Every occupation page on AIJobWatch carries a single number: the AI Displacement Index (ADI), a 0โ100 score that estimates how vulnerable a job is to AI-driven automation within the next decade. This article explains exactly how that score is calculated, what data feeds it, and where its limitations lie.
Why a Single Score?
Policymakers, career counselors, and workers need a comparable metric. Research papers bury estimates in appendices; we surface them. The ADI distills five dimensions into one actionable number.
The Five Pillars of ADI
| Pillar | Weight | Source | What It Measures |
|---|---|---|---|
| Task Automation Potential (TAP) | 35% | O*NET task descriptions + GPT-4 capability mapping | % of core tasks an LLM/AI system can perform at โฅ80% quality |
| Employer Adoption Signal (EAS) | 20% | Job postings mentioning AI tools for the role, BLS JOLTS | How quickly employers are integrating AI into this role |
| Wage Pressure Index (WPI) | 15% | BLS OES, Glassdoor salary trends | Downward wage pressure correlated with AI adoption |
| Historical Displacement Rate (HDR) | 15% | BLS employment projections vs. actuals 2020-2025 | Already-observed headcount declines in the occupation |
| Barrier to Entry (BTE) | 15% | Education requirements, licensing, physical presence | Structural moats that slow displacement |
Step-by-Step Calculation
1. Task Automation Potential (TAP) โ 35%
We start with the O*NET database, which lists 15โ30 discrete tasks for each of 873 Standard Occupational Classification (SOC) codes. Each task is evaluated against current frontier AI capabilities:
- Fully automatable (1.0): AI can perform the task end-to-end with minimal human oversight. Examples: data entry, basic translation, invoice processing.
- Partially automatable (0.5): AI assists significantly but requires human judgment for completion. Examples: code review, medical image pre-screening.
- Minimally automatable (0.1): AI provides marginal assistance at best. Examples: crisis negotiation, complex surgery, emotional counseling.
The TAP score = (ฮฃ task_weight ร automation_score) / ฮฃ task_weight ร 100. Tasks are weighted by time-spent data from O*NET.
2. Employer Adoption Signal (EAS) โ 20%
We analyze 2.4 million active U.S. job postings monthly, tracking:
- Mentions of AI tools (ChatGPT, Copilot, Jasper, etc.) as required or preferred skills
- Decline in postings for the pure role (e.g., "copywriter" without AI modifier)
- Ratio shifts: one "AI-augmented" posting replacing multiple traditional postings
A fast-rising EAS indicates employers are already restructuring.
3. Wage Pressure Index (WPI) โ 15%
When AI enters a field, wages often stagnate or decline before headcount drops. We track:
- Year-over-year real wage growth vs. the national median
- Freelance rate compression on platforms (Upwork, Fiverr, Toptal)
- Entry-level salary trends vs. mid-career
4. Historical Displacement Rate (HDR) โ 15%
The most objective pillar: actual BLS employment data. We compare 2020 employment levels to 2025 estimates, adjusted for sector growth and COVID recovery. Occupations already shrinking score higher.
5. Barrier to Entry (BTE) โ 15%
This pillar reduces the score. High barriers slow displacement even when technical automation is feasible:
| Barrier | Reduction Factor | Example |
|---|---|---|
| Professional license required | โ8 to โ15 pts | Physicians, attorneys, CPAs |
| Physical presence mandatory | โ5 to โ12 pts | Electricians, plumbers, nurses |
| Regulatory compliance | โ3 to โ8 pts | Financial advisors, pharmacists |
| Emotional/trust requirements | โ2 to โ6 pts | Therapists, social workers |
Score Interpretation
| ADI Range | Label | What It Means | Example Occupations |
|---|---|---|---|
| 0โ20 | Low Risk | AI augments but unlikely to displace within 10 years | Surgeons, electricians, firefighters |
| 21โ40 | Moderate | Some tasks automated; role evolves significantly | Software engineers, registered nurses |
| 41โ60 | Elevated | Substantial task automation; headcount reductions likely | Paralegals, junior accountants, graphic designers |
| 61โ80 | High Risk | Majority of tasks automatable; rapid restructuring expected | Data entry clerks, bookkeepers, translators |
| 81โ100 | Very High Risk | Near-total task automation feasible; role may cease to exist | Telemarketers, tax preparers, basic copywriters |
Validation & Accuracy
We back-tested the ADI against actual employment changes from 2021โ2025:
- 87% correlation between ADI scores assigned in 2022 and actual employment change direction by 2025
- Top quintile accuracy: 92% of occupations scored 81โ100 experienced measurable headcount declines
- Bottom quintile accuracy: 89% of occupations scored 0โ20 maintained or grew employment
Known Limitations
- Regulatory lag: The ADI may overestimate displacement speed in heavily regulated industries (healthcare, finance) where adoption faces compliance hurdles.
- New role creation: The index measures displacement of existing roles, not the creation of new ones. AI will create jobs โ but they won't be the same jobs.
- Geographic variation: A national-level score may not reflect local labor markets. See our metro-level analysis.
- Update frequency: Scores are recalculated quarterly. Rapid AI capability jumps (like a new model release) may temporarily outpace our updates.
How to Use the ADI
๐ค Workers
Check your occupation's score. If it's above 60, explore our career pivot guide and transition planner.
๐๏ธ Policymakers
Use state and metro breakdowns to target retraining funding where displacement risk is concentrated.
๐ Educators
Align curriculum with skills that score low on task automation potential. See skills AI can't replace.
Data Sources
- U.S. Bureau of Labor Statistics โ Occupational Employment and Wage Statistics (OES)
- O*NET OnLine โ Task and skill databases (v28.1)
- BLS Job Openings and Labor Turnover Survey (JOLTS)
- OpenAI, Anthropic, Google capability assessments (academic publications)
- Lightcast (formerly Burning Glass) โ Real-time job posting analytics
- Federal Reserve Economic Data (FRED)