Data Scientist Map From 40,000 Real Job Ads: Analytics or Machine Learning?

A dataset of forty thousand postings

Real postings beat monthly guesses. The speaker built a collection pipeline over months: the system pulls open roles daily, cleans them and reports back. The result spans nearly 40,000 real ads from more than 10,000 companies across roughly 100 countries, with 6,000-plus teams hiring data scientists. The narrative follows that ground: opening question first, then concepts, then numbers.

The demand ranking breaks cliches. Data engineering sits first, data science second, followed by data analyst, BI developer, data architect and analytics engineer. The Bureau of Labor Statistics Handbook (www.bls.gov) puts median pay at $112,590, employment at 245,900 in 2024, growth at 34% toward 2034 and about 23,400 yearly openings. GetUHired (getuhired.co) 2026 flow agrees live: 41,240 postings in 90 days, 3,108 new roles a week and a 32% remote share. The Axial Search (axialsearch.com) 2026 demand map completes the picture with state, city and seniority splits.

Three axes place every role: finding answers in analytics, preparing data in engineering, building models in machine learning and AI. The data analyst sits squarely in analytics; the data engineer stands in engineering while reaching into AI engineering; the ML engineer tops the stack. The data scientist spreads everywhere, from analysis to pipelines to modeling, which makes the title the market's blurriest label.

The blur has a price: at least 19% of postings are not real data science jobs. Some seek analysts, AI researchers, BI developers, data engineers or product managers instead. The speaker tells of four master's-holding friends who joined as data scientists and ended up doing data engineering; three stayed, one returned. The practical lesson: read the job description line by line and ask what the work really is at interview.

Two distinct jobs: analytics and machine learning

Cleaned postings split in two: 54% analytics, 46% machine learning data scientists. The first answers a company's hardest questions; the second trains models from scratch and embeds them in production apps. The recommender example sums up the second lane: build the model, ship it live, feed the feed millions see. The roadmap grows out of this split.

The analytics lane starts its day with a question: data analysis, answer-oriented modeling , experiment design and requirements translation. A churn model flagging next month's leavers is the classic case; A/B tests prove causality; business language gets translated into data. Output means dashboards, reports and recommendations; the decision table expects all three.

The machine learning lane turns models into product: building, shipping to production , upfront analysis and agent systems. The model never stays on a laptop; it serves live traffic. LLMs and agent stacks appear in nearly every posting, and 40% demand a full AI system such as a chatbot or agent. That expectation winks at the AI engineer's job description.

Degrees, code and libraries

The degree row surprises: 72% of postings mention one; PhD at 32%, master's at 25%, bachelor's at 14%. Yet only 2% require a doctorate. A bachelor's suffices for about half the roles, and a 25% slice stays open to degree-free profiles. BLS entry typically wants a bachelor's, while 40 to 50% of other roles never mention a degree at all.

Code leadership is clear: Python first, SQL second. Wherever the title carries data, SQL follows. R counts at 36% in analytics and a marginal 12% in machine learning. Posting skill lists are wish lists , not checklists; nobody matches all of them, and ordering signals priority.

Methods and libraries diverge by lane. Analytics leads with statistics, then machine learning and A/B testing; deployment stalls at 9%. Machine learning cites the method at 32% and deployment at 30% against 9% in analytics, with statistics, A/B, NLP and deep learning trailing, so no neural-network mastery is required. Libraries say PySpark and Pandas for analytics as big data moves to Databricks scale. Machine learning crowns PyTorch with TensorFlow, then PySpark, scikit-learn, Pandas and NumPy. Skillenai (skillenai.com) logs 2,161 PySpark postings in 90 days, up 8%, against 5,534 PyTorch postings, up 38% with San Francisco on top; PyTorch appears in 9.3% of data scientist ads.

Platforms and the roadmap

Cloud and visualization finish the frame. AWS leads, especially in machine learning, with Google Cloud second; Databricks runs slightly ahead of Snowflake and the gap widens. Generative AI already ranks top three among machine learning-track skills, past PyTorch mentions; it sits mid-table at 16% in analytics while agents still trail. Tableau beats Power BI on advanced visuals and big-data speed while Power BI owns standard reporting. Refonte Learning (refontelearning.com) cross-industry BI demand confirms the split. The order reads: Python plus SQL foundations, statistics, math and machine learning theory, A/B for analytics, LLMs and generative AI for machine learning, scikit-learn basics in both lanes, PySpark for analytics, PyTorch with TensorFlow for machine learning, Tableau, AWS basics and a free Databricks account. The finale is one end-to-end project : the analyst closes with a dataset and hard questions, the modeler with a recommender and a live-deployment story, shared on LinkedIn. The DataCamp (datacamp.com) 23-course, 90-hour associate track packages that order; the first chapter is free. University programs teach theory, and this map closes the gaps.