Methodology v1.0 ยท April 2026

How we score AIโ€™s impact on 50 roles.

Rule-based, deterministic, and open. No LLM in the scoring path. The same inputs always produce the same risk score. Reviewed by an external labour economist at every major version.

Step 01 ยท The five dimensions

Five dimensions per role, scored 0โ€“100.

Every role in the catalogue is scored on five dimensions, calibrated against O*NET task data and external research. The dimensions combine into a base risk score before industry and geography overlays are applied.

  • task_automation

    Task automation

    Share of role tasks current and projected AI can do without human supervision. Calibrated against O*NET task taxonomies.

  • augmentation_potential

    Augmentation potential

    Share of role tasks where AI yields 2โ€“5ร— productivity for the human still in the loop.

  • judgment_requirement

    Judgment requirement

    Share of tasks requiring ethics, fiduciary duty, or durable customer relationships.

  • physical_requirement

    Physical requirement

    Share of tasks requiring physical presence or embodied dexterity in the workplace.

  • regulatory_barrier

    Regulatory barrier

    Licensure, certification, or legal protection that slows AI substitution.

Step 02 ยท Formulas

The math is short. We publish all of it.

Risk score (0โ€“100)

riskScore = clamp(
  round(automation - 0.45 * judgment - 0.35 * physical - 0.25 * regulatory + 35),
  0,
  100
)

Classification

if (judgment > 70 || physical > 70 || regulatory > 70)
  โ†’ UNCHANGED
else if (automation >= 60 && judgment < 40 && physical < 30)
  โ†’ REPLACE
else if (augmentation > 50)
  โ†’ AUGMENT
else
  โ†’ UNCHANGED

Year of major impact

yearOfMajorImpact = baseYear
  + clamp(
      round(judgment / 30 + regulatory / 40 + (4 - adoption_pace[country])),
      1, 8
    )

Confidence

confidence = clamp(
  round(90 - dataAgeMonths - industryGap - regionalGap),
  0, 100
)
label = score >= 75 ? "High" : score >= 50 ? "Medium" : "Lower"

Step 03 ยท Overlays

Industry context and country adoption pace.

After the base score, we multiply by per-industry ร— per-role factors (e.g. healthcare ร— 0.5 on doctor) and per-country adoption-pace factors (e.g. KR ร— 1.0 vs JP ร— 0.85). Defaults are 1.0; we only override where research evidence supports a directional pull.

Lead score (Path B only)

Submissions are also lead-scored for routing

Weights: exec role +30, HR +20, 200+ employees +20, ICP industry +15, 1-year horizon +10, AI-mature stack +10, โ‰ฅ20% replace ratio +15. Tier A โ‰ฅ 70. Lead-score rules are documented here so it is never a black box.

Step 04 ยท Data sources

Where the numbers come from.

  • O*NET (US Department of Labor)

    Task taxonomies, SOC codes, and skill-occupation mappings for every Phase 1 role.

    Open source

  • Goldman Sachs โ€” Generative AI economic impact

    Macro estimate of jobs exposed to GenAI augmentation; informs adoption-pace priors.

    Open source

  • McKinsey Global Institute โ€” Generative AI

    Per-task automation curves and productivity uplift envelopes by sector.

    Open source

  • Brookings โ€” AI exposure index

    Geographic exposure index informing per-country adoption pace.

    Open source

  • WEF Future of Jobs (2025)

    Global employer survey on hiring intent, skill shifts, and AI adoption.

    Open source

Versioning & refresh

v1.0 โ€” April 2026

Refreshed quarterly. The refresh log lives on this page; major version increments are reviewed by an external labour economist before publication.

Disclaimers

Workforce planning, not termination.

Risk scores are directional, never personnel decisions. Confidence is shown on every result. This is workforce planning + augmentation guidance, not legal or financial advice.

Ready when you are

Run the calculator on a role or your roster.

60 seconds for a single role. Five minutes for a full company analysis with a board-ready PDF.

Open the tool