Most workforce tools are black boxes. Ours is not. This page documents the full methodology: the data graph underneath, the four pillars of the score, the formula that combines them, and the limitations we're candid about. If you're a CHRO evaluating a vendor, you should be able to read this page and know exactly what you'd be buying.
Every score we produce starts with a normalized graph. We ingest six public or commercially-licensable data sources and project them into a single graph where nodes are occupations, tasks, and skills, and edges carry weights for transferability, exposure, and adjacency.
O*NET provides the occupational backbone. ESCO adds the multilingual skills layer. Lightcast brings real-time job posting signal. The Anthropic Economic Index provides actual task-level AI adoption evidence at scale. WEF supplies macro context. BLS grounds the wage and employment base in public record.
The graph is where the crosswalks live. O*NET to ISCO to ESCO to Lightcast. Wage bands from BLS joined to role nodes. AEI task usage patterns joined to task nodes. That's why one-source products are weaker than this: a skills taxonomy alone can't tell you about exposure, and exposure data alone can't tell you where to route.
For each task in your role (from O*NET), we compute the share of that task's observed execution that already shows automation or augmentation patterns in the Anthropic Economic Index. Higher exposure means higher score, which means higher risk.
Per-skill durability based on Lightcast posting trend direction and ESCO skill category. Skills losing demand in postings compound the score. Skills gaining demand mitigate it. Stakeholder communication trends up, report drafting trends down.
BLS employment projections for your occupation family and wage band direction over the last 8 quarters, cross-checked against WEF macro signals. A role with shrinking employment and flat wages gets a worse trajectory score than one growing in both.
How many adjacent roles you can transition to with limited retraining, weighted inversely by their own exposure. A worker with many low-exposure escape hatches has a better score than one with few high-exposure ones. This is the optimism pillar.
We deliberately do not use an opaque neural network to produce the top-line score. The four pillars are weighted and combined in a single transparent formula so that any visitor, CHRO, or auditor can reproduce the math and challenge the weights.
The neural work happens underneath the pillars, in the graph enrichment layer. The top-line score stays legible.
The exposure pillar uses Anthropic's anonymized analysis of millions of real Claude conversations mapped to O*NET tasks. This is evidence of how AI is actually used, not how vendors claim it will be used. That's a meaningful upgrade from consultant narratives.
AI capability is advancing faster than any data source can refresh. The score is best understood as a well-grounded estimate of near-term displacement risk, not a long-horizon forecast. It cannot predict specific layoffs, capture firm-specific dynamics, or replace a conversation with your manager.
Transparency on freshness matters as much as transparency on method. Different inputs refresh at different cadences. The score inherits the slowest one.
Skill posting trends, real-time demand signal, new skill emergence.
Task-level usage patterns. New data drops accompany each AEI report release.
Occupation taxonomy updates, wage bands, employment projections.
We'd rather answer the hard questions before the procurement cycle than during it. If you're a CHRO, workforce analyst, or researcher with a specific challenge to our approach, we want to hear it.