Methodology · 10 min read
Compression Exposure Index (CEI) Explained
Methodology
Deterministic. Reproducible. Auditable.
The AI Career Index does not guess. It measures. Here is exactly how.
Methodology v3 · 2026
How the framework works
Structural scoring, not predictive modelling
The AI Career Index reads the structure of work. It does not forecast when automation will arrive. It runs two layers. Your personal assessment turns your answers into a score using a fixed weighting of five dimensions. The role pages classify each occupation into a low, medium, or high exposure band using one consistent structural rubric. Both are deterministic: the same inputs always produce the same result, with no randomness, no black-box reasoning, and no LLM deciding the outcome.
That is deliberate. Forecasts depend on adoption speed and company decisions nobody can predict. A structural reading depends on the work itself, which is stable enough to build a 24-month plan around. The five dimensions and their exact weights are below, followed by how the role pages are scored.
Five Principles
The model's design constraints
01
Deterministic, not probabilistic
Every input maps to a number. Every number maps to a weight. Every weight produces a score. There is no AI guesswork inside the model and no probabilistic interpretation of your answers. Two professionals with identical inputs always produce identical outputs.
02
Structural, not behavioural
The framework measures the configuration of your work, not the configuration of you. Personality, temperament, and disposition are explicitly excluded from the model. The AI economy changes pay for roles, and the model measures roles. See why personality tests fail in the AI economy.
03
Reproducible by design
Reproducibility is the property that separates measurement from interpretation. The same answers always produce the same score. Updates to the engine are versioned and disclosed. Your reading is auditable.
04
Calibrated to current market conditions
The compression curves and migration probabilities inside the model are calibrated against observed market conditions, not against speculation about long-horizon AGI scenarios. The framework is deliberately scoped to the next 24 months.
05
Action-oriented output
The model produces a structural reading and a sequenced intervention plan, not a personality profile or a list of suggested careers. The output is built to be acted on, not contemplated. See the 6-month Acceleration Plan.
06
Honest, not encouraging
The model's job is to produce an accurate structural reading, not a comforting one. Scoring does not adjust based on how a user answers earlier questions. Compression risk is not softened. Authority bands are not inflated. The reading is calibrated to the market, not to the reader.
The Method
Six steps from input to plan
1
Structured input collection
The assessment collects 35 structured questions across five weighted dimensions: Strategic Clarity, AI Adaptability, Income Leverage, Skill Resilience, and Market Alignment. Questions are deliberately structural. They ask about the configuration of your work, not how you feel about it.
2
Dimension scoring
Each answer maps to a 0 to 100 dimension score using fixed weights. Dimension weights are determined by the relative impact of each dimension on structural durability and authority migration in current market conditions, not by user customisation.
3
Composite index calculation
The five dimension scores combine into a composite AI Career Index score on a 0 to 1,000 scale. The composite is not a simple average. It applies the dimension weights and produces a scaled reading that is comparable across users.
4
Derived metric generation
From the dimension scores and the composite, the model derives the Compression Exposure Index, the Authority Band, the Structural Leverage Index, the Authority Migration Probability, and the Structural Archetype. These derivations are deterministic and traceable.
5
Projection modelling
The 24-month projection compares two structural paths: the Drift Path (no intervention) and the Strategic Path (Acceleration Plan executed). Both paths are calibrated against the user's current readings and the structural pressure on their archetype.
6
Acceleration Plan generation
The 6-month plan is built around the user's weakest dimension and is sequenced into three structural phases: foundation, migration, and consolidation. Every action in the plan ties to a measurable structural outcome.
Outputs
What the model produces
Every reading is a deterministic function of your inputs. The eight outputs below are what get generated from a completed assessment.
0 to 1,000
AI Career Index score
The composite reading. A single number comparable across users.
Read explainer →
0 to 100 · CEI
Compression Exposure Index
How exposed your role is to AI-driven pay shifts. Higher is more exposed.
Read explainer →
0 to 1.00 · SLI
Structural Leverage Index
The ratio of judgment-bearing work to routine production in your role.
Read explainer →
4 bands · Execution to Authority
Authority Band
The structural altitude at which your role operates, independent of job title.
Read explainer →
12-month window · AMP+
Authority Migration Probability
The probability of moving one authority band upward in the next year.
Read explainer →
1 of 6 postures
Structural Archetype
Your posture inside the AI economy, mapped from your dimension profile.
Read explainer →
Drift vs Strategic paths
24-month Projection
Two parallel trajectories showing your position under inaction versus plan execution.
Read explainer →
Sequenced 3-phase plan
6-month Acceleration Plan
Your weakest dimension as the anchor. Foundation, migration, consolidation.
Read explainer →
Reviewed by AI Career Index Research · Methodology v3 · 2026 ·See the framework →
Role-Page Scoring
How the role pages are scored
The personal assessment scoring above is one layer. The 1,000+ role pages on the site sit on a separate, role-level scoring layer described here.
Each role maps to a specific U.S. SOC code. Its page combines two public datasets, O*NET (what the work is) and BLS OEWS (what it pays and how many people do it), with our structural exposure classification, and it shows one real-world adoption signal from the Anthropic Economic Index alongside the reading. Every source below is public and citable.
O*NET 30.2: Task and skill structure
The U.S. Department of Labor's O*NET database supplies the descriptive occupation data shown on every role page: the task statements, work activities, skills, knowledge, and the SOC code itself, in standards-grade language. It is also the reference we classify each role's task mix against (the automatable-versus-durable split and the "tasks AI can do" figure). Licence: CC BY 4.0. Release: O*NET 30.2.
BLS OEWS (May 2025): Wages and employment
The U.S. Bureau of Labor Statistics Occupational Employment and Wage Statistics file supplies the wage and employment baselines per SOC: the median annual wage, the junior and senior wage bands (percentiles), total US employment, and the economic-impact figure on each role page (median wage × employment). Published each May; the site is on the May 2025 release and refreshes to the May 2026 release around mid-May 2027.
Structural classification: the exposure band
The headline reading on each role page, its low, medium, or high exposure band, is a structural classification, not a figure calculated from the datasets above. Each occupation is read against one fixed rubric: high when current AI tooling can do the core production work directly, medium when the role is split (part of the work compresses while judgement, relationships, or oversight hold), and low when it is insulated by judgement, accountability, licensure, or in-person physical presence that AI cannot assume. The rubric is applied the same way for every role and recorded per role, so the same role always carries the same band, and that band drives the 0 to 100 exposure score, the task-mix split, and the category ranking shown on the page.
It is transparent and auditable, with no black-box model and no LLM assigning the result. We classify this by hand rather than fitting a formula to wages or job titles on purpose: for AI specifically, high-paid cognitive work is often more exposed, not less, so a wage-driven or seniority-driven formula would get many roles backwards.
Anthropic Economic Index: real-world adoption signal
The Anthropic Economic Index (CC BY 4.0) samples conversations on Claude.ai mapped to O*NET task structures, measuring the share of work in each occupation already showing AI usage. We show this per-role figure as a real-world signal on each role page, next to the band but deliberately never blended into the band or the exposure score. A 0.0% value is a real reading (no measurable usage observed for that SOC at sample time), not a placeholder; specialised medical, legal, and trade roles are common zeros because practitioners use domain-specific tools rather than general-purpose chat assistants.
Anthropic publishes other cuts of this index that do not feed the role scoring: a geographic cut powers our AI Adoption by Country page, and a developer-API cut powers our Fastest-Automating Tasks page.
Every role page cites its O*NET task source, its BLS wage source, and its Anthropic adoption figure inline, so each number traces back to a public dataset. The exposure band itself is a structural classification applied by the same rubric for every role, which is why the same role always reads the same way.
Sources: O*NET 30.2 (USDOL/ETA, CC BY 4.0) · BLS OEWS, May 2025 · Anthropic Economic Index (CC BY 4.0)
Model Integrity
What the model is not
The AI Career Index is not a personality test, not a coaching tool, not a job-matching service, not a prediction engine, and not a generative-AI advice generator. It does not produce different output for the same input. It does not soften its readings to be encouraging. It does not customise scoring to user preference.
What it is, instead, is a structural measurement instrument. The same property that makes it useful (reproducibility) is also what disqualifies it from being a substitute for the human-layer support that a coach or therapist or mentor provides. Those tools are complementary, not competing.
View the full glossary →FAQ
Common questions
Articles in this guide
Go deeper on each part of the pipeline
12 articles, grouped by which step of the methodology they expand on. Click a step to see its articles.
Further Reading
Related articles from the index
Framework · 10 min read
Authority Bands: Title vs Leverage
Two professionals with the same job title can have very different authority. The Authority Band measures the structural altitude at which you actually operate.
Strategy · 11 min read
The 6-Month Acceleration Plan: A Structural Reset
The Acceleration Plan is not a list of self-improvement goals. It is a sequenced 6-month structural intervention that targets your weakest dimension first.
"My category label on the index was completely accurate and completely uncomfortable. I had been drifting for two years without realising it. The 6-month plan broke the drift into four specific moves. I am three months in and the difference in the quality of the work I get offered is already visible."
TY
Takeshi Yamamoto
Product Marketing Lead · Tokyo, Japan
"Finance is supposed to be safe from AI because of regulation. The report pushed back on that complacency by showing which parts of my role were substitutable regardless of sector. That precision is rare in career content. I trust the framework because it refused to tell me what I wanted to hear."
OB
Olivia Bennett
Finance Business Partner · Manchester, UK
"The projection chart was the most honest piece of career analytics I have seen. It did not promise me growth. It showed me the probability weighted outcomes of two clear strategies and let me pick. I have recommended this to every senior engineer on my team. Nine of them have now taken it."
MH
Miguel Hernandez
Director of Engineering · Barcelona, Spain
Measure Your Position with AI Career Index
Take the AI Career Snapshot to see your structural leverage, automation exposure, and growth trajectory with AI Career Index.