Benchmark Analysis

How Do AI-Related Roles and Our Pathway Compare With the Whole Job Market?

Overview

The other pages study one pathway in one industry (NAICS 5182). This page steps back and benchmarks it against the whole job market in the same data: every industry and every job type in the course Jobs_2026 files. It answers three questions a job seeker would ask before committing to the Data Analyst to Data Scientist path:

1. Does AI exposure change pay? How do salaries differ between AI-related and non-AI postings?
2. How do data careers compare with neighboring ones? Where do data roles sit against software engineering, IT, cybersecurity, and business fields?
3. Which fields combine strong pay with many openings? Where do data careers sit on demand and pay?

What is in the data. The files hold 815,193 postings. After removing duplicates, possible ghost listings, and internships, 733,064 valid postings remain, and 69,246 of them state a salary we can convert to an annual figure (the folder’s own normalization of annual, hourly, weekly, and biweekly pay, kept between $25,000 and $600,000). All pay figures on this page use those 69,246 postings. The script is build_benchmark_tables.py, which saves only small summary tables.

How we define the groups.

• AI-related posting: the job title names AI or machine learning, or the posting text names at least one core AI term (machine learning, artificial intelligence, deep learning, neural networks, natural language processing, large language models, generative AI, computer vision, PyTorch, or TensorFlow). About 5% of postings meet this rule. A posting that merely mentions AI in passing counts, so this measures AI exposure and not only dedicated AI jobs.
• Occupation field: assigned from the job title with transparent keyword rules (listed in the script) into ten computer and business fields: Data Science & ML, Data Engineering, Data Analytics & BI, Software Engineering, IT & Cloud Infrastructure, Cybersecurity, Product & Project Management, Business Analysis & Consulting, Finance & Accounting, and Marketing & Sales. Everything else is “All other occupations.”

import numpy as np
import pandas as pd
import plotly.graph_objects as go

PALETTE = ["#457b9d", "#2a9d8f", "#e9c46a", "#e76f51", "#6d597a", "#f4a261"]
GRAY = "#b8bec6"


def style(fig, title, subtitle, height=480, legend=False):
    fig.update_layout(
        title=dict(text=f"<b>{title}</b><br><span style='font-size:12px;color:gray'>{subtitle}</span>",
                   x=0, xanchor="left"),
        template="simple_white", height=height, showlegend=legend,
        font=dict(family="Segoe UI, Helvetica, Arial, sans-serif", size=13),
        margin=dict(l=10, r=20, t=95, b=50), hoverlabel=dict(bgcolor="white", font_size=13),
        legend=dict(orientation="h", x=0.5, xanchor="center", y=-0.15, yanchor="top"))
    fig.update_xaxes(gridcolor="#eceff3")
    fig.update_yaxes(gridcolor="#eceff3")
    return fig


def show(fig):
    fig.show(config={"displaylogo": False, "responsive": True, "modeBarButtonsToRemove": ["lasso2d", "select2d"]})


quant = pd.read_csv("data/processed/benchmark_quantiles.csv")
fields = pd.read_csv("data/processed/benchmark_fields.csv")
print(f"{int(fields['postings'].sum()):,} valid postings, {int(fields['salary_postings'].sum()):,} with a usable annual salary")

733,064 valid postings, 69,246 with a usable annual salary

Salary Differences Between AI-Related and Non-AI Postings

How do salaries differ across AI-related and non-AI postings?

A raw comparison would mislead, because most non-AI postings are lower-paid jobs such as retail and healthcare support. So the chart shows two views: all jobs, and only the computer and business fields, where the comparison is closer to like with like. Each box covers the middle half of salaries (25th to 75th percentile), the line inside is the median, and the whiskers reach the 5th and 95th percentiles.

scopes = ["All jobs", "Computer and business fields"]
counts = quant.pivot(index="scope", columns="ai_related", values="n").loc[scopes]
x_labels = [f"{sc_}<br>({int(counts.loc[sc_, False]):,} non-AI | {int(counts.loc[sc_, True]):,} AI-related)" for sc_ in scopes]
fig = go.Figure()
for ai, color, label in [(False, GRAY, "Non-AI postings"), (True, PALETTE[3], "AI-related postings")]:
    sub = quant[quant["ai_related"] == ai].set_index("scope").loc[scopes]
    fig.add_trace(go.Box(
        x=x_labels,
        q1=list(sub["p25"]), median=list(sub["median"]), q3=list(sub["p75"]),
        lowerfence=list(sub["p05"]), upperfence=list(sub["p95"]), mean=list(sub["mean"]),
        name=label, marker_color=color, boxpoints=False,
        hovertemplate="<b>%{x}</b><br>Median: $%{median:,.0f}<br>Middle half: $%{q1:,.0f} to $%{q3:,.0f}<br>5th to 95th percentile: $%{lowerfence:,.0f} to $%{upperfence:,.0f}<extra>" + label + "</extra>"))
fig.update_layout(boxmode="group")
fig.update_yaxes(title="Annual salary (USD)", tickprefix="$", tickformat=",.0f", range=[0, 320000])
style(fig, "Salary Spread: AI-Related vs. Non-AI Postings",
      "All industries, 69,246 postings with a usable annual salary | box = middle half, line = median, whiskers = 5th to 95th percentile",
      height=520, legend=True)
show(fig)

Interpretation: AI-related postings pay clearly more. Across all jobs, the median is $139.5K against $54.1K (about 2.6 times). That gap is inflated by job mix, so the fairer comparison is within the computer and business fields: $172K against $97.5K, which is $75K or 77% more, and the middle half of AI-related postings ($123K to $215K) sits mostly above the middle half of non-AI postings ($52K to $150K). The top end is higher too (95th percentile $300K against $227K). This is an association in job postings, not proof that adding AI skills raises anyone’s pay, because AI-related roles also tend to be more senior and more technical.

Data Careers vs. Neighboring Fields

How does pay in data fields compare with software engineering and other computer and business fields?

The bars show each field’s median salary and the diamond shows the median for its AI-related postings (only where there are at least 20 of them). The line through each bar covers the middle half of the field’s salaries. The gray bar is “All other occupations,” the baseline for the rest of the job market.

family = {
    "Data Science & ML": "Data fields", "Data Engineering": "Data fields", "Data Analytics & BI": "Data fields",
    "Software Engineering": "Software engineering", "IT & Cloud Infrastructure": "IT and security",
    "Cybersecurity": "IT and security", "Product & Project Management": "Business fields",
    "Business Analysis & Consulting": "Business fields", "Finance & Accounting": "Business fields",
    "Marketing & Sales": "Business fields", "All other occupations": "All other occupations",
}
family_color = {"Data fields": PALETTE[1], "Software engineering": PALETTE[0], "IT and security": PALETTE[4],
                "Business fields": PALETTE[2], "All other occupations": GRAY}
fq = fields.sort_values("median").reset_index(drop=True)

fig = go.Figure()
for fam, color in family_color.items():
    sub = fq[fq["field"].map(family) == fam]
    fig.add_trace(go.Bar(
        x=sub["median"], y=sub["field"], orientation="h", name=fam, marker_color=color,
        error_x=dict(type="data", symmetric=False, array=list(sub["p75"] - sub["median"]),
                     arrayminus=list(sub["median"] - sub["p25"]), color="#374151", thickness=1.3, width=4),
        customdata=np.stack([sub["salary_postings"], sub["p25"], sub["p75"], sub["postings"]], axis=1),
        text=[f"${v / 1000:,.0f}K" for v in sub["median"]], textposition="inside", insidetextanchor="start",
        textfont=dict(color="#1d2733"),
        hovertemplate="<b>%{y}</b><br>Median: $%{x:,.0f}<br>Middle half: $%{customdata[1]:,.0f} to $%{customdata[2]:,.0f}<br>Postings with salary: %{customdata[0]:,.0f} (of %{customdata[3]:,.0f} postings)<extra></extra>"))
ai_pts = fq[fq["n_ai"] >= 20]
fig.add_trace(go.Scatter(
    x=ai_pts["median_ai"], y=ai_pts["field"], mode="markers", name="Median of AI-related postings",
    marker=dict(symbol="diamond", size=12, color=PALETTE[3], line=dict(color="white", width=1)),
    customdata=ai_pts["n_ai"],
    hovertemplate="<b>%{y}</b><br>AI-related median: $%{x:,.0f}<br>AI-related postings with salary: %{customdata:,.0f}<extra></extra>"))
fig.update_layout(barmode="overlay")
fig.update_yaxes(categoryorder="array", categoryarray=list(fq["field"]))
fig.update_xaxes(title="Median annual salary (USD)", tickprefix="$", tickformat=",.0f", range=[0, 330000])
style(fig, "Median Salary by Occupation Field",
      "All industries | bars = field median, lines = middle half, diamonds = AI-related postings", height=600, legend=True)
fig.update_layout(legend=dict(orientation="h", x=0.5, xanchor="center", y=-0.12, yanchor="top"), margin=dict(b=120))
show(fig)

Interpretation: the data fields are not all equal. Data Science & ML pays the most of any field ($187K median), about $50K above Software Engineering ($137.5K), while Data Engineering ($160K) also beats software engineering. Data Analytics & BI sits lower at $120K, below software engineering and the IT, cybersecurity, and product-management fields ($132K to $143K), though still above business analysis ($109K) and finance ($107.5K). For our pathway, that means moving from Data Analyst toward Data Scientist is the biggest pay step in this set of fields, roughly $67K at the median. The diamonds show where AI exposure adds the most inside a field: Software Engineering ($200K against $128K for non-AI postings, +56%), Cybersecurity (+44%), and Product & Project Management (+31%). In the data fields the gap is small or zero because almost every Data Science & ML posting (91%) is AI-related to begin with, and Data Engineering shows no premium. (Marketing & Sales also shows a large gap, but its AI-related postings are likely technology sales roles, so treat that one with caution.)

Our Own View: Demand vs. Pay

Which fields combine strong pay with a lot of openings?

Pay alone can hide how hard a field is to enter. This chart puts demand (number of postings, on a log scale) against pay (median salary) for every field, with bubble size showing how AI-related the field’s postings are. Fields toward the top are the best paid, fields to the right have the most openings, and the upper-right corner combines both.

fields_b = fields.copy()
fields_b["family"] = fields_b["field"].map(family)
short = {"Data Science & ML": "Data Science & ML", "Data Engineering": "Data Engineering",
         "Data Analytics & BI": "Data Analytics & BI", "Software Engineering": "Software Eng.",
         "IT & Cloud Infrastructure": "IT & Cloud", "Cybersecurity": "Cybersecurity",
         "Product & Project Management": "Product & Project Mgmt", "Business Analysis & Consulting": "Business Analysis",
         "Finance & Accounting": "Finance & Accounting", "Marketing & Sales": "Marketing & Sales",
         "All other occupations": "All other occupations"}

label_pos = {"Business Analysis & Consulting": "middle left", "Finance & Accounting": "bottom center",
             "Software Engineering": "bottom center", "Data Analytics & BI": "bottom center",
             "Cybersecurity": "middle left"}
fig = go.Figure()
for fam, color in family_color.items():
    sub = fields_b[fields_b["family"] == fam]
    fig.add_trace(go.Scatter(
        x=sub["postings"], y=sub["median"], mode="markers+text", name=fam,
        text=sub["field"].map(short), textposition=[label_pos.get(f_, "top center") for f_ in sub["field"]],
        textfont=dict(size=11),
        marker=dict(color=color, opacity=0.85, size=14 + 70 * np.sqrt(sub["ai_share_postings"]), line=dict(color="white", width=1.5)),
        customdata=np.stack([sub["ai_share_postings"] * 100, sub["salary_postings"], sub["remote_or_hybrid_share_stated"] * 100], axis=1),
        hovertemplate="<b>%{text}</b><br>Postings: %{x:,.0f}<br>Median salary: $%{y:,.0f}<br>AI-related share: %{customdata[0]:.0f}%<br>Remote or hybrid (of those that state it): %{customdata[2]:.0f}%<extra></extra>"))
fig.update_xaxes(title="Valid postings (log scale)", type="log", range=[np.log10(1800), np.log10(1000000)],
                 tickvals=[2000, 5000, 10000, 20000, 50000, 100000, 500000],
                 ticktext=["2,000", "5,000", "10,000", "20,000", "50,000", "100,000", "500,000"])
fig.update_yaxes(title="Median annual salary (USD)", tickprefix="$", tickformat=",.0f", range=[20000, 230000])
style(fig, "Demand vs. Pay by Occupation Field",
      "Bubble size = AI-related share of postings | top = better paid, right = more openings", height=620, legend=True)
fig.update_layout(legend=dict(orientation="h", x=0.5, xanchor="center", y=-0.14, yanchor="top"), margin=dict(b=120))
show(fig)

Interpretation: the chart separates fields that are well paid but crowded from fields that are well paid and less crowded. Software Engineering has the most openings of the high-pay fields (20,355 postings at $137.5K), and Product & Project Management (18,590 at $142.5K) and Data Science & ML (13,603 at $187K) follow. Data Science & ML is the standout: the highest pay of any field, a large volume of postings, and by far the biggest bubble (91% AI-related), so it is the field where AI exposure is the norm and not an add-on. Data Engineering (3,720 postings at $160K) and Cybersecurity (3,607 at $132K) are smaller but well paid. Data Analytics & BI (5,652 postings at $120K) is a mid-sized, mid-paid field, which fits its role as the entry step of our pathway. The two big low-pay blocks, Marketing & Sales (56,905 postings at $56K) and all other occupations (567,145 at $52K), show how far the computer and data fields sit above the rest of the market. Roughly 72% to 84% of the postings in the data, software, and IT fields that state a work arrangement are remote or hybrid, against 41% for all other occupations, so flexibility is part of what these fields offer.

What This Means for Our Pathway

• Moving up the pathway pays. The jump from Data Analytics & BI ($120K) to Data Science & ML ($187K) is the largest step among the data and software fields, and Data Engineering ($160K) is a second route to the same level.
• AI exposure is associated with higher pay almost everywhere, most strongly in software engineering, cybersecurity, and product management, and least in fields that are already almost entirely AI-related.

These are patterns across job postings, and they feed the career evaluation on the Predictive Modeling page.

Limitations

• Salary is stated in only about 9% of postings (69,246 of 733,064), and posting pay ranges are not the same as what people are paid. Postings that disclose pay may differ from those that do not.
• Fields come from job-title keywords, so some postings land in the wrong field and a title can hide the real job. The Marketing & Sales and “All other occupations” groups are broad and mix very different jobs.
• “AI-related” means the posting names AI terms, which includes passing mentions. It measures AI exposure and not whether a job is a dedicated AI role.
• This is one dataset, the MET job postings for 2026, and its mix of occupations is not a representative sample of all US jobs.