The "Data Scientist" Title Is Quietly Demoting Itself

The "data scientist" title isn't dying. It's hollowing out. It became a prestige umbrella over three different jobs (analyst, ML engineer, data engineer), and the market is now pulling that umbrella apart into its parts. Many postings labeled "data scientist" now describe analyst work, while the hard modeling moved to ML and AI engineering, where the pay is higher.

So the real question stopped being "how do I become a data scientist?" It became "which of the three jobs hiding inside that title do I actually want, and which skill survives no matter which one I pick?" Picking the label was the easy part. Picking the job under it is where careers are made or stalled.

What made "data scientist" the sexiest job in the first place?

In 2012, Harvard Business Review called it the sexiest job of the 21st century. The conditions that made that true were specific and temporary. Cloud ML tooling was immature. There was no AutoML, no managed feature stores, no LLM you could paste a messy dataset into. If a company wanted a model in production, it needed one rare person who could do statistics, write production code, wrangle infrastructure, and explain the result to an executive.

That person was scarce, so the title carried real weight. The pay followed the scarcity, not the label.

By the 2022 HBR revisit, the same authors noted the role had changed a lot, with more weight on non-technical skills like ethics and change management. That's the sound of a job maturing. The scarcity that justified the prestige was already eroding. The tools caught up. The bootcamps graduated thousands. And the "unicorn" who could do all four things turned out to be mostly a hiring fantasy.

When did the title and the skills stop matching?

Somewhere between 2016 and 2021, the title inflated faster than the skill behind it. Every SQL analyst who learned pandas added "data scientist" to their LinkedIn. Job postings asked for a unicorn: a PhD-level statistician who also ships production code, builds pipelines, and presents to the board. Almost nobody is all of those at once, which is the same reason the job description is mostly a fictional document: it lists the wish, not the role.

A Harvard Data Science Review study looked at 2,726 LinkedIn profiles and 487 job postings and found a real split underneath the shared title: 74% of data scientists rated machine learning theory as a must-have, against only 23% of analysts. That's not title noise. That's two different jobs wearing the same badge.

That gap is the useful part for anyone hiring or job-hunting. The work was already two different jobs, even while the postings kept asking for one person who could do both. The split was real in the day-to-day long before the job descriptions admitted it.

How did the 2022 contraction expose the three real jobs?

The downturn did the disaggregating that the titles wouldn't. When budgets tightened, companies stopped buying the umbrella and started buying the specific job they actually needed.

Postings for data scientists, analysts, and engineers fell sharply after the end of 2022, down about 15% by mid-2024 even as the broader market recovered. Then the data jobs snapped back. After the July 2023 floor, data science postings rebounded 130% year over year, with data analyst openings up 63% in the same window. So the contraction was largely a macro layoff cycle, not a death.

But the rebound came back wearing different clothes. The work split into three functional jobs: the analyst (business and reporting), the ML engineer (production models), and the data engineer (infrastructure). A 2025 study of 16,348 job postings describes the same shift, finding the field moving from generalist data scientist expectations toward specialized roles, each supporting a distinct stage of the data work.

What is the market actually paying for now?

Different jobs, different money. Here's the divergence the shared title used to hide.

Role	Reported median pay	What it actually does
ML Engineer	~$161,407	Models in production, serving, monitoring
Data Scientist	~$123,069	Experiments, analysis, some modeling
Analytics Engineer	Strong, often six figures	Transformation layer, trusted data models
Data Analyst	Lower, climbs with scope	Dashboards, reporting, business questions

On those same DataCamp medians, the ML engineer commands roughly a 31% premium over the data scientist, about $38,000 a year. The generic "data scientist" title sits in the middle of the pack, out-earned by the role that ships models to production. The hard modeling is where the money moved.

The healthy version of this split has a name. Analytics engineers now make up 48% of individual-contributor respondents in dbt Labs' survey, where the report says over 80% of individual contributors earn more than $100k a year. That title didn't exist a few years ago. It describes a specific set of tasks at the analyst-engineer boundary, which is what role maturity looks like: a name that maps to real work, not to ambition.

Doesn't the label keep moving anyway?

It does, and faster than you can chase it. "AI engineer" postings overtook ML engineer postings in May 2023, under a year after the term entered common use, while AI engineering jobs kept growing faster than the older ML engineer label they sprang from.

The title half-life is shrinking. Today's hot label is next year's over-saturated search term. Which is exactly why chasing the label is the wrong move. The label is the thing the market re-prices fastest. The underlying skill is what holds value through every rename.

Here's the weak-versus-strong version, same person, two paths.

Weak move: A 2019 analytics manager wants the title. They finish a Coursera ML certificate, add PyTorch to LinkedIn, and call themselves a data scientist. Six months later the market has renamed the title twice and they're competing with ten thousand people who did the identical certificate.

Strong move: That same manager deepens experiment design and causal reasoning. They learn to run A/B tests with proper power analysis and can now own the question, "did this product change actually work?" A BI tool can't answer that on its own. An ML pipeline can't either. It needs someone who knows which question is worth asking and can explain the answer to the person who'll act on it.

The first person bought a label. The second built real advantage. Compounding beats the certificate every time.

But isn't data science still growing? What's the real problem?

Take the counterargument seriously, because it's true. The role isn't disappearing. The rebound was fast and real. Senior practitioners with deep domain expertise in pharma, fintech, or insurance still clear $190k and are still described as scarce by employers. There's an honest case that some firms are even re-centralizing analytics under one scientist function after over-specialized teams underdelivered.

So what's actually compressing?

The prestige premium and the entry barrier, not the absolute job count. The glut is concentrated at the entry-level and generalist end. AI tools also bite the analyst leg of the work harder than the ML engineering leg, which means the "analyst wins" read of the split is too clean. The honest framing: total jobs keep growing, but the title no longer guarantees a coherent trajectory. You can be employed under the label and still be in a slowly demoting seat if your work isn't tied to a decision someone acts on.

Look at who got cut. In the 2022 to 2023 layoffs, data scientists were only 2.7% of Amazon's layoffs and analysts plus scientists combined were 4.3% at Meta. Neither the prestige title nor the analyst title was a shield. What protected people was whether their role sat inside a revenue-generating decision loop. The title was never the safety. The proximity to a real decision was.

What should you actually do now?

Stop optimizing for the label. Optimize for the stack the label can't protect.

1. Pick one leg and go deep. Analyst, ML engineer, or data engineer. Depth in one is worth more than a shallow claim on all three. The unicorn posting is dead, and so is the unicorn career. A "do all three" title usually just means you're an underpaid specialist quietly doing two jobs.
2. Build the skill that survives all three: causal reasoning plus stakeholder communication. Figuring out what question is worth asking, designing the test that answers it cleanly, and explaining the result to someone who can act. The WEF Future of Jobs Report 2025 puts AI and ML specialists among the fastest-growing roles to 2030 and data analysts close behind, with over 90% of employers in the top industries expecting demand for AI and big data skills to rise. The technical floor is rising for everyone. The judgment layer is the edge, and it's the same shift happening across knowledge work: when AI does the analysis, the premium moves to the conversation.
3. Tie your work to a decision loop. If your output doesn't change what someone does, it's the first thing cut. Map your current role to the decision it feeds. If you can't find one, that's the problem to fix before the next downturn finds it for you.
4. Track the work, not the word. When the title renames itself again (it will), you want to be the person whose skills transfer, not the person whose LinkedIn headline expired.

The trade-off, said plainly: going deep in one leg means closing doors on the other two for a while, and the durable skills (experiment design, causal inference, clear communication) take years to build and won't show up on a certificate. That's the cost. It's also exactly why a bootcamp can't replicate it and AutoML can't automate it. You're paying time for something that doesn't deflate.

Not sure which of the three jobs your current skills actually point to, or what to build next? Tell Praxy on WhatsApp what you do today and what you want in three years. We'll map your real stack against where the pay and the durability are going, and name the one skill worth your next six months.