AITakeover Tracker
Full Transparency

How We Calculate Risk

A two-tier hybrid index blending hard quantitative data with AI-classified signals. Every score is reproducible. No black boxes.

The Doomsday Clock

Score-to-clock mapping: 0% = 12 o'clock, 100% = midnight

0%25%50%75%STARTEarly SignalsAcceleratingRestructuringPoint of No ReturnWe are here36.0%AI Displacement Index

Reading the Clock

The clock hand moves clockwise from 12 o'clock (0%) toward midnight (100%). Each quarter represents a phase of AI displacement:

0-25%Early Signals
25-50%Accelerating
50-75%Restructuring
75-100%Point of No Return

Embed the clock on your site: <iframe src="https://takeovertracker.com/embed/clock" width="400" height="450"></iframe>

Part 1: Daily Index

Two-Tier Hybrid Index

Hard data (50%) + AI-classified signals (50%)

Pipeline Flow

Tier 1 Hard Data50%
+
Tier 2 Signals50%
Blend50/50
Hype Discount
EMA Smoothα=0.08
Round 0.5
Final Score0-100

Tier 1 — Hard Data Anchors (50%)

Quantitative economic data is z-score normalized against trailing historical windows, then converted to a 0-100 scale: score = clamp(50 + z × 15, 0, 100)

BLS Monthly Employment35%

Month-over-month change in NAICS 51 (Information), 54 (Professional/Scientific/Technical), 71 (Arts/Entertainment). Z-score normalized against trailing 12 months. Employment drops push the score up.

FRED Initial Claims25%

Weekly initial unemployment claims (ICSA). Z-score against trailing 52 weeks. Higher claims = higher displacement signal.

Job Posting Trends20%

Indeed Hiring Lab data + displacement keyword counts from collected signals. Z-score against 90-day window.

Corporate Layoff Count20%

WARN Act firehose employee count (70%) + GDELT displacement article count (30%). Z-score against 90-day window.

tier1 = bls × 0.35 + fred × 0.25 + posting × 0.20 + layoff × 0.20

Tier 2 — Categorical Signal Classification (50%)

Each signal is classified into a 5-level displacement scale by AI, then aggregated mathematically (no LLM scoring calls).

5-Level Displacement Scale

Each signal is classified into one of five levels

L0
No SignalNo displacement evidence
0
L1
SpeculationDiscussion or speculation only
15
L2
Plans/PilotsConcrete plans or pilots announced
35
L3
MeasurableMeasurable displacement occurring
60
L4
StructuralStructural/permanent displacement
85

Signal Category Weights

How each data source type contributes to Tier 2

Labor market dataLagging

BLS employment, job postings, layoff data — most reliable ground truth

25%
Corporate adoptionCoincident

Earnings calls, enterprise AI spending, hiring patterns

25%
AI capability benchmarksLeading

Model benchmarks, task performance — hype-discounted

15%
Economic indicatorsCoincident

GDP per worker, productivity data, sector output

15%
Sentiment & hypeLeading

Social media, news volume — heavily discounted

10%
Regulatory signalsModifying

AI legislation, executive orders, industry standards

10%

Job Sector Weights

30%

Knowledge

Legal, finance, consulting

25%

Service

Support, sales, admin

25%

Technical

Software, data, IT

20%

Creative

Writing, design, marketing

Composite Math

Two-Tier Blend

Tier 1 hard data (50%) and Tier 2 categorical signals (50%) are combined via simple arithmetic mean. Neither tier can dominate alone.

Hype Discount

Tier 1 hard data serves as the reality anchor. When hype signals exceed reality, the raw blend is discounted to keep the index grounded.

Single EMA Smoothing

A single exponential moving average (alpha = 0.08, ~25-day window) filters noise while responding to genuine trends faster than the old dual-EMA.

0.5 Rounding

The final score is rounded to the nearest 0.5 percentage point. This replaces the old movement cap, providing stability without artificial constraints.

What Changed from v1

×
Geometric MeanArithmetic 50/50 blend
×
72 Gemini scoring callsMath aggregation (~70% cost reduction)
×
Dual-EMA (slow + fast)Single EMA (α=0.08, ~25-day window)
×
2-point movement cap0.5 rounding

Source Credibility Hierarchy

1.0xPeer-reviewed research, BLS/government data
0.8xIndustry surveys, consulting firm reports
0.5xCEO predictions, company announcements
0.3xSocial media, opinion pieces

Phase Thresholds

Each score level maps to observable labor market conditions

0%Pre-AI baseline (~2019)
10%Early tools — AI assists with drafts, code suggestions
20%Established assistance — entry-level task automation beginning
25%Task transformation — AI handles 25-30% of routine cognitive tasks
40%Role compression — teams doing more with fewer people
50%Role restructuring — most firms halved entry-level hiring
60%Human premium emerging — standard knowledge work is AI-first
75%Human premium dominant — humans add value only through creativity and judgment
90%Near-complete automation — minimal human oversight
100%Economic non-viability — human labor offers no cost advantage

Data Sources

Organized by tier

Tier 1 (Hard Data)

FRED

Federal Reserve Economic Data — initial claims, employment

BLS

Bureau of Labor Statistics — monthly employment by NAICS sector

WARN Firehose

Worker Adjustment and Retraining Notification Act filings

Indeed Hiring Lab

Job posting trends and AI hiring data

Tier 2 (Signal Classification)

25 RSS Feeds

DeepMind, HuggingFace, Meta AI, Import AI, IEEE Spectrum, BAIR, Stratechery, FT, Brookings, SHRM, HR Dive, and more

GDELT

Global Database of Events — displacement-related news articles

Reddit & Hacker News

Community sentiment on AI and job displacement

SEC EDGAR

Corporate filings mentioning AI workforce changes

ArtificialAnalysis.ai

AI model benchmarks and capability tracking

SerpAPI

Google Trends for displacement-related search terms

Part 2

Per-Occupation Risk Scoring

Individual displacement scores for 1,100+ occupations using O*NET task data

Task Risk Formula

Every task in an occupation is classified into one of five categories and scored for AI capability:

task_risk = base_risk × 0.6 + ai_capability_score × 0.4

Tasks are weighted by estimated time fraction to produce the raw score:

raw_task_score = Σ(task_risk × time_fraction)

Task Categories & Base Risk

Higher base risk = more automatable task type

Routine Cognitive82/100
Routine Manual55/100
Non-Routine Analytical45/100
Non-Routine Interpersonal25/100
Non-Routine Manual15/100

Protective Factors

Five factors can reduce the raw score by up to 55%:

final_score = raw_task_score × (1 - protective_discount)
Social Intelligence15%

Empathy, negotiation, reading social cues

Creativity12%

Novel ideation, artistic expression, innovation

Decision Complexity10%

Ambiguous judgment, ethical reasoning, strategic calls

Regulatory Barriers10%

Licensing, legal requirements, safety standards

Fine Manipulation8%

Dexterous physical tasks, precision work

Risk Tiers

75 – 100

Critical Risk

50 – 74

High Risk

25 – 49

Medium Risk

0 – 24

Low Risk

See It In Action

Explore the data yourself

Search 1,100+ occupations, compare risk scores, and see exactly how each job is scored.