Learn Python, SQL and statistics first, then pick a lane. Employers hire into six distinct data jobs that differ by $83K in median pay and sixfold in how often they ask for generative-AI skills, so the right plan depends on which one you want. What separates a junior hire from a senior one is leading others and shipping systems, and data skills pay off well beyond data titles: finance and operations together post more roles than the entire data/AI market.
- R1Python everywhere, SQL to production standard
- R2Causal inference and experimental design in your core
- R3A shared core, then commit to a lane
- R4Data literacy even outside data roles
- R5Ship something operational for a real stakeholder
- R6Look outside tech first
- R7AI everywhere; depth in one lane
- R8Quant trading and semiconductors if you’re strongly quantitative
Six findings that should shape your plan
Each is a claim with the number behind it. Select a card’s label to jump to its evidence.
Data science is six specializations with six different doors.
Analytics & BI and ML Engineering are the two largest specializations, at 22.4% and 22.3% of data/AI hiring, and sit at opposite ends of pay: $129,750 median against $213,000. Aiming at one takes a different plan than aiming at another.
Data skills already left the data team.
27.3% of Finance roles want SQL and 22.2% want data visualization. Marketing asks for experimentation as often as AI/ML roles do (7.7% against 7.4%). Most of the people who need these skills are headed for roles without a data title.
The steepest climb on the ladder is leading others.
Among individual contributors, asks for leadership skills rise from 17.5% at entry to 47.6% at staff level, a 30.1-point climb no technical skill matches (the largest, LLMs, is +17.4). Entry postings ask for more Python (56.0% against 51.3%). Managers are counted separately.
Building systems pays. Reporting on them doesn't.
Pay against comparable postings at the same employer, function and level. Python, in half of all postings, adds +0.1% once the specialty is held fixed, but +2.6% paired with ML skills and −5.4% paired with analytics skills; at entry level it reads +5.3% on 141 employers.
AI fluency climbs with seniority.
33.7% of entry-level data roles ask you to build with AI. At senior it's 41.6%. At staff and above, 57.2%. Generative-AI skills appear in the qualifications of 37.8% of data/AI postings: 72.6% in applied AI and 11.8% in analytics.
More than a third of data jobs sit outside the tech industry.
38% of data/AI roles are at banks, insurers, retailers, health systems and manufacturers, spread across 31% of the employers hiring for this work. They pay less ($161,923 against $212,196 average) and hire entry-level at a slightly higher rate (12.8% against 11.1%).
Data science is six jobs, not one
Grouped by what the work is, the 50,639 postings form six specializations that differ by $83K in median pay, sixfold in how often they ask for generative-AI skills, and from 1.0% to 25.9% in PhD demand. A plan built around one of them serves the other five poorly. See R3 →
| Role family | Share | Employers | Median pay | Entry-level | Staff+ / mgmt | Bachelor’s · Master’s · PhD |
|---|---|---|---|---|---|---|
| Analytics & BIData analyst, reporting, product & marketing analytics | 22.4%n = 11,348 | 79%484 of 610 | $129,750 | 15.4% | 21.5% | 87.0 · 6.7 · 2.5 |
| Marketing Science | 0.8%n = 403 | 22%134 of 610 | $168,000 | 6.5% | 37.6% | 86.4 · 11.7 · 1.9 |
| Product Analytics | 0.7%n = 365 | 20%121 of 610 | $129,900 | 16.7% | 13.5% | 94.4 · 2.8 · 1.4 |
| ML EngineeringModel training, serving, ML platform | 22.3%n = 11,280 | 76%461 of 610 | $213,000 | 8.9% | 34.1% | 74.6 · 16.6 · 8.3 |
| Data EngineeringPipelines, warehousing, analytics engineering | 15.1%n = 7,651 | 80%485 of 610 | $175,500 | 8.9% | 25.0% | 92.1 · 5.6 · 1.0 |
| Applied & Generative AILLM product, forward-deployed, evaluation | 13.8%n = 7,005 | 64%389 of 610 | $200,250 | 8.5% | 34.8% | 78.0 · 9.4 · 5.0 |
| AI/ML ResearchResearch science, NLP, vision, quantitative ML | 11.8%n = 5,952 | 50%308 of 610 | $201,775 | 16.0% | 22.0% | 33.3 · 40.4 · 25.9 |
| Data Science & StatisticsData scientist, quantitative analyst, economist | 11.8%n = 5,951 | 69%420 of 610 | $177,300 | 14.1% | 29.9% | 57.6 · 30.2 · 11.4 |
What each family actually asks for
Skill signatures diverge sharply. Top requirements inside each family, as a share of that family's postings naming them in qualifications. Each capability includes its named tools; where one tool reaches 10% of the family, it's shown in brackets:
| Family | Signature requirements |
|---|---|
| Data Science & Statistics | Python 73% · Statistics 71% · SQL 59% · Machine learning 51% (Deep learning 15%) · R 31% · Data visualization 24% (Tableau 15%) · Big data 24% (Spark 16%) |
| ML Engineering | Machine learning 54% (Deep learning 32%) · Python 54% · LLMs 33% · Cloud platforms 25% (AWS 19%) · C++ 25% · AI agents 23% · Java 20% |
| Data Engineering | Python 57% · SQL 54% · Big data 52% (Spark 43%) · Data engineering 51% · Data warehousing 50% (Snowflake 20%) · Cloud platforms 47% (AWS 37%) · Data modeling 31% |
| Analytics & BI | SQL 46% · Data visualization 41% (Tableau 29%) · Python 34% · Statistics 30% · Excel 23% · Data analysis 22% · Data warehousing 18% |
| Applied & Generative AI | LLMs 52% · Python 46% · AI agents 41% · Cloud platforms 32% (AWS 24%) · Machine learning 29% (Deep learning 14%) · GenAI 27% · RAG 24% |
| AI/ML Research | Machine learning 69% (Deep learning 43%) · Python 59% · C++ 34% · LLMs 29% · Java 25% · Statistics 23% · Distributed systems 16% |
AI/ML Research posts the highest entry-level share of any family at 16.0%, ahead of Analytics & BI on 15.4% and every engineering-side family. Research internships are 39% of it, and four employers (Amazon, Meta, Capital One and NVIDIA) post 41%.
But it's the narrowest door in the dataset. Only 308 employers post these roles at all, and 66% of them want a graduate degree (40.4% master's, 25.9% PhD) against 9.2% in analytics and 6.6% in data engineering. It's a PhD pipeline that hires at the bottom of its own ladder.
Python and SQL are the price of entry; most else is a preference
Where employers separate must-haves from nice-to-haves, Python is a hard minimum in 80% of its mentions and SQL in 76%. Cloud, big-data and LLM skills sit on the line between required and preferred, which makes them elective material. See R1 →
Employers split hard minimums from preferences in the posting itself. It's the cleanest proxy we have for what you must have versus what merely helps.
| Skill | Named in qualifications | Required where employers split | Reading |
|---|---|---|---|
| Python | 50.4% | 80.3%67% of mentions | Required |
| Machine learning | 34.4% | 58.9%83% of mentions | Mixed |
| Deep learning | 16.3% | 59.2%79% of mentions | Mixed |
| NLP | 6.0% | 50.7%71% of mentions | Mixed |
| Computer vision | 4.0% | 62.5%84% of mentions | Mixed |
| Scikit-Learn | 3.5% | 55.9%72% of mentions | Mixed |
| Reinforcement learning | 2.9% | 51.8%75% of mentions | Mixed |
| SQL | 30.4% | 75.6%62% of mentions | Required |
| Statistics | 23.7% | 71.8%70% of mentions | Required |
| Statistical modeling | 2.1% | 72.4%65% of mentions | Required |
| Time series analysis | 2.1% | 54.9%67% of mentions | Mixed |
| Cloud platforms | 23.3% | 47.7%75% of mentions | Mixed |
| AWS | 17.5% | 47.6%72% of mentions | Mixed |
| Azure | 10.5% | 52.7%68% of mentions | Mixed |
| GCP | 10.0% | 52.5%68% of mentions | Mixed |
| S3 | 3.1% | 31.5%79% of mentions | Preferred |
| LLMs | 22.4% | 54.1%71% of mentions | Mixed |
| Big data | 18.0% | 48.1%72% of mentions | Mixed |
| Spark | 12.7% | 47.4%67% of mentions | Mixed |
| Hadoop | 4.0% | 30.5%72% of mentions | Preferred |
| Hive | 2.7% | 29.8%66% of mentions | Preferred |
| Data warehousing | 16.9% | 52.6%72% of mentions | Mixed |
| Databricks | 6.4% | 53.1%66% of mentions | Mixed |
| Snowflake | 6.3% | 51.7%65% of mentions | Mixed |
| Redshift | 3.5% | 51.7%83% of mentions | Mixed |
| dbt | 3.1% | 52.6%58% of mentions | Mixed |
| BigQuery | 2.9% | 50.1%63% of mentions | Mixed |
| Data visualization | 16.4% | 54.3%66% of mentions | Mixed |
| Tableau | 10.3% | 57.5%66% of mentions | Mixed |
| Power BI | 6.3% | 53.8%64% of mentions | Mixed |
| Looker | 2.4% | 61.1%53% of mentions | Mixed |
| AI agents | 16.1% | 43.2%73% of mentions | Preferred |
| MCP | 3.0% | 36.9%73% of mentions | Preferred |
| Tool calling | 3.0% | 52.6%70% of mentions | Mixed |
| Java | 15.2% | 86.9%73% of mentions | Required |
| Data engineering | 14.8% | 67.5%69% of mentions | Required |
| R | 8.6% | 60.9%73% of mentions | Mixed |
| Experimentation | 7.9% | 53.8%63% of mentions | Mixed |
How to read this table
Named in qualifications: share of 50,639 data/AI postings naming the skill as a minimum or preferred qualification. Required where employers split: among the 58.8% of postings that separate minimum from preferred, the share of this skill's mentions that sit in the minimum. Read as required at 65% or more, preferred at 45% or less, mixed in between. A capability includes the specific tools and methods that roll up to it; select a capability's count to show or hide them. Specific skills are listed where at least 2% of postings name them.
Reading the split
Python is the entry ticket. It's named in the qualifications of 50.4% of data/AI postings, and where employers separate minimum from preferred, 80.3% of its mentions are hard minimums. R is named in 8.6%, most often inside Data Science & Statistics (31% of that family), so treat it as a specialism.
Five skills behave as requirements. Java (86.9% of mentions are minimums), Python (80.3%), SQL (75.6%), statistics (71.8%) and data engineering (67.5%). SQL is named in 59% of data science postings and 46% of analytics postings. If you can't write a window function, a large share of both screen you out.
Platforms sit on the line. Cloud platforms (47.7%), big data (48.1%) and Spark (47.4%) are placed as preferences about as often as requirements, and LLMs sit at 54.1%. Inside big data, Hadoop (30.5%) and Hive (29.8%) read as preferences. AI agents is the one capability here that splitting employers list as a preference more often than as a requirement (43.2% minimums).
What employers pair Python and SQL with
Python and SQL are named across nearly every family, so the name alone says little about the job. The other skills in the same posting say more.
| Python asks, by family | With LLM or agent skills | With ML skills | With pipeline & warehouse tools | With analysis tools |
|---|---|---|---|---|
| ML Engineering | 65.2% | 25.3% | 0.2% | 0.1% |
| Analytics & BI | 22.9% | 17.5% | 13.0% | 27.1% |
| Data Engineering | 30.9% | 18.9% | 32.6% | 1.4% |
| Data Science & Statistics | 41.5% | 37.3% | 4.0% | 2.7% |
| Applied & Generative AI | 86.1% | 0.6% | 0.0% | 0.0% |
| AI/ML Research | 66.7% | 23.4% | 0.1% | 0.1% |
Share of each family's postings that name Python in qualifications, by what else the posting names. Pairings are checked in a fixed order, so a posting naming both LLM and ML skills counts once, under LLM. Rows don't sum to 100%: the rest pair with other skills or none.
Python means different jobs. Across data/AI postings, 51.4% of Python asks come with LLM or agent skills, 21.6% with ML skills, 8.3% with pipeline and warehouse tools and 4.9% with analysis tools such as pandas or Tableau. Analytics postings break the pattern: their Python sits most often next to analysis tools (27.1%), and data engineering's next to pipeline tools (32.6%).
SQL is mostly warehouse SQL. 59.0% of SQL asks come with pipeline or warehouse tools and 22.3% with analysis tools. Even in analytics postings the split is 48.1% to 41.1%.
Data skills are now required outside the data team
Finance names SQL in 27.3% of its roles and operations in 12.8%, and those two functions together post more roles than the whole data/AI market. Experimentation and AI-agent skills have spread the same way, into marketing and product. See R4 →
Across the full 362,785-posting corpus, data competencies now show up as stated requirements in functions that don’t carry a data title.
| Function | SQL | Python | Data analysis | Statistics | Data viz | Experi- mentation | Machine learning | AI agents | Metrics & KPIs |
|---|---|---|---|---|---|---|---|---|---|
| Data & Analytics 19,860 | 52.9 | 48.5 | 19.8 | 39.5 | 34.5 | 10.8 | 21.4 | 6.0 | 6.1 |
| Finance 10,458 | 27.3 | 6.7 | 7.8 | 6.4 | 22.2 | 0.5 | 1.3 | 1.8 | 8.2 |
| Operations 45,283 | 12.8 | 4.4 | 8.3 | 3.4 | 9.2 | 0.8 | 2.0 | 2.8 | 7.1 |
| Product Management 17,319 | 7.5 | 3.2 | 6.2 | 5.7 | 6.9 | 6.6 | 6.1 | 9.3 | 10.1 |
| Marketing 10,065 | 8.2 | 2.7 | 6.8 | 1.8 | 10.8 | 7.7 | 1.6 | 5.1 | 6.4 |
| Engineering 139,447 | 8.8 | 33.9 | 4.4 | 4.6 | 3.2 | 1.7 | 6.9 | 6.6 | 1.7 |
| Customer Success 8,676 | 8.4 | 9.8 | 4.7 | 0.8 | 5.7 | 0.6 | 1.8 | 3.7 | 5.9 |
| Sales & Business Dev 50,466 | 5.0 | 6.1 | 4.3 | 1.6 | 4.1 | 0.7 | 3.6 | 4.5 | 4.4 |
| People & HR 6,934 | 1.9 | 1.3 | 4.8 | 1.0 | 3.9 | 0.6 | 1.2 | 1.6 | 2.9 |
| Design 7,542 | 1.3 | 4.4 | 2.6 | 3.4 | 3.0 | 5.0 | 2.0 | 5.9 | 2.1 |
| Legal & Policy 5,549 | 6.5 | 2.4 | 4.5 | 1.1 | 3.3 | 0.3 | 2.7 | 3.6 | 2.2 |
| Infrastructure & IT 14,367 | 6.5 | 13.5 | 3.3 | 1.4 | 5.2 | 0.5 | 3.0 | 4.5 | 2.6 |
| Content / Editorial 1,114 | 1.4 | 3.3 | 2.1 | 0.8 | 2.2 | 1.3 | 1.7 | 2.0 | 2.8 |
About this chart
Every cell is directly labelled and colour is a redundant encoding. Counts beside each function name are that function's total postings in the corpus. Data visualisation and AI agents include the named tools that roll up to them. Read down a column to see how far a competency has travelled.
Three columns matter most.
SQL became general business literacy. 27.3% of finance roles and 12.8% of operations roles name it in their qualifications. Those two functions alone are 15.4% of all the hiring we observe — more postings than the entire data/AI market (14.0%).
Experimentation crossed into marketing and product at 7.7% and 6.6%, level with its rate inside AI/ML (7.4%). Designing and reading a controlled test stopped being a specialist skill.
AI-agent asks reach product management ahead of engineering. 9.3% of product roles name AI agents in their qualifications, against 6.6% of engineering roles; design (5.9%), sales (4.5%) and customer success (3.7%) trail. Count any AI expectation at all and 32.8% of design roles and 32.5% of product roles carry one, above engineering's 25.8%.
“Proficiency in Python and SQL to build analytics and tools, backed by a strong foundation in statistics, time series, and risk measurement”
“Deep experimentation experience. You know how to build an experiment backlog, design rigorous tests, and use results to drive compounding improvements”
The data-literate versions of finance, operations, marketing and product careers are, taken together, a bigger market than data science itself.
What transfers, in order of demand outside the data organisation: SQL, data visualisation, data analysis and metrics definition, then, at smaller shares, experiment design and applied AI-tool fluency, which concentrate in product, marketing and design. Machine learning stays inside the technical functions, 6.1% or less everywhere outside data, AI/ML and engineering, so leave it to the specialists unless you're becoming one.
Seniority is bought with leadership and production skills
From entry to staff level, asks for leadership skills rise 30 points, more than any technical skill. The technical climbs are LLMs, AI agents, cloud platforms and distributed systems, while entry postings ask more often for Python, SQL and statistics. See R5 →
Entry and intern roles are 11.8% of the data/AI postings where we can read a level. Staff-and-above individual contributors are 15.2%, and managers 13.0%. Comparing entry with staff-level individual contributors shows what experience buys on the technical ladder. Managers are shown separately, because leading people is their job description.
About this chart
Bars extend from a zero midline. Axis spans ±17.4 percentage points. Staff-and-above pools staff, senior staff, principal, distinguished and fellow individual contributors; managers are left out of this comparison.
The technical core is front-loaded. Employers ask entry candidates for Python, SQL, statistics and data analysis at higher rates than they ask staff-level candidates, because at that level those are assumed and the posting spends its words on what's scarce.
What's scarce is everything after the model works. Getting it into production. Instrumenting it. Running it across distributed systems. Staff-level postings also name more methods: 1.81 techniques per posting against 0.91 at entry, while named tools rise from 4.71 to 5.64.
| Soft skill | Entry | Staff-level IC | Managers |
|---|---|---|---|
| Leadership skillsAny of 15 skills, most often stakeholder management, leadership, mentoring and executive influence | 17.5% | 47.6% | 71.0% |
| Interpersonal skillsAny of 14 skills, most often communication, collaboration, cross-functional work and influence | 55.7% | 64.1% | 69.4% |
| Cognitive skillsAny of 24 skills, most often analytical thinking, problem solving, agility and ambiguity tolerance | 47.8% | 45.3% | 54.7% |
| Influence | 3.8% | 22.9% | 25.1% |
| Leadership | 4.4% | 21.4% | 32.9% |
| Stakeholder management | 9.5% | 19.4% | 29.9% |
| Mentoring | 2.5% | 14.4% | 11.6% |
| Technical leadership | 0.2% | 13.1% | 6.8% |
| People management | 0.6% | 2.2% | 24.3% |
| Analytical thinking | 23.1% | 11.8% | 21.5% |
| Communication | 45.1% | 43.0% | 52.1% |
About this table
Soft skills group into subclasses; a group counts when any skill in it is named. A single-skill row counts that skill by its own name: communication does not absorb written or verbal communication.
“Experience thriving in multi-team, multi-stakeholder environments with the ability to build trust quickly and drive outcomes through influence”
“Experience leading teams or serving as a technical lead for multi-person data science initiatives, including mentoring, roadmap-setting, and quality review.”
Entry postings already ask for more Python, SQL and statistics than staff-level postings do. On the individual-contributor ladder the steepest climb is in leading others: leadership skills rise 30.1 points and interpersonal skills 8.4, while cognitive skills fall 2.5. The gap worth closing is the ambiguous, stakeholder-facing, ship-it end of the work: take a vague business question, scope it, build something that runs unattended, defend the result to someone who didn’t ask for a model, and bring a less experienced teammate along.
That argues for building something real for a real stakeholder, end to end. A notebook alone won't do it.
Building systems pays; reporting on them doesn’t
Against comparable postings at the same employer, function and level, foundation models, deep learning and distributed systems pay 6–8% more; Excel, office suites and reporting tools pay 5–12% less. Python, asked for in half of all postings, adds +0.1% once the specialty is held fixed. See R1 →
28,821 of the 50,639 roles publish a salary range. Two questions follow: what the ladder looks like, and which competencies move pay once you control for where someone sits on it.
How this chart is built
What a bar is. The pay gap between postings that ask for the skill and comparable postings at the same employer, function and level that don’t, averaged across employers, as a percent of base pay. The bar weights the estimate inside each function this population spans by its postings (AI / ML 43%, Data & Analytics 39%, engineering’s data ladders 18%). Dollars use the $187,938 median published midpoint; published ranges carry base only, so dollar values understate senior and AI roles.
- Table stakes: asked for in a quarter or more of postings; plotted with the specialty held fixed, pooled figure in the tooltip.
- Sub-ladder: the premium mostly disappears once the specialty is held fixed; the skill marks a better-paid specialty.
- Bundle: the premium is credited jointly with the skills it usually appears with.
- Varies by level: fewer than half the seniority bands agree with the pooled sign.
A capability row opens into the tools inside it that carry their own price; a pairing prices a broad skill by the other skills named with it. Meta-cognitive skills are not nested. 138 of 149 skills asked for in at least 2% of postings are charted; 11 are withheld because one or two employers dominate them or too few price them. 33 rows carry a sub-ladder or bundle label. Premiums don’t add up across skills.
Hover a row for its three function figures, its specialty or co-occurring skills, and its entry-level read where eight or more employers price it at entry (129 rows).
“Conduct deep dives into product and usage data to surface actionable insights, size opportunities, and influence roadmaps across product, engineering, and go-to-market teams”
“Product-minded and impact-driven, with a passion for applying cutting-edge ML and AI techniques to real-world problems”
“Collaborate with P/E/D/D (product, engineering, data and design) teams to develop product roadmaps and measure success”
“Optimization: Develop methodologies to explore/simulate the impact of new interventions and develop data products to optimize product/operational strategies”
“Analyze and profile complex workloads (e.g., LLM training, Rendering pipelines) to identify bottlenecks in compute, memory bandwidth, and latency”
“Awareness of modern machine learning approaches (e.g. embeddings, sequence models, deep learning) and interest in applying them to real-world problems”
“Deep expertise in both machine learning systems and backend distributed systems, with the ability to build production systems that scale as the business grows.”
“Strong technical foundation in either applied machine learning or full-stack / distributed systems engineering (strength in one, fluency in the other).”
How to read the pattern
The line runs between building and running systems and reporting on them. Foundation models (+8.0%), deep learning (+6.6%, +1.9% once its specialty is held fixed), causal inference (+6.6%, +1.8% held fixed) and machine learning (+6.1% pooled, +2.2% once its specialty is held fixed) sit above comparable roles at the same employer; Excel (−12.0%), Microsoft Office (−10.4%), attention to detail (−7.7%), Tableau (−4.7%) and troubleshooting (−3.3%) sit below them. Technical leadership (+7.4%) pays as well as the engineering skills, and the whole bottom of the chart is office tools, reporting tools and the generic personal qualities that appear in every posting. The pairings say more than the flat numbers: SQL is −5.9% in analytics pairings; Python runs from +2.6% with ML skills to −5.4% with analytics skills around a pooled figure that fits neither pairing.
Read a negative figure as a signal about role shape rather than a penalty. Holding employer, function and level fixed, naming a reporting tool marks a role as more reporting-oriented, and the market prices the difference. Most of the differences are small: half the charted skills sit inside ±3.2% of base. The tails are wider, from +8.0% for foundation models, roughly $15K on this population's median range, to −12.0% for Excel, roughly $23K the other way.
Three limits worth stating. A capability is priced only from the postings that name it directly, so machine learning (+6.1% pooled) and data visualisation (−2.8%) carry their own figures, but a capability and the tools inside it are different measurements and are never ranked against each other: each capability row expands to its own tools instead. And where the estimator flags that a premium tracks the sub-ladder or the stack a skill identifies rather than the skill itself, the row is labelled rather than dropped, which is 33 of the 138 charted rows. Python is asked for in 50.4% of these postings; its pooled +2.9% is the data-engineering and ML ladders it marks, and held to the sub-function it is +0.1%, which is the figure the chart plots for it and the other three table-stakes skills: expected everywhere, a raise nowhere, unless it comes paired with ML or LLM skills. And the bar averages over five seniority bands. Each tooltip carries the entry-level read where eight or more employers price the skill at entry, and it is often the more useful number if you are starting out: Python reads +5.3% at entry on 141 employers and SQL +1.2% on 93, against −3.7% for SQL with the specialty held fixed. Those per-level re-fits carry no confidence interval, so they annotate the chart rather than order it; agile, c++, data governance, GCP, git, prototyping, R, verbal communication and written communication are marked varies by level because fewer than half of their readable bands agree with the pooled sign.
Entry pay, in context
At entry level across the full 362,785-posting corpus, AI/ML roles pay the most of any function at $163,632 average (n = 791), ahead of engineering at $131,397. Data & Analytics entry roles average $120,027, above product management, operations, marketing and finance, but well behind the AI track.
Which of the six families a graduate enters sets their starting ladder more than the degree does.
AI raises the bar with seniority; it isn’t closing the junior door
Asks to build with AI climb from 34% of entry-level postings to 57% at staff level, and generative-AI skills appear in 38% of data/AI postings. Among employers tracked throughout, the entry share of new postings rose from 157.0% in May to 205.0% in September. See R7 →
About this chart
This curve alone can't rule out a closing junior door, because a requisition that was never posted never appears in it. The flow check below is the direct test: among employers tracked throughout, the entry-level share of new postings stayed between 9.5% and 13.3% from May to September. What the curve does show is that the expectation to build with AI concentrates at the levels furthest from entry.
37.8% of data/AI postings name at least one generative-AI skill in their qualifications. The specific asks: LLMs 22.4%, AI agents 16.0%, generative AI broadly 12.1%, retrieval-augmented generation 7.7%, foundation models 5.3%, AI coding assistants 4.8%, prompt engineering 4.5%, LLM frameworks 4.2% and fine-tuning 3.8%.
The qualification: penetration is uneven across the six families.
- Applied & Generative AI: 72.6%
- ML Engineering: 55.3%
- AI/ML Research: 48.3%
- Data Science & Statistics: 28.8%
- Data Engineering: 23.7%
- Analytics & BI: 11.8%
LLM depth is essential if you're heading into ML or applied AI and close to irrelevant if you're heading into business analytics. That argues for a shared statistical and engineering core with a genuinely divergent second half, rather than one AI checkbox bolted onto every plan.
“Practical experience integrating AI/LLM capabilities into applications or workflows (e.g., prompt-based interfaces, retrieval-augmented generation, AI agents, copilots…”
“Build and maintain the metric definitions, metadata, and business context that enable AI agents to help stakeholders explore trusted data and answer business questions independently”
“Architect and scale Retrieval-Augmented Generation (RAG) pipelines — including chunking, embedding, indexing, and retrieval strategy — for accuracy and latency at production scale”
Movement over the window
Trends below follow the same 37 employers from May to September.
About this chart
Weekly, 4 May to 28 Sept 2026, on the same 37 employers tracked throughout, as in the rest of this section. Each point is one week's new postings at those employers (382–564 of them), so a single week moves easily: the range runs 31.3% to 48.3% with no trend behind it.
Week to week, generative-AI demand moved between 31.3% and 48.3% with no consistent direction. Month to month it is steadier and just as flat: 37.2% in May, 39.3% in September.
About this chart
Each point is a job function. The dashed curve is twice May’s share, so anything above it more than doubled; no function did, and Finance came closest, from 5.2% to 9.5% (1.8×). Two functions did not rise: Customer Success (11.0% to 10.8%) and AI / ML (49.0% to 47.5%, still the highest share); Data & Analytics, the other marked data/AI function, edged up from 16.2% to 16.3%. Corpus-wide the share rose from 13.6% in May to 17.0% in September. Hover a point for its May share and its multiple; the smallest September cell is 311 postings.
The flat line above is a vantage point, not the market. Hold the employer set fixed and eleven of the thirteen functions asked for generative-AI skills more often in September than in May — corpus-wide, 13.6% of new postings to 17.0% — and the steepest climbs are all outside the data organisation: design (+13.3 points), product management (+5.3), sales (+4.5), finance (+4.3). The data functions barely moved: AI / ML slipped from 49.0% to 47.5% and data & analytics edged up from 16.2% to 16.3%; customer success (11.0% to 10.8%) is the other function that did not rise. By September a design posting was about twice as likely to carry a generative-AI qualification as a data-and-analytics posting (30.2% against 16.3%). Both data functions started above the corpus-wide rate and the wave has moved past them — the same diffusion section 4 shows for SQL and experimentation, moving faster.
Among new postings where we can read a level, the entry share held in a 9.5%–13.3% band: 9.5% in May, 13.3% in September. Whatever generative AI is doing to this market in this window, it isn't closing the junior door.
Why these 37 employers
Month-over-month changes across all employers would mostly reflect which employers we added and when. An employer counts here only if it was tracked for the whole comparison, which holds the group fixed. Across all employers the same months would show growth that is mostly new tracking.
When tracking starts, an employer's first week brings in roles that had been open for a while, and those over-represent hard-to-fill roles. That's all of April for most of these employers, so April isn't compared with later months. Five months of new postings is short, so read these as levels and treat movement of a few points as noise.
AI is changing what existing jobs ask for, not what they’re called
Eight AI capabilities, led by AI agents and AI coding assistants, are named by 30% or more of employers. Dedicated AI job titles are 3.6% of hiring; the change is happening inside titles that already existed. See R7 →
We looked for new skills and new job titles separately. They tell different stories, and the difference is the finding.
The skill side is moving. Eight AI capabilities are already named in the qualifications of postings at 30% or more of the 628 employers.
About this chart
Breadth rather than volume, because a skill named by half the employers is a market and a skill named 3,000 times inside 5 companies is a house style. A capability counts the named tools and protocols that roll up to it, so each employer counts once. For scale, Python reaches 97% of employers and SQL 95%.
For direction, the same 37 employers in May against September:
About this chart
Each point is one capability: the share of September’s new postings naming it, against how much that share moved since May. Shares run from 0.22% to 7.52%, so the horizontal axis is logarithmic and the dashed curve marks twice May’s share — a point above it more than doubled. Prompt engineering is the only capability that both doubled and reached a share worth building a course around (1.42% to 3.51%). AI agents moved further in absolute terms off a much larger base (5.13% to 7.52%). Fine-tuning is the one capability that slipped (0.71% to 0.67%). The cluster at the left is real but thin: a tenth of a point on a 0.2% base is a handful of postings.
Three patterns in that data
Prompt engineering asks more than doubled. They rose from 1.42% of new postings in May to 3.51% in September at the same 37 employers, the only capability on the chart to more than double. AI agents added more points (+2.39) off a base of 5.13%.
Evaluation is spreading across employers. Evals reach 53% of employers, human-in-the-loop 37% and guardrails 33%. Read together, those describe a job: deciding whether a model output is good enough to ship, and building the machinery that decides it at scale. Their share of new postings is small and moved less than a fifth of a point.
About this chart
A capability line includes its named tools: a posting asking for GitHub Copilot counts toward AI coding assistants. Instrument shares run 0.2–0.9% of new postings — 35 to 150 postings a month — so single months move easily — the zoomed panel is there for the shapes, not for precision.
The pattern in one sentence: AI coding assistants are broad and roughly level — 78% of employers name one, and the ask held between 1.3% and 1.5% of new postings — while Claude Code and GitHub Copilot fade from qualifications; the AI-agents capability climbed from 5.1% to 7.5% while every framework under it moved erratically. None of this vocabulary is new (every term was already appearing before May), and no agent framework holds a direction for more than two consecutive months. Where a posting names a specific tool, 79.6% ask for fluency in using it and 9.8% ask the role to build on it.
“Establish evaluation frameworks to measure the quality, cost, and impact of AI-generated code and agentic workflows across the organization”
“Design safety boundaries and guardrails for model outputs, including tool-call allowlists and fallback logic”
“Design rigorous evaluation frameworks using offline benchmarks, online A/B experiments, golden datasets, regression suites, LLM-as-a-Judge, and human evaluation.”
New roles are a different story
Most frontier role categories we track are flat or shrinking as a share of new postings among employers tracked throughout; AI-prefixed engineering titles and Forward Deployed are the two that rose.
| Role category | Share of hiring | Employers | May share | Sept share | Direction |
|---|---|---|---|---|---|
| AI-prefixed, non-AI functionAI Product Manager, AI Architect, AI Recruiter | 1.09%n = 3,944 | 52%325 of 628 | 0.96% | 1.0% | flat |
| AI-prefixed, engineeringAI Engineer, AI Scientist | 1.24%n = 4,491 | 55%345 of 628 | 0.54% | 0.79% | rising |
| GenAI-prefixedGenAI Engineer, Generative AI Lead | 0.56%n = 2,018 | 20%128 of 628 | 1.16% | 0.8% | falling |
| Forward DeployedForward Deployed Engineer and variants | 0.52%n = 1,903 | 27%168 of 628 | 0.68% | 0.95% | rising |
| Applied AI / ML | 0.35%n = 1,279 | 25%158 of 628 | 0.43% | 0.35% | flat |
| Agent-prefixedAgentic AI Engineer, AI Agent Engineer | 0.21%n = 744 | 20%127 of 628 | 0.26% | 0.24% | flat |
| AI safety & governance | 0.11%n = 390 | 17%106 of 628 | 0.1% | 0.09% | flat |
| AI evaluation / evalsIn titles, as opposed to in requirements | 0.03%n = 110 | 7%42 of 628 | n/a | n/a | too small |
| GTM Engineer | 0.01%n = 51 | 6%40 of 628 | n/a | n/a | too small |
Together those categories are 3.6% of all hiring we observe; on the same 37 employers they went from 3.4% of May’s new postings to 3.7% of September’s. The AI shift is happening inside job titles that already existed. A posting called Senior Data Scientist now asks for evals and AI agents; it doesn't get renamed.
About this chart
Titles containing AI, GenAI, LLM, agent or eval as a word; 16,650 of the 50,639-posting data/AI population (32.9%), across 481 employers. A different denominator from the corpus-wide table above, which counts title-novelty categories against all 362,785 postings. On the 37 employers tracked throughout, AI-forward titles were 38.1% of May–June’s new data/AI postings and 36.1% of August–September’s. Broad, and not spreading.
Inside the data/AI population the renaming has a sharp edge. The builder families wear AI in the title, at 82.7% of applied-AI postings, 60.4% of ML engineering, and the one genuinely consolidating new ladder is AI Engineer: 2,154 postings across 250 employers, with senior, staff, lead and principal rungs already attached. The analyst-side families participate far less: 1.6% of Analytics & BI and 6.0% of Data Science & Statistics postings name AI in the title (357 postings in the latter, titles like “Data Scientist, Agents”). Yet generative-AI skills sit in the qualifications of 37.8% of the whole population: the title tells you which side of the build/decide line a role sits on, not whether it touches AI.
For career planning, build the competencies these titles ask for, not the titles.
One genuinely new role, and one to watch
Forward Deployed Engineer is the real one. It reaches 27% of employers and carries a full seniority ladder already, with senior, staff, principal and manager variants all appearing. It's an engineer who sits with the customer and builds the integration in the room.
GTM Engineer is tiny at 0.01% of hiring, but it appears at 40 employers across 51 postings, almost one per company. The job is automating go-to-market with AI tooling.
What this section can and can’t see
A new skill only shows up once we recognise it, and our check for unrecognised terms found no new AI vocabulary in this window; a few recognised terms (Devin, Cline, Ollama) first appear after May, each in fewer than 100 postings. The title check counts open and closed postings alike, but only titles it recognises as one of these categories; a new role under an unfamiliar title won't appear. Some AI terms signal a posting's general AI posture rather than a stated requirement, and those aren't counted as requirements here.
Five months of new postings is a short read for anything moving less than a percentage point.
The widest entry doors aren’t at the biggest employers
Seven companies post more than a fifth of data/AI roles and hire at entry level slightly below the market rate. Quant trading firms post 42% of their roles at entry level and semiconductor makers 23%. See R8 →
We grouped the roles into eight employer cohorts. MAANGOS here means Meta, Amazon, Apple, Netflix, Google, OpenAI and SpaceX. Seven companies, 22.7% of every data/AI role in the corpus.
| Cohort | Share | Employers | Entry | Staff+ / mgmt | Median pay | GenAI ask | PhD |
|---|---|---|---|---|---|---|---|
| Everyone elseStartups, SaaS, media, health, gaming | 25.3%≈ 12,076 roles | 363 | 8.5% | 34.8% | $205,000 | 46.3% | 12.3% |
| MAANGOSMeta, Amazon, Apple, Netflix, Google, OpenAI, SpaceX | 22.7%≈ 10,848 roles | 7 | 10.9% | 19.8% | $197,750 | 35.4% | 9.4% |
| Enterprise non-techRetail, healthcare, pharma, industrial | 20.2%≈ 9,657 roles | 143 | 12.8% | 28.1% | $143,642 | 30.1% | 4.7% |
| Finance & insurance | 19.8%≈ 9,462 roles | 61 | 10.3% | 30.2% | $175,000 | 36.1% | 3.6% |
| SemiconductorsNVIDIA, Qualcomm, Micron, AMD and peers | 5.2%≈ 2,465 roles | 11 | 22.9% | 24.4% | $203,025 | 51.8% | 17.0% |
| Other big techMicrosoft, IBM and peers | 4.4%≈ 2,105 roles | 3 | 20.8% | 32.6% | $190,400 | 25.2% | 9.7% |
| Frontier AI labsAnthropic, Mistral, xAI, Cohere and peers | 1.5%≈ 716 roles | 8 | 17.5% | 29.0% | $310,000 | 68.8% | 38.3% |
| Quant tradingSystematic and high-frequency trading firms | 1.0%≈ 494 roles | 14 | 42.2% | 3.6% | $212,500 | 9.9% | 29.0% |
What each cohort buys that the others don't
Take the skills a cohort names in qualifications, divide by how often the rest of the data/AI market names them, and the house style shows up. Multiples below are that ratio. Where one employer supplies most of a cohort's mentions, it's named in brackets.
| Cohort | Over-indexed skills (share of cohort · lift vs rest of market) |
|---|---|
| MAANGOS | Design patterns 6.5% · 5.4× (Amazon 94%) · Data structures 15.0% · 4.9× (Google 54%) · Model evaluation 7.4% · 4.3× (Google 90%) · C++ 26.7% · 3.1× (Amazon 51%) |
| Frontier AI labs | JAX 22.1% · 9.5× · Distributed training 8.7% · 6.1× · Rust 7.5% · 5.8× · Reinforcement learning 12.7% · 4.6× (DeepMind 51%) |
| Semiconductors | Computer architecture 6.7% · 15.4× (NVIDIA 52%) · CUDA 19.8% · 15.0× (NVIDIA 69%) · Systems engineering 8.3% · 12.9× (Qualcomm 84%) · Compilers 10.1% · 12.3× |
| Finance & insurance | Responsible AI 9.7% · 3.6× (JPMorgan Chase 53%) · Agile 13.0% · 3.0× · Troubleshooting 7.3% · 2.4× (JPMorgan Chase 55%) · Scala 9.2% · 2.4× (Capital One 69%) |
| Enterprise non-tech | Power BI 13.9% · 3.2× · Databricks 12.2% · 2.5× · DevOps 6.4% · 2.5× · Azure 18.7% · 2.2× |
| Quant trading | JAX 7.7% · 3.0× · Linux 8.9% · 2.5× · C++ 28.0% · 2.2× · Statistics 34.6% · 1.5× |
Four things worth your attention
Quant trading has the most lopsided level mix in the dataset. 42.2% of its postings are entry-level, the highest of any cohort, and only 3.6% are staff or management. These firms hire quantitative talent young, train it in-house, and post almost nothing senior.
They also care least about generative AI, at 9.9% against a 37.8% market average, while 29.0% want a PhD. Their stated skills barely stand out from the rest of the market (C++ at 2.2×), so what distinguishes them is the level mix. If you're strong in probability and can code, it is the widest entry door in the corpus.
Amazon and Google screen on classical CS fundamentals harder than anyone. Design patterns shows up in 6.5% of MAANGOS data/AI postings, 5.4× the rest of the market, 94% of it at Amazon. Data structures appears in 15.0%, 4.9×, with Google supplying 54%. C++ at 26.7%, 3.1×. If you're aiming at those two, you need the algorithms material, whatever your specialism.
MAANGOS is also the flattest of the big cohorts: 19.8% staff-and-above against 34.8% for everyone else, and 10.9% entry-level.
Frontier labs are a ceiling, not a target. $310,000 median, 38.3% want a PhD, 68.8% name a generative-AI skill in qualifications. Also 8 employers and 1.5% of the market.
Semiconductors is a hidden ML market. GPU computing appears in 26.4% of their data/AI postings, 11.3× the rest of the market, and CUDA, the most-named GPU tool, in 19.8%, 15.0×. NVIDIA supplies about two-thirds of both. 51.8% name a generative-AI skill, second only to frontier labs. And 22.9% are entry-level at a $203,025 median. Pairing computer architecture with deep learning matches this market's ask more closely than a standard data-science curriculum does.
| Employer | Share | Entry | Staff+ / mgmt | Median pay | GenAI ask | Largest family |
|---|---|---|---|---|---|---|
| Amazon | 10.6%≈ 5,087 roles | 14.5% | 8.0% | $169,050 | 18.6% | AI/ML Research |
| 6.7%≈ 3,212 roles | 4.0% | 32.0% | $213,000 | 45.6% | ML Engineering | |
| JPMorgan Chase | 4.8%≈ 2,285 roles | 10.1% | 25.9% | $175,000 | 52.1% | Analytics & BI |
| Apple | 2.8%≈ 1,353 roles | 8.6% | 23.6% | $231,150 | 61.6% | ML Engineering |
| Capital One | 2.3%≈ 1,117 roles | 5.8% | 32.1% | $216,200 | 35.3% | Applied & Generative AI |
| Microsoft | 2.0%≈ 953 roles | 13.7% | 35.2% | $190,400 | 44.6% | AI/ML Research |
| NVIDIA | 1.9%≈ 914 roles | 19.0% | 15.6% | $235,750 | 62.6% | ML Engineering |
| IBM | 1.9%≈ 901 roles | 54.5% | 5.7% | — | 8.6% | Data Engineering |
| Booz Allen Hamilton | 1.4%≈ 684 roles | 24.2% | 11.0% | $126,800 | 24.3% | Analytics & BI |
| UnitedHealth Group | 1.4%≈ 682 roles | 8.4% | 26.7% | $137,250 | 46.1% | Analytics & BI |
| Walmart | 1.4%≈ 649 roles | 1.1% | 65.1% | $165,000 | 15.2% | Data Science & Statistics |
| Mastercard | 1.3%≈ 642 roles | 6.9% | 25.5% | $142,500 | 25.5% | Analytics & BI |
| Citi | 1.3%≈ 628 roles | 14.9% | 59.7% | $157,000 | 23.3% | Analytics & BI |
| Johnson & Johnson | 1.3%≈ 605 roles | 13.4% | 28.1% | $134,400 | 25.0% | Analytics & BI |
| Meta | 1.2%≈ 560 roles | 29.4% | 32.1% | $212,000 | 69.4% | AI/ML Research |
| Salesforce | 1.2%≈ 557 roles | 2.9% | 40.1% | $201,500 | 54.2% | Applied & Generative AI |
Entry-level share swings from 1.1% at Walmart to 54.5% at IBM across employers of similar market share. Meta runs an open junior door at 29.4% and Qualcomm at 25.5%. Capital One (5.8%) and Adobe (7.1%) hire at entry level well below the market's 11.8%, whatever their headcount suggests.
Generative-AI demand splits the same way. Meta names a GenAI skill in the qualifications of 69.4% of its data/AI postings, NVIDIA 62.6%, Apple 61.6%, Adobe 61.0%. IBM sits at 8.6% and Walmart at 15.2%. Same calendar year, very different jobs.
Cohort caveats
Frontier labs and quant trading are each around 1% of the market (n = 747 and n = 517 postings). Percentages hold at n ≥ 100 but treat single-digit differences there as noise. Median pay appears only where 30+ postings publish a range, which is why IBM has none.
Cohorts are mutually exclusive and assigned in the order listed, so OpenAI counts under MAANGOS rather than frontier labs, and Salesforce falls into Everyone else. Amazon and Google are 17.4% of the data/AI market between them, so MAANGOS numbers move with those two.
Lift multiples leave out industry labels, which restate a cohort's own industry, and products a cohort's own employers make, such as Claude at Anthropic and S3 at Amazon. Other big tech is omitted from the lift table because Microsoft supplies nearly all of its mentions.
More than a third of the market sits outside tech, and Boston leans AI
38% of data/AI roles are at banks, insurers, retailers, health systems and manufacturers, which hire at entry level slightly more often than tech. Boston is 2.6% of the market, led by applied AI and research roles, with financial services its largest employer base. See R6 →
Boston, specifically
Boston is 2.6% of the global data/AI market in this window (n = 1,308). The mix is distinctive and, for a new program, favourable. Shares below are of Boston's own 1,308:
- Boston skews AI. Applied & Generative AI (26.7%) and AI/ML Research (18.6%) are the two largest families locally. Analytics & BI comes third at 17.0%; nationally it is the largest family.
- Where the demand sits: Financial Services (tech) 26.7%, Consulting 15.5%, Big Tech 14.4%, InsurTech 6.3%, Healthtech 5.5%.
- Top local employers, as a share of Boston's estimated 1,205 data/AI roles: Capital One (25.9%), Amazon (11.8%), Accenture (11.8%), Whoop (5.7%), Wayfair (3.4%), Klaviyo (3.1%), McKinsey & Company (2.5%), Boston Consulting Group (2.4%).
- Entry-level share is 11.7%, in line with the national rate. The local market is about as open to new graduates as anywhere else.
More than a third of the market sits outside tech
38% of the data/AI market comes from companies whose business isn't technology: banks, insurers, retailers, health systems, pharmaceutical firms, industrial manufacturers. That's 187 of the 610 employers posting these roles, so nearly one in three.
They pay less on average ($161,923 against $212,196) but hire entry-level at a slightly higher rate, 12.8% against 11.1%. And they concentrate in the Northeast corridor rather than the Bay Area.
For placement, that's the more accessible part of the market. Programs pointed entirely at big-tech and AI-lab hiring miss it.
| Sub-industry | Share | Average pay | Entry |
|---|---|---|---|
| Big Tech / Hyperscalers | 25.2%n = 12,761 | $198,408 | 11.4% |
| Financial Services (tech) | 14.4%n = 7,271 | $181,428 | 11.8% |
| Semiconductors | 5.0%n = 2,522 | $212,162 | 22.9% |
| Consulting / Professional Services | 3.4%n = 1,711 | $145,840 | 19.2% |
| Healthcare (tech) | 3.3%n = 1,674 | $140,442 | 8.7% |
| Productivity & Enterprise SaaS | 3.2%n = 1,635 | $211,900 | 6.3% |
| Fintech | 3.0%n = 1,494 | $194,894 | 4.1% |
| Social Platforms | 2.3%n = 1,178 | $236,361 | 18.6% |
| Insurance (tech) | 2.1%n = 1,077 | $147,931 | 8.7% |
| Frontier AI Labs | 2.1%n = 1,059 | $338,840 | 13.5% |
| Pharma & Life Sciences (tech) | 2.1%n = 1,050 | $173,201 | 12.7% |
| Grocery & General Merchandise (tech) | 1.6%n = 808 | $162,222 | 1.5% |
On the highest-paying rows
Frontier AI Labs and Social Platforms are small, self-selecting cohorts driven by a handful of top-of-market payers. Between them they account for 4.4% of the data/AI market. Read them as a ceiling.
What to do with this
Eight moves, each tied to specific evidence above, ordered by how much of the observed demand they address.
Take Python everywhere and SQL to production standard
Python in everything you build. SQL to the level of window functions, CTEs and query performance, against a warehouse rather than a toy database. R only if you're on the statistics or life-sciences path. Learn Python inside your lane's own tools: analysis libraries for analytics, pipeline tools for data engineering, LLM and agent frameworks for applied AI.
Why: Python is named in the qualifications of 50.4% of data/AI postings, and where employers separate minimum from preferred, 80.3% of its mentions are hard minimums. It reaches 73% of data science and 57% of data engineering postings. SQL is named in 30.4% overall with 75.6% required, and 46–59% in the two analytics-facing families. R appears in 8.6%. What surrounds Python follows the family: analysis tools lead in analytics postings (27.1%), pipeline tools in data engineering (32.6%). SQL asks come with warehouse or pipeline tools 59.0% of the time, 48.1% even in analytics postings.
Keep causal inference and experimental design in your core
A full course's worth: randomised experiment design, power and sample size, sequential testing, quasi-experimental methods (difference-in-differences, regression discontinuity, instrumental variables), and how to report an effect to a non-technical decision maker.
Why: experimentation is named in 7.9% of data/AI qualifications (mixed required/preferred, section 3) and 10.8% of Data & Analytics function postings — and it is the one method that has crossed functions: marketing names it as often as AI/ML roles do (7.7% against 7.4%, section 4). Section 6 prices causal inference at +6.6% against comparable roles at the same employer, function and level, but the estimator flags it twice: hold the specialty fixed and it falls to +1.8%, and the credit is shared with the econometrics, R and experimentation stack it travels with. Experimentation itself reads +3.9%, and the estimator does not trace that one to a specialty. So causal inference earns its place through breadth across roles and functions, and experimental design carries a premium of its own.
Pick your family, then commit to a lane
A shared core first (programming, statistics, data management, ethics), then a lane: Analytics & decision science Data & ML engineering Applied AI. Each lane has its own door, its own pay band and its own employer set.
Why: the six families differ by $83K in median pay, by a factor of 6 in generative-AI penetration (11.8% to 72.6%), and in PhD expectation from 1.0% to 25.9%. One undifferentiated plan serves all six badly.
Not going into data? Get data-literate anyway
The sequence that transfers: SQL and data access; then metrics definition, experiment reading, visualisation and applied AI-tool fluency. No advanced math required.
Why: 27.3% of finance roles and 12.8% of operations roles name SQL in their qualifications, 22.2% of finance roles name data visualisation, and 10.1% of product roles name metrics and KPIs. Those three functions are 20.1% of all the hiring we observe, against 14.0% for the entire data/AI market.
Ship something operational for a real stakeholder
A project with a real client, ending in something that runs unattended: a deployed model or pipeline with monitoring, plus a defence to the client's own stakeholders. The handover is the skill.
Why: on the individual-contributor ladder, asks for leadership skills rise from 17.5% at entry to 47.6% at staff level, influence from 3.8% to 22.9%, distributed systems from 4.6% to 17.1% and productionization from 3.4% to 12.6%, while entry postings ask for more Python, SQL and statistics. Communication is named in 42.6% of qualifications, behind Python's 50.4%.
Look outside tech first
Financial services, insurance, healthcare, consulting and retail hire the same roles with less competition. The Boston-area employers already posting these roles: Capital One, Accenture, Whoop, Wayfair, Klaviyo, McKinsey & Company, and the regional health and insurance systems.
Why: 38% of data/AI roles sit at non-technology companies, spread across 187 of the 610 employers hiring for this work, at a slightly higher entry-level rate than tech (12.8% against 11.1%). In Boston specifically, financial services is the largest category at 26.7% of the local market, ahead of consulting (15.5%) and big tech (14.4%).
Use AI everywhere; go deep in one lane only
Universal: using AI assistants competently, evaluating model output, knowing the failure modes, and the governance and responsible-use questions. Deep, if you choose this lane: LLM architecture, RAG, fine-tuning and agent design.
Why: AI-build expectation climbs from 33.7% at entry to 57.2% at staff level, so it's what seniority looks like here. But generative-AI skills appear in the qualifications of 72.6% of applied-AI roles against 11.8% of analytics roles, so universal depth would be wasted while universal fluency wouldn't. Responsible AI is named in the qualifications of 4.1% of data/AI postings and AI governance in 0.9%. Where postings name a specific AI tool, 79.6% ask for fluency in using it and 9.8% ask the role to build on it.
Strongly quantitative? Look at quant trading and semiconductors
Two cohorts hire young and get almost no attention from data science candidates. The 14 systematic trading firms in this corpus and the chip designers are the most open doors for probability-strong coders.
Why: quant trading posts 42.2% of its data/AI roles at entry level, the highest share of any cohort here, against a market average of 11.8%. Semiconductors posts 22.9% entry at a $203,025 median, and 51.8% of its postings name a generative-AI skill. Between them they're 6.2% of the market.
Limits of this evidence
What to discount, and by how much.
Show the limits
Postings tell you what employers asked for. They don't tell you who got hired, at what salary, or whether the role was filled at all. We see the requisition and how long it stayed open.
We chose the 637 companies. They're picked for relevance to technology careers, so big technology employers are over-represented against the US economy, and small firms, startups below roughly Series B, government, academia and non-profits are largely absent. Read every figure as applying to employers of this kind.
Shares are the figure that travels. Because the roster is a deliberate sample rather than a census, a raw posting count says as much about which companies we watch as about the market. Shares are stable under that: if Analytics & BI is 22.4% of data/AI hiring here, adding 50 more employers of the same kind moves that number very little, while the count would jump. Every table leads with a share and keeps the underlying n underneath it so you can check the base.
Pay coverage is 57% and non-random. Companies that publish ranges skew US-based and subject to pay-transparency law. Medians from a low-disclosure sub-industry come from a self-selected slice.
Skill figures are floors. A skill counts only once we recognise it, and some postings are written in ways that are hard to read reliably, so true shares are likely somewhat higher.
Skill pairings are inferred. Where the brief says what Python, SQL or an AI tool is paired with, that's read from the other skills and wording in the same posting.
Some roles are posted more than once. Large employers open separate requisitions for the same role and location, so 50,639 data/AI postings are an estimated 47,824 distinct roles. That tilts shares slightly toward the heaviest posters: Amazon is 11.0% of postings and 10.6% of distinct roles.
Seniority and job family are inferred. Both are read from each posting, and not every posting states a level. The Other data/AI family in section 2 is a catch-all, so don't read it as a coherent role type.
Six months is a short window. Long enough for a level read of the market. The trends we report (sections 7 and 8) are restricted to 37 employers tracked for the whole comparison, and even those readings are provisional.
We left remote out. Only 18% of these postings state a remote policy at all, and silence doesn't mean on-site. Of those that do state one, 71% are remote-eligible, which is too selection-biased to act on.
Where this evidence comes from
How the figures in this brief were produced, in short.
Show how the figures were produced
Peakbound tracks job postings published on the career pages of 637 employers. This brief covers postings first seen between 13 April and 10 October 2026: 362,785 postings across 71 sub-industries and 77 countries, of which 50,639 are data, analytics or AI roles (14.0%). Every query behind the figures is written up separately, so any figure here can be audited back to its query.
Closed postings count. Every figure includes roles that have since closed. A posting usually comes down because the role was filled, so leaving closed roles out would undercount the fastest-moving demand.
Pay is employer-published only. Pay figures use the ranges employers publish, which covers 57% of data/AI roles. We don't estimate the rest.
Comparisons are controlled. Where we compare groups, we hold the obvious differences fixed and say which. Where a finding weakens under that control, we report the weaker number.
Skills count where employers ask for them. A skill counts when a posting lists it as a requirement or a preference. A named tool also counts toward the capability it belongs to, so a posting that asks for Tableau counts toward data visualization.
Supporting detail
Skill pairings, how demand splits across rival tools, and credentials: the detail behind section 3.
Show the detail
Skills that travel together
Build skills in the bundles employers actually ask for. Most frequent co-occurrences in qualifications, leaving out pairs where one skill rolls up into the other (Spark always counts toward big data):
- Python + machine learning: 24.3% of all postings
- Python + SQL: 23.7%
- Python + statistics: 17.5%
- Python + cloud platforms: 16.8%
- Python + LLMs: 15.0%
- Python + big data: 14.1%
- SQL + statistics: 13.6%
Employers name a capability, then pick a tool
Demand for most tools is split across rivals. Share of data/AI postings naming each capability in qualifications, with its most-named tools:
- Cloud platforms: 23.3%, with AWS 17.5%, Azure 10.5%, GCP 10.0%, S3 3.1%
- Data warehousing: 16.9%, with Databricks 6.4%, Snowflake 6.3%, Redshift 3.5%, dbt 3.1%
- Data visualization: 16.4%, with Tableau 10.3%, Power BI 6.3%, Looker 2.4%, QuickSight 1.4%
- Deep learning: 16.3%, with PyTorch 11.1%, TensorFlow 8.2%, JAX 2.6%, Transformers 2.4%
- Workflow orchestration: 6.0%, with Airflow 5.1%, Dagster 0.6%, Prefect 0.3%
No warehouse is named in more than 6.4% of postings, while the capability reaches 16.9%. Teaching warehousing through one platform covers the ask; the market doesn't converge on a single platform.
The cloud ask leans toward ML: 56.1% of cloud-platform asks come with ML skills and 28.6% with pipeline tools.
Certifications barely register
Only 2.6% of data/AI postings list a professional certification in their qualifications (7.6% mention one anywhere in the posting), and the ones named come from adjacent functions. PMP leads at 0.4%, then SAFe at 0.3%; no other credential reaches 0.3%. Six Sigma (0.7%) and SAFe (0.1%) also appear as methodology asks without the credential.
Among the 56% of postings that state a degree level, the signal is academic: 71.2% bachelor's, 17.6% master's, 8.9% PhD. Where a field of study is named, computer science dominates at 15.6% of all data/AI postings, but mathematics (4.3%) and statistics (3.9%) get named on their own terms.
This is one report. The engine runs nightly.
Peakbound reads these 600+ career pages every night. The app points the same evidence at your career: which skills to build next, what they actually pay, and where the doors are for someone with your profile.
Join the Peakbound waitlist →