nico's blog

Reading preferences

Text size

← All posts
19 min read

The State of Data & AI Hiring (2026)

An analysis of 2026 data and AI hiring trends, including in-demand skills, salary ranges, experience levels, and how roles differ.

On this page

On this page

Note. I built the scraper, the cleaning and deduplication pipeline, the keyword-based skill extractor, the LLM-based enrichment job, and the public dashboard behind this dataset. This blog post is the analytical backbone. A follow-up blog post will detail the architectural choices I made.

Motivation. When I began studying data science in 2016, the industry essentially recognized two paths: Data Scientists who built predictive algorithms, and Data Analysts who queried data to build dashboards. Yet, even in that simplified landscape, companies constantly scrambled the labels. It was very common to open a “Data Scientist” job description only to find a pure dashboarding role, or vice versa.

Today, the ecosystem has fractured into highly specific niches — Data Engineer, MLOps Engineer, AI Engineer, Forward Deployed Engineer. Seeing this explosion of distinct job titles, I asked myself: are we now exposed to an even greater amount of misclassification? Does every job title actually represent a distinct technical reality?

To find out, I built a scraper to strip away the titles from a substantial amount of job offers and analyze the raw skill requirements underneath. But as the data rolled in, what started as a simple skill-mapping exercise evolved into a comprehensive audit of the current data job market, capturing not just what companies expect candidates to know (i.e., the skill requirements), but how roles are distributed, what they pay, and how they operate.

TL;DR

  • Core roles still dominate volume: the market is still anchored by traditional titles, with Data Engineers (24.4%), Data Analysts (19.2%), and Data Scientists (16.6%) making up the bulk of the sample.

  • Distinct fingerprints: roles differentiate themselves through specific tooling: Data Analysts over-index on Power BI and Tableau, Data Engineers rely heavily on ETL/ELT and Spark, and AI Engineers are defined by LLMs, RAG, and Agentic AI.

  • Python as the quasi-universal requirement: while skills like Machine Learning and SQL are popular, Python is the only true foundational requirement that acts as a baseline across all data families.

  • The GenAI divide: GenAI is not uniformly required across the data landscape. It appears in 89.6% of AI Engineer postings and 70.3% of Forward Deployed Engineer postings, but only 38.7% of Data Scientist and 7.7% of Data Analyst roles.

  • The entry-level bottleneck: the vast majority of roles target mid-level professionals (2-7 years of experience), with Data Analysts offering the highest share of 0-1 year entry opportunities at just 6.5%.

  • Experience alters the skill mix: as experience requirements increase, the requested skill mix fundamentally shifts. Senior bands place a progressively heavier emphasis on cloud infrastructure (AWS, Azure, Databricks, GCP) and architectural competencies compared to junior bands.

  • Remote work dynamics: work arrangement is heavily influenced by company size and role. The share of remote work declines significantly as company size increases. ML/AI Engineers and MLOps Engineers secure the highest remote shares (over 25%), while Research roles remain predominantly on-site.

What is observable in the data?

  1. Canonical skills100,575 / 107,28493.7%
  2. Company size74,012 / 107,28469.0%
  3. Business sector72,346 / 107,28467.4%
  4. Required experience53,898 / 107,28450.2%
  5. Work arrangement46,550 / 107,28443.4%
  6. Usable salary35,915 / 107,28433.5%

Share of all job postings with an observable field

This chart shows that field availability varies across job postings: skill matches are available for 93.7% of postings, compared with 50.2% for required experience, 43.4% for work arrangement, and 33.5% for usable salary.

The overall leaders and the strongest cross-role skills

The following analysis includes both technical skills and broader conceptual or foundational knowledge. The left panel below shows the most frequently required skills overall: Python (71.6%), SQL (52.8%), Machine Learning (45.4%), ETL/ELT (33.5%), and LLMs (27.2%). The right panel ranks skills by their minimum within-family share across the 12 supported job families, showing which skills have the broadest coverage across roles. Python is the only skill mentioned in at least 48.4% of postings even in the family where it is least common.

Here, this minimum-rate provides a stricter insight into cross-role breadth. A high pooled share can coexist with low coverage in particular families (e.g., SQL clearly stands out here).

Most common overall

  1. Python71.6%
  2. SQL52.8%
  3. Machine Learning45.4%
  4. ETL/ELT33.5%
  5. LLMs27.2%
  6. AWS24.9%
  7. Azure24.2%
  8. CI/CD22.9%
  9. Statistics18.9%
  10. Power BI18.7%
  11. Spark18.1%
  12. GCP16.9%

Lowest rate across eligible families

  1. Python48.4%
  2. Machine Learning11.9%
  3. ETL/ELT4.9%
  4. SQL4.4%
  5. C/C++4.4%
  6. LLMs3.8%
  7. Computer Vision3.6%
  8. R3.5%
  9. AWS2.9%
  10. GCP2.5%
  11. Generative AI2.3%
  12. Git2.1%

The table below compares within-family rates for the five pooled leaders:

Skill Pooled share Equal-family mean Weighted minus equal-family Lowest family rate Highest family rate
Python 71.6% 70.9% +0.7 pp Research Scientist, 48.4% Data Scientist, 86.2%
SQL 52.8% 36.0% +16.8 pp Research Engineer, 4.4% Analytics Engineer, 90.5%
Machine Learning 45.4% 56.8% -11.3 pp Data Analyst, 11.9% ML Scientist, 94.5%
ETL/ELT 33.5% 23.4% +10.1 pp Research Scientist, 4.9% Data Engineer, 71.6%
LLMs 27.2% 33.8% -6.5 pp Data Analyst, 3.8% AI Engineer, 71.4%

The biggest composition effects run in opposite directions. SQL and ETL/ELT gain 16.8 and 10.1 percentage points in the volume-weighted sample, consistent with their concentration in larger data-oriented families. Machine Learning and LLMs are 11.3 and 6.5 points higher under equal-family weighting, indicating stronger prevalence in several smaller AI, ML, and research families.

Job Fingerprints

In the visualizations below, each panel asks: which skills are mentioned more often in this family’s ads than in ads for all the other families combined? The number is the difference between those two shares, measured in percentage points (pp) and rounded to the nearest integer. For example, +41 pp means that Power BI appears in Data Analyst ads 41 percentage points more often than it appears in ads for the other families.

Data Analyst

n=18,062

  1. Power BI+41 pp
  2. Tableau+28 pp
  3. SQL+25 pp
  4. Statistics+14 pp
  5. R+8 pp
  6. Looker+7 pp

Enrichment vs other families (pp)

Data Engineer

n=24,999

  1. ETL/ELT+51 pp
  2. SQL+34 pp
  3. Spark+31 pp
  4. Databricks+24 pp
  5. Data Warehousing+24 pp
  6. CI/CD+22 pp

Enrichment vs other families (pp)

Data Scientist

n=17,041

  1. Machine Learning+39 pp
  2. Statistics+36 pp
  3. R+21 pp
  4. Python+18 pp
  5. Statistical Modeling+15 pp
  6. pandas+12 pp

Enrichment vs other families (pp)

AI Engineer

n=12,468

  1. LLMs+50 pp
  2. RAG+43 pp
  3. Agentic AI+27 pp
  4. Prompt Engineering+24 pp
  5. LangChain+22 pp
  6. Generative AI+22 pp

Enrichment vs other families (pp)

MLOps Engineer

n=1,130

  1. MLOps+66 pp
  2. CI/CD+58 pp
  3. Kubernetes+54 pp
  4. Docker+47 pp
  5. Machine Learning+37 pp
  6. MLflow+37 pp

Enrichment vs other families (pp)

Forward Deployed Engineer

n=4,387

  1. LLMs+23 pp
  2. Agentic AI+21 pp
  3. Go+12 pp
  4. RAG+12 pp
  5. Prompt Engineering+9 pp
  6. Java+8 pp

Enrichment vs other families (pp)

These fingerprints are coherent with the distinctions suggested by the family labels. The charts also show overlap: SQL is distinctive to Data Engineer but is also mentioned by most of Data Analyst ads, and Machine Learning (as conceptual knowledge) is prevalent in Data Scientist offers while also common for more engineering-focused roles such as ML Engineer.

Some roles advertise a much longer skill list

This plot counts distinct canonical skill mentions per skill-bearing ad:

Dot = median; thick interval = middle 50%; thin interval = P10-P90.

  1. MLOps Engineern=1,13013 middle 50%: 9 to 17. P10 to P90: 6 to 23.
  2. AI Engineern=12,46810 middle 50%: 6 to 15. P10 to P90: 3 to 21.
  3. Data Engineern=24,99910 middle 50%: 7 to 14. P10 to P90: 4 to 19.
  4. ML Engineern=10,60710 middle 50%: 7 to 15. P10 to P90: 4 to 20.
  5. Data Scientistn=17,0419 middle 50%: 5 to 13. P10 to P90: 3 to 18.
  6. ML Scientistn=6739 middle 50%: 6 to 12. P10 to P90: 4 to 16.
  7. Analytics Engineern=3,6068 middle 50%: 5 to 11. P10 to P90: 3 to 14.
  8. Applied Scientistn=1,2658 middle 50%: 5 to 11. P10 to P90: 4 to 14.
  9. Forward Deployed Engineern=4,3876 middle 50%: 3 to 10. P10 to P90: 2 to 15.
  10. Research Engineern=2,7985 middle 50%: 3 to 8. P10 to P90: 1 to 12.
  11. Data Analystn=18,0624 middle 50%: 2 to 7. P10 to P90: 1 to 9.
  12. Research Scientistn=3,5394 middle 50%: 2 to 7. P10 to P90: 1 to 10.

Distinct detected mentions per posting

MLOps Engineer has the highest median at 13 (middle 50%: 9–17; P10–P90: 6–23). AI Engineer, Data Engineer, and ML Engineer each have a median of 10. On the other extreme, Data Analyst and Research Scientist have medians of 4.

The distribution is strongly role-dependent. However, note that a longer list is not evidence that every item is a hard requirement or that those roles demand a larger amount of skills in practice.

Count of concrete technologies, not just concepts

Total canonical labels (the chart above) include conceptual and methodological skills, while the chart below counts the defined technology subset only. The gap between the two counts can suggest whether ads emphasize broad, conceptual knowledge (e.g. Machine Learning, NLP, ETL/ELT, Computer Vision, etc.) or named tools.

Dot = median; thick interval = middle 50%; thin interval = P10-P90.

  1. MLOps Engineern=1,1309 middle 50%: 5 to 12. P10 to P90: 3 to 17.
  2. Data Engineern=24,9997 middle 50%: 5 to 11. P10 to P90: 3 to 14.
  3. Analytics Engineern=3,6066 middle 50%: 4 to 8. P10 to P90: 2 to 11.
  4. AI Engineern=12,4685 middle 50%: 2 to 9. P10 to P90: 1 to 13.
  5. Data Scientistn=17,0415 middle 50%: 3 to 8. P10 to P90: 1 to 11.
  6. ML Engineern=10,6075 middle 50%: 3 to 9. P10 to P90: 1 to 13.
  7. Forward Deployed Engineern=4,3874 middle 50%: 1 to 7. P10 to P90: 1 to 10.
  8. ML Scientistn=6734 middle 50%: 2 to 6. P10 to P90: 0 to 8.
  9. Applied Scientistn=1,2653 middle 50%: 2 to 5. P10 to P90: 0 to 8.
  10. Data Analystn=18,0623 middle 50%: 2 to 5. P10 to P90: 1 to 7.
  11. Research Engineern=2,7982 middle 50%: 1 to 4. P10 to P90: 0 to 7.
  12. Research Scientistn=3,5392 middle 50%: 1 to 3. P10 to P90: 0 to 5.

Distinct detected mentions per posting

This repeats the same distribution design for the taxonomy’s concrete-technology subset. MLOps Engineer again leads, with a median of 9 technologies (middle 50%: 5–12), while Data Analyst, Research Engineer, and Research Scientist have medians of 3 and 2 respectively.

How often do job postings mention GenAI?

In order to mark a job posting as mentioning any GenAI skill, we first need to exhaustively enumerate the list of skills and competencies we are looking for. The defined marker rule counts a posting once if it contains any of the following GenAI markers:

AWS Bedrock, Agentic AI, Anthropic API, AutoGPT, Azure OpenAI, Chroma, CrewAI, DSPy, DeepEval, FAISS, Fine-tuning, Gemini API, Generative AI, LLMOps, LLMs, LanceDB, LangChain, LangFuse, LangGraph, Langsmith, LiteLLM, LlamaIndex, LoRA, Marqo, Milvus, MCP, Ollama, OpenAI API, OpenAgents, OpenRouter, PGVector, Pinecone, Prompt Engineering, QLoRA, Qdrant, Quantization, RAG, Smolagents, Vector DBs, Vertex AI, Vespa, Weaviate, Zilliz, vLLM

  1. AI Engineer11,175 / 12,46889.6%
  2. Forward Deployed Engineer3,086 / 4,38770.3%
  3. ML Engineer6,672 / 10,60762.9%
  4. ML Scientist413 / 67361.4%
  5. Applied Scientist755 / 1,26559.7%
  6. Research Engineer1,320 / 2,79847.2%
  7. MLOps Engineer506 / 1,13044.8%
  8. Data Scientist6,589 / 17,04138.7%
  9. Research Scientist1,283 / 3,53936.3%
  10. Analytics Engineer722 / 3,60620.0%
  11. Data Engineer4,254 / 24,99917.0%
  12. Data Analyst1,383 / 18,0627.7%

Share of skill-bearing postings that contain at least one GenAI marker

Overall, 37.9% of skill-bearing job offers contain a GenAI marker. However, this is far from being fairly distributed across job families. The clearest finding here is a strong role stratification: at least one selected GenAI marker appears in nearly 9 in 10 AI Engineer ads and 7 in 10 Forward Deployed Engineer ads, versus fewer than 1 in 5 Data Engineer ads and fewer than 1 in 10 Data Analyst ads. ML and research roles sit between those ends, suggesting this vocabulary is concentrated in AI-facing work but is not exclusive to AI titles.

Which role portfolios actually resemble each other?

The heatmap below shows how similar each role family’s skill-mention profile is to every other family’s. Comparing these profiles helps reveal which roles emphasize similar skills. I calculated cosine similarity from the share of skill-bearing ads mentioning each of the 342 skills, then order families using exploratory average-linkage clustering.

Data AnalystData EngineerAnalytics EngineerData ScientistML EngineerResearch ScientistResearch EngineerApplied ScientistML ScientistMLOps EngineerAI EngineerForward Deployed Engineer
Data Analyst1.000.710.840.740.430.430.400.410.430.340.400.48
Data Engineer0.711.000.890.720.620.400.460.460.470.640.610.71
Analytics Engineer0.840.891.000.700.510.380.410.400.430.480.500.60
Data Scientist0.740.720.701.000.860.810.790.830.860.670.750.74
ML Engineer0.430.620.510.861.000.850.900.890.930.830.880.82
Research Scientist0.430.400.380.810.851.000.970.930.920.540.670.64
Research Engineer0.400.460.410.790.900.971.000.940.920.630.730.70
Applied Scientist0.410.460.400.830.890.930.941.000.940.580.720.68
ML Scientist0.430.470.430.860.930.920.920.941.000.640.730.66
MLOps Engineer0.340.640.480.670.830.540.630.580.641.000.760.73
AI Engineer0.400.610.500.750.880.670.730.720.730.761.000.95
Forward Deployed Engineer0.480.710.600.740.820.640.700.680.660.730.951.00

Cell labels are cosine-similarity scores rounded to two decimals; higher means more aligned skill-rate profiles

Each of the 12 families is represented by a list of 342 numbers: for every skill, the share of that family’s skill-bearing ads that mention it. The heatmap compares those lists. A higher cosine-similarity score means the two families have a more similar pattern of skill mentions. I really want to emphasize that the score does not mean they share that percentage of skills. It means their skill-mention rates have fairly similar shapes. In fact, if every rate in one list (e.g., Data Engineer) were exactly half the corresponding rate in the other (e.g., Data Analyst), cosine similarity would be 1, even though the overall rates differ. This is because cosine similarity is about the angle between the vectors, not how far apart their components percentages are. Importantly too, neither the number of ads in each family or the overall level of skill mentions have any effect on the computed values.

Also, in case you wonder what average-linkage clustering is and why it is useful here: this clustering is a separate step performed after computing all the cosine pairs. It arranges the families so that similar ones appear near each other in the heatmap (which is why you see a cluster of high values in the center). It doesn’t generate or change the similarity scores.

Coming back to our heatmap: the most similar pair is Research Scientist–Research Engineer (0.97). AI Engineer–Forward Deployed Engineer, ML Scientist–Applied Scientist, and Research Scientist–Applied Scientist are also strongly related. Unsurprisingly, we also find the related pair: Data Engineer–Analytics Engineer at 0.89 and Data Analyst–Analytics Engineer at 0.84.

Many labels, how many distinct skill portfolios?

−1.5−1−0.500.511.52−0.500.511.5PC1 (41.1% of profile variance)PC2 (29.3%)Data EngineerAnalytics EngineerData AnalystData ScientistML ScientistResearch EngineerApplied ScientistResearch ScientistMLOps EngineerAI EngineerML EngineerForward Deployed Engineer
  • Core data
  • Science/research
  • AI/ML engineering

Core data, Science/research, and AI/ML engineering are aditorial groups, not learned clusters. One point per search-derived family; equal centroid weighting. Axis signs are arbitrary. Distances outside the first two components are not displayed. Point positions are read off the original figure (approximate to about 0.01).

PCA (Principal Component Analysis) turns each family’s 342 skill-mention rates into a dot on a two-dimensional map. Each dot represents the family average, and each family counts equally even though some have far more ads than others. Considering that the question I care about is “How do the typical skill profiles of these 12 job families differ?”, performing PCA this way - giving each family equal weight - seemed like a reasonable choice to me.

Each axis combines many skill rates into one score; a dot’s position shows the family’s score on those two combinations. PCA chooses the combinations that preserve as much as possible of the vqriqnce among the 12 family averages. The first axis captures 41.1% of the total variance and the second 29.3%, for 70.4% combined. This describes how much of the variation among the family averages the map retains. Because the map leaves 29.6% out, families that look close may still differ in ways that these two axes do not show.

The skills that contribute most to the first axis help explain what the map is showing. SQL (+0.45), ETL/ELT (+0.24), Power BI (+0.23), and dbt (+0.20) point in one direction; Machine Learning (-0.42), LLMs (-0.25), PyTorch (-0.24), and Deep Learning (-0.22) point the other way. In simple terms, this axis contrasts a data and analytics toolset with an ML and AI toolset. The second axis is mostly associated with CI/CD (+0.43), MLOps (+0.33), Kubernetes (+0.31), and Docker (+0.29), and points away from Statistics (-0.16), Deep Learning (-0.14), R (-0.11), and Reinforcement Learning (-0.10). In this dataset, it roughly contrasts operations and deployment terms with statistical and research-oriented skills. The plus and minus signs could be flipped without changing the result, and the markers and colors (Core data, Science/research, AI/ML engineering) are chosen labels of mine — not groups discovered by PCA.

The three colors broadly line up with visible patterns, but they do not form three perfectly separate clusters:

  • Core data sits on the positive/right side of PC1, consistent with that axis’s SQL, ETL/ELT, and Power BI emphasis. The three points are spread vertically rather than forming a tight ball: Data Engineer is higher, Data Analyst lower, and Analytics Engineer between them. This vertical alignment makes sense when compared to the main skills associated with PC2 (CI/CD, MLOps, Kubernetes, Docker), for which intuition dictates that Data Engineer is generally a more technical position than Data Analyst and has a higher chance to also require those skills, while Analytics Engineer sits in-between.

  • AI/ML engineering is mostly in the upper-left or near the center of the map. MLOps Engineer is much higher than the other points in this group, matching the operations-heavy PC2 direction; Forward Deployed Engineer sits near the center rather than deep inside the group. The heatmap previously depicted supports some of these relationships: for example, AI Engineer and Forward Deployed Engineer have a 0.95 similarity score.

  • Science/research is mostly in the lower-left area. Research Scientist and Research Engineer are close in the full-profile heatmap (0.97), as are ML Scientist and Applied Scientist (0.94). Data Scientist is nearer the center of the PCA map, so it does not sit as distinctly with the other science/research points in this two-axis view.

So the coloring reflects a broad pattern in the skill profiles: data-focused families tend toward one side, while many AI/ML and science/research families occupy different parts of the other side. But keep in mind that PCA simplifies and displays profiles; it does not decide how many groups exist or assign group membership.

Which jobs have a more accessible entry route?

The heatmap below uses all eligible postings in each family, including those without skills, and displays the share of job postings with a specific band of years of experience required. It retains an explicit Unknown experience column.

0-12-45-78-1112+Unknown
Data Engineer2.2%14.9%23.6%9.3%4.2%45.8%
Data Analyst6.5%22.8%16.7%3.5%1.9%48.4%
Data Scientist3.2%16.1%20.5%8.0%3.9%48.4%
AI Engineer2.5%13.8%21.3%8.8%4.6%49.0%
ML Engineer1.5%10.7%18.0%7.1%3.4%59.4%
Forward Deployed Engineer1.6%15.6%23.2%8.5%4.5%46.5%
Research Scientist2.8%18.3%17.5%6.6%5.5%49.3%
Analytics Engineer1.5%16.6%20.9%5.1%1.5%54.3%
Research Engineer3.0%16.5%14.7%4.6%2.8%58.5%
Applied Scientist0.9%6.9%17.4%9.9%3.5%61.4%
MLOps Engineer0.3%10.6%21.3%6.0%2.2%59.7%
ML Scientist0.4%10.7%14.8%7.6%2.9%63.4%

All eligible postings in each family

Required experience is absent in 53,386 postings (49.8%). Only 3,330 postings in the full cohort (3.1%) explicitly target Junior/Graduate profiles with 0–1 year of previous experience. These figures reveal how brutal and limited the data & AI market id for newcomers. It offers concrete evidence for what many entry-level job seekers have been complaining about lately.

The most accessible entry-level job is by far Data Analyst, but with a relatively low share of 6.5% of postings. We see that more “niche” roles such as MLOps Engineer and ML Scientist are almost impossible to land with no previous experience. However, a word of warning here: notice that these roles also have the highest share of Unknown experience, and we do not know how thes shares would re-distribute if we were given the ground truth. Thefore, it is not an estimate of entry-level opportunities across all postings. Missing experience cannot be interpreted as zero years or as an entry-level opening.

0-12-45-78-1112+
Intern91.7%6.0%1.3%0.5%0.5%
Junior48.7%42.0%7.5%1.5%0.3%
Mid-Level1.3%41.2%38.7%11.8%7.1%
Senior0.2%15.3%57.8%19.3%7.5%
Principal/Lead0.2%5.7%30.7%39.8%23.7%
Principal/Lead+0.0%6.8%35.6%29.4%28.2%

Required experience within each observed seniority label; unknown labels/experience are discarded. Both attributes known: n=22,172

This heatmap conditions on both seniority and required experience being known (n=22,172), then normalizes experience shares within each seniority label. The heatmap reads this way: Intern ads are 91.7% in the 0–1 band, and Junior ads are 48.7% in that band plus 42.0% in the 2–4 years band. Senior ads are concentrated at 5–7 years (57.8%), with 15.3% in 2–4 and 19.3% in 8–11.

The broad gradient is sensible, but it is not a one-to-one mapping. The reason is that seniority may be inferred from job-title language, while experience is extracted from ad text. Therefore, disagreement is expected and useful; it indicates that “Senior” does not imply the same level of experience and responsibilities universally, but ultimately depends on each company’s internal classification. YoE (years of experience) is a factual and more reliable marker. This is also why I favoured it over seniority in all the Datalent’s visualizations.

Salary visualizations

Each panel below uses advertised salary midpoints (i.e., the average between the lower and upper brackets usually advertised in job offers, e.g., “Our base compensation is $100,000 - $115,000”) in one native currency and displays the median, middle 50% (P25–P75), and P10–P90.

A few notes:

  • The intervals show cross-ad spread, not confidence intervals for the median.

  • A midpoint can conceal the width of an advertised range.

  • Advertised pay is not always total compensation.

Salary records with missing or unsupported currency are excluded, as well as Intern salaries which would bias the analysis.

Country, source, seniority, experience, employer, and role composition are not adjusted in these descriptive panels. Finally, do not interpret the EUR pooling plot as cost-of-living-adjusted pay, because this is not the case.

Advertised pay within United States

Annualized USD midpoints; dot = median, thick interval = middle 50%, thin interval = P10-P90. n=21,929

  1. ML Scientistn=224$211,100 middle 50%: $175,500 to $244,500. P10 to P90: $151,000 to $298,000.
  2. Applied Scientistn=587$196,600 middle 50%: $176,500 to $223,500. P10 to P90: $168,000 to $254,000.
  3. Forward Deployed Engineern=989$195,000 middle 50%: $160,000 to $217,500. P10 to P90: $135,000 to $250,000.
  4. ML Engineern=2,746$195,000 middle 50%: $162,000 to $232,000. P10 to P90: $133,500 to $275,000.
  5. Research Engineern=791$192,000 middle 50%: $131,000 to $237,500. P10 to P90: $99,000 to $298,000.
  6. AI Engineern=2,160$172,000 middle 50%: $141,500 to $210,000. P10 to P90: $113,500 to $250,000.
  7. MLOps Engineern=221$165,000 middle 50%: $141,000 to $200,000. P10 to P90: $125,000 to $240,000.
  8. Data Scientistn=3,973$157,500 middle 50%: $126,500 to $189,000. P10 to P90: $103,500 to $222,500.
  9. Research Scientistn=1,525$153,199 middle 50%: $100,000 to $217,500. P10 to P90: $79,500 to $287,500.
  10. Analytics Engineern=804$145,000 middle 50%: $116,500 to $175,000. P10 to P90: $99,000 to $200,500.
  11. Data Engineern=4,132$145,000 middle 50%: $122,500 to $172,500. P10 to P90: $101,000 to $205,000.
  12. Data Analystn=3,777$101,450 middle 50%: $80,500 to $127,000. P10 to P90: $67,000 to $155,000.

Annual advertised midpoint (USD)

Only families with 30+ observations

Advertised pay within United Kingdom

Annualized GBP midpoints; dot = median, thick interval = middle 50%, thin interval = P10-P90. n=4,929

  1. ML Scientistn=41£82,500 middle 50%: £61,500 to £98,000. P10 to P90: £45,000 to £111,000.
  2. MLOps Engineern=47£80,000 middle 50%: £59,500 to £90,000. P10 to P90: £52,000 to £97,500.
  3. ML Engineern=388£78,483 middle 50%: £60,000 to £100,000. P10 to P90: £47,500 to £130,000.
  4. Forward Deployed Engineern=236£77,600 middle 50%: £62,000 to £110,500. P10 to P90: £50,500 to £143,000.
  5. Applied Scientistn=45£75,000 middle 50%: £60,500 to £105,000. P10 to P90: £52,500 to £135,000.
  6. AI Engineern=571£74,000 middle 50%: £59,500 to £93,500. P10 to P90: £49,500 to £125,000.
  7. Data Scientistn=764£64,445 middle 50%: £51,500 to £80,500. P10 to P90: £44,000 to £99,500.
  8. Data Engineern=1,278£63,583 middle 50%: £52,500 to £79,500. P10 to P90: £44,500 to £95,000.
  9. Analytics Engineern=157£62,748 middle 50%: £53,000 to £76,500. P10 to P90: £46,000 to £91,000.
  10. Research Engineern=119£62,245 middle 50%: £50,500 to £102,500. P10 to P90: £43,000 to £146,500.
  11. Research Scientistn=152£53,223 middle 50%: £43,500 to £92,500. P10 to P90: £38,500 to £149,500.
  12. Data Analystn=1,131£45,010 middle 50%: £37,500 to £55,500. P10 to P90: £31,500 to £70,000.

Annual advertised midpoint (GBP)

Only families with 30+ observations

Advertised pay across EUR-currency countries

Annualized EUR salary midpoints; median, P25-P75 and P10-P90. n=6,108 across 11 countries

  1. Forward Deployed Engineern=200€90,000 middle 50%: €67,500 to €106,500. P10 to P90: €55,000 to €144,500.
  2. ML Engineern=458€72,500 middle 50%: €56,000 to €89,000. P10 to P90: €39,000 to €109,500.
  3. MLOps Engineern=53€72,500 middle 50%: €60,000 to €87,000. P10 to P90: €45,000 to €90,000.
  4. Research Scientistn=42€67,000 middle 50%: €50,000 to €88,500. P10 to P90: €42,500 to €107,000.
  5. AI Engineern=985€64,500 middle 50%: €48,000 to €85,000. P10 to P90: €35,000 to €102,500.
  6. Analytics Engineern=168€60,500 middle 50%: €51,000 to €74,500. P10 to P90: €39,000 to €84,500.
  7. Data Scientistn=1,090€57,500 middle 50%: €45,000 to €75,000. P10 to P90: €35,500 to €90,000.
  8. Research Engineern=100€56,784 middle 50%: €44,000 to €73,000. P10 to P90: €32,500 to €100,500.
  9. Data Engineern=1,979€56,400 middle 50%: €44,000 to €72,000. P10 to P90: €36,000 to €84,000.
  10. Data Analystn=1,013€45,000 middle 50%: €35,000 to €57,000. P10 to P90: €30,000 to €71,500.

Annual advertised midpoint (EUR)

Recorded EUR salaries from countries whose local currency is EUR; country codes: AT, BE, DE, ES, FI, FR, IE, IT, LU, NL, PT. Only families with 30+ observations

I wanted a currency-neutral comparison of hierarchy across job families, i.e. to compare the shape of each local salary hierarchy without converting USD, GBP, CAD, and EUR.

To achieve this currency-neutral comparison, we can leverage the fact that Data Analyst has the lowest median among the supported job families in each displayed market, so it provides a consistent base of 100. Each other index is that family’s median divided by the local Data Analyst median, multiplied by 100: an index of 150 means a median salary 50% higher than Data Analyst in that market. A dash means fewer than 30 usable salaries for that family in the panel.

Family USD — United States GBP — United Kingdom EUR — pooled
ML Scientist 208 183 —
Research Engineer 189 138 126
Applied Scientist 194 167 —
Forward Deployed Engineer 192 172 200
ML Engineer 192 174 161
AI Engineer 170 164 143
Research Scientist 151 118 149
MLOps Engineer 163 178 161
Data Scientist 155 143 128
Data Engineer 143 141 125
Analytics Engineer 143 139 134
Data Analyst 100 100 100

In the US, Data Engineer and Analytics Engineer both index at 143; in the UK they index at 141 and 139, which is extremely close. We observe slightly lower figures for Europe, but nothing dramatic.

The specialist tier is less consistent: Research Engineer indexes at 189 in the US but only 138 in the UK and 126 in Europe, while Research Scientist is at 151 in US and 149 in Europe, whereas it is at only 118 in UK. These differences show that “research” is not one uniform pay tier in the job postings, and the central ranges reveal substantial spread over markets.

We also observe that the Forward Deployed Engineer has the highest index at 200 in Europe, very far ahead of the runner-up, which is ML Engineer (161). In contrast, the US and UK feature less extreme indices for this role: in both markets, it sits at the 4th position. One possible explanation I have is that Forward Deployed Engineer emerged first in the US and spread to the UK faster than to the European markets. Therefore, job seekers with ad-hoc knowledge and skills for this relatively new position are in shortage in Europe. This means companies have to offer a premium to attract these good candidates.

Conclusion

This analysis of over 100K job postings highlights a dynamic yet structured data and AI hiring market in 2026. While foundational roles like Data Engineers, Data Analysts, and Data Scientists continue to drive overall hiring volume, emerging specialization — particularly across GenAI and MLOps — has introduced distinct skill fingerprints and expectations across job families.

Key findings demonstrate that Python remains the universal technical baseline across virtually all data disciplines, whereas GenAI tooling is heavily concentrated within specialized AI and engineering roles rather than standard analytics positions. Furthermore, entry-level opportunities remain highly competitive, with mid-level experience dominating demand across most domains. Finally, compensation structures reflect clear functional hierarchies across global markets, rewarding highly specialized technical and engineering profiles with significant salary premiums.