs without much prompting from the interviewer. The candidates who learned A/B testing from a course will say “we set significance to 0.05 and ran it for two weeks.” Different signal.
LLM evaluation is the responsibility growing fastest in the 2026 JD and the one most postings handle worst. “Familiarity with LLMs” tells you nothing about whether the candidate has ever shipped a real eval harness or whether they have only read a few papers from last year. The work is structured: defining the eval set, picking the right metrics for the use case (faithfulness for RAG, instruction-following for agentic tasks, calibration for routing decisions inside a model gateway), running human evaluation at the cadence the use case actually warrants, and quantifying hallucination rates in a way that is reproducible across model swaps and across the inevitable upgrade cycle when the underlying foundation model is replaced. Ask candidates how they evaluated the last LLM-based feature they shipped. The strong ones describe a specific eval harness, a specific failure mode they caught before launch, and a specific tradeoff they made between latency and quality. The weak ones describe a chatbot demo.
Stakeholder communication is the responsibility that gets the most lip service and the least interview pressure. Most data scientist JDs include “strong communication skills” and then never test it. In practice, the data scientists who succeed long-term are the ones who can stand in a room with a CEO who does not believe the model, defend the methodology under genuine scrutiny rather than retreating into jargon, and either change the CEO’s mind or learn that the model itself is wrong, both of which are valuable outcomes for the company even if the second one is uncomfortable for the data scientist that day. The interview question is not “are you a strong communicator.” It is “tell me about the last time a senior leader pushed back on your analysis, what they said, and what you did.” Generic answers signal a problem.
Data Scientist Salary in 2026
Salary data on this role is messier than for cloud or DevOps engineering. The four profiles do not pay the same, the geographic spread is wide, and the gap between non-tech employer comp and big-tech total compensation is enormous. Five sources, with what each one is actually measuring.
| Source | Metric | Base / Range | Notes |
|---|---|---|---|
| BLS, May 2024 | Median, all U.S. | $112,590 | Base only. Captures the broadest population, including non-tech employers, which pulls the median below tech-heavy aggregators. |
| Glassdoor, April 2026 | Average, U.S. | $155,117 base, range $122,575–$198,897 | Self-reported. Skews toward larger employers and tech metros. Sample size near 57,000. |
| ZipRecruiter, March 2026 | National average | $130,000–$140,000 | Job posting data. Reflects what employers are actually offering, which trends below self-reported figures. |
| Built In, 2026 | Average, U.S. | $135,000–$160,000 base | Funded tech companies actively hiring through Built In. Skews toward Series B and later, where base ranges run higher. |
| Levels.fyi, 2026 | Median, tech companies | $176,276 median total comp | Late-stage and big tech, self-reported. At Google or Meta, total comp at senior levels routinely clears $315,000. |
The variance between BLS and Levels.fyi is not noise. It is a real signal about which slice of the market you are hiring from. A regional healthcare company hiring its first data scientist is benchmarking against the BLS median plus a metro adjustment. A growth-stage SaaS company is benchmarking against Glassdoor or Built In. A FAANG-adjacent AI company is benchmarking against Levels.fyi and pricing a senior data scientist at $250,000 to $400,000 in total comp because that is what the candidate has on the table from three other companies that all moved fast on the offer and circled back the following week to ask whether the candidate had decided yet. Pulling one number off one source and posting it as your range is how you end up off-market by $40,000 or more in either direction.
By profile, the rough rank order in 2026:
- Applied / GenAI data scientists are the highest-paid right now. Senior comp routinely lands $40,000 to $80,000 above an equivalent ML-forward data scientist at the same company.
- ML-forward data scientists track close to ML engineers. The line between them blurs further every year, and the comp follows.
- Classical / experimental data scientists pay similarly to ML-forward at top companies, with a wider spread at smaller employers where the role is less well understood.
- Analytics-leaning data scientists run 10 to 20 percent below the other three at most companies. Strong candidates often retitle to senior analytics engineer for the same money and clearer scope.
What Most Data Scientist JDs Get Wrong
A handful of patterns show up in nearly every search that lands on our desk after running ninety days somewhere else. None of them are subtle. All of them are fixable in an afternoon if someone notices.
Posting one job and describing four. The opening paragraph reads like an analytics-leaning role, the bullets read like ML-forward, the preferred section drops in “experience with LLMs and generative AI” as a single tacked-on line, and the comp band sits at a $145,000 generalist mid-level number that would not move the needle for any of the candidates who could genuinely do all four. The candidate who fits all four is at a hyperscaler making twice that. Pick one profile. Write the JD for that person. Mention the other capabilities as plus-factors at most.
The Master’s-or-PhD reflex. Half of the data scientist postings we see require an advanced degree by default, almost always copy-pasted from a JD template that was last revised somewhere around 2018 when the academic credential was a more reliable signal than it is today. In 2026, that requirement excludes a meaningful share of the strongest mid-career candidates, who came up through software engineering or analytics teams and built their statistical chops on the job under the supervision of senior data scientists who themselves did not necessarily have a doctorate. For an applied or ML-forward role, a strong portfolio and four to six years of production experience often beats an academic credential on day-one productivity. State the degree as preferred, not required, unless the work genuinely needs it.
“Familiarity with LLMs and generative AI.” This sentence has appeared in roughly every data scientist JD posted since mid-2024 and tells a candidate exactly nothing about whether the role is real GenAI work or a checkbox. Replace it with what you actually want. Have they shipped a RAG system in production? Do they know how to build an eval set? Have they measured hallucination rates and reported them to a product team? The specifics signal a real job. The buzzwords signal a posting that was not read closely before it went live.
Conflating data scientist with ML engineer. A real fintech intake call last year, on a search that had been open at the company for almost three months without a single offer extended. The hiring manager asked for a senior data scientist with strong PyTorch skills, low-latency model serving experience in a regulated environment, Kubernetes familiarity at the operator-CRD level, and a track record building feature stores against a real-time fraud detection use case. We pointed out, gently, that the description was an ML engineer with a data scientist title, that the candidate pool he was sourcing from did not have those skills at any meaningful rate, and that the comp band of $165,000 was about $40,000 below where that hire actually closed. The role was retitled, repriced, and filled in five weeks. Worth saying upfront so the next hiring manager reading this can save the same nine weeks we saw burned in the version where someone did not.
Asking for ten plus years of data science experience. The discipline has only existed under the title “data scientist” since around 2012, and the first wave of hires did not become senior until about 2017. Asking for ten to fifteen years of data science specifically narrows the candidate pool to a handful of people who are currently very expensive and not reading the job posting. If the role genuinely requires that level of seniority, fine. Then post for a principal data scientist or staff data scientist with the comp to match. Otherwise, “five plus years of data science experience or equivalent quantitative work” is the cleaner signal.
Common Questions
So What Does a Data Scientist Actually Do, Versus a Data Analyst or ML Engineer?
A data scientist builds models and runs experiments to predict or change what happens next. A data analyst describes what already happened. An ML engineer ships and operates the model in production. The roles share Python and SQL fundamentals; after that the work diverges. If the JD reads like more than one of these jobs, it is going to attract candidates who fit none of them well.
Realistically, What Should a Data Scientist Make in 2026?
Mid-level base lands $130,000 to $185,000 in most U.S. metros. Senior runs $180,000 to $260,000. Big-tech total comp at senior levels routinely crosses $300,000 once equity is included, per Levels.fyi self-reported data. The variance is real and reflects which slice of the market the role is in. A regional non-tech employer benchmarks against BLS plus a metro adjustment; a hyperscaler benchmarks against Levels.fyi.
Do You Actually Need a PhD to Hire a Strong Data Scientist?
For most production-facing roles, no. Roughly half of the strong data scientists we have placed in the last two years did not have a PhD. The exceptions are research-leaning teams, regulated environments where credentialed expertise is part of the model’s defensibility, and a handful of FAANG research orgs that have made the doctorate a hard filter. Outside those, gating on the degree narrows the pool without improving the hire.
What’s the Single Biggest JD Mistake You See on Data Scientist Searches?
Writing one posting that describes four different jobs. The opening reads analytics. The bullets read ML. The “preferred” section drops in GenAI as a one-liner. The comp band assumes a generalist mid-level. The candidate who can do all four does not exist at that price. Pick the profile. Write the JD for that person specifically, and let the others be plus-factors.
How Long Should a Data Scientist Search Realistically Take?
Five to ten weeks for an analytics-leaning or ML-forward role with a clear scope and a market-rate comp band. Twelve weeks plus for senior GenAI specialists, where the candidate pool is small and the offers competing are aggressive. KORE1’s average time-to-hire across IT roles is 17 days for clean contract and direct hire searches, though data scientist roles trend longer than the IT average given the interview rounds and the case study or take-home most companies still run.
Should the Role Be Contract, Contract-to-Hire, or Direct?
Direct hire is the default for data scientists, since most of the work is embedded in product roadmaps and benefits from continuity. Contract makes sense for project work, model migrations, or when the team needs senior expertise on a fixed deliverable. Contract-to-hire is underused on this role and worth considering when the hiring manager wants to evaluate a senior candidate’s actual fit before the equity vest starts. See contract staffing if that is the model you are weighing.
What Should the JD Say About Tools and Frameworks?
Be specific about what your stack actually is. “Python, SQL, and one cloud ML platform (SageMaker, Vertex AI, Databricks, or Azure ML)” tells a candidate the truth. “Experience with various ML frameworks and tools” tells them you have not decided yet, which is a different signal than you intended. List the things the team really uses, mark the rest as preferred, and let the candidate read between the lines accurately.
Is the Data Scientist Role Going to Be Replaced by AI Tools?
No, but the work is shifting. The parts of the job that involve writing boilerplate Python or generating a starter notebook are increasingly handled by coding assistants, which is fine. The parts that involve framing a problem, choosing the right method, defending a result against pushback, and evaluating an LLM-based feature are all growing. The data scientists who are pulling away in 2026 spend more time on the second list and noticeably less on the first, and the JDs that screen for the second list specifically are the ones closing strong senior candidates inside the standard search window rather than running open for fourteen weeks.
If you are working through a data scientist hire and the JD is starting to feel like it covers three different jobs, that is the call to have before the posting goes live. Reach out to our team or browse our broader data and analytics staffing work. KORE1 places data and analytics talent across 30+ U.S. metros and has enough searches behind us to have specific opinions on which JDs close in five weeks and which ones run fourteen.