Crystal Ball: How We Track AI Displacement Data Across 21 Industries
Every Protocol recommendation and every Oracle reading stands on the same foundation: the Crystal Ball Intelligence Database — currently1,822 documented evidence entries covering506 companies across 21 industries, generation date2026-04-05. This post explains what is actually inside it, how an exposure score is computed, and — just as important — what the data cannot tell you.
An evidence entry is a sourced, dated record of something a real organization actually did or stated about AI and its workforce — a deployment, a restructuring, an executive statement — never a prediction someone merely published.
whatsmyedge, Crystal Ball Intelligence DatabaseWhat one company record contains
Each of the 506 tracked companies carries a structured record: the business functions it runs (HR, finance and accounting, customer service, IT operations…), its documented evidence highlights with sources, a headline signal — the single strongest public statement about its AI trajectory — and two derived values: an exposure score from 0 to 100 and a risk tier from 1 to 5. Nothing in the derived values exists without the underlying evidence lines; that is the design rule that keeps the database honest.
How the exposure score works
The score estimates how much of an organization's core work could be performed or significantly augmented by AI, based on industry data, company signals, and publicly reported automation initiatives. The bands read like this:
- 0–20 — Strategic position: AI is an opportunity, not a threat.
- 21–40 — Moderate exposure: some functions face automation; early action is an advantage.
- 41–60 — Significant exposure: multiple job functions are actively being automated in the sector.
- 61–80 — High exposure: core tasks are in the automation corridor; the tools exist today.
- 81–100 — Critical exposure: displacement is already underway for similar roles and profiles.
In the current generation, 35 of the 506 companies sit in the top two bands. That distribution is itself a finding: documented displacement is concentrated, not universal — which is why we publish the full per-industry table instead of a scary headline number.
How the data reaches you
Three surfaces consume the same evidence base. The public Market Intelligence page computes its industry table directly from the dataset at build time — the numbers you see there are derived, not typed in. The free Oracle reading benchmarks your role against a companion sector evidence base covering 15 sectors. And the paid Protocol uses the evidence to calibrate urgency: someone in a critical-exposure sector gets a different day one than someone in a strategic-position one. The full pipeline — scoring dimensions, quality gate, generation — is documented on the methodology page.
What we deliberately do not claim
Curated evidence has limits and pretending otherwise would defeat the point. The database over-represents organizations that make public announcements — quiet automators are undercounted. Exposure describes what is automatable, not what any given company will decide to do about specific people. And company-level scores say nothing about you personally: your exposure lives in your task mix, which is why the entry point to everything we sell is a free assessment of your own role rather than a subscription to sector doom.
Frequently asked questions
Where does the AI displacement data come from?
From public, documented signals: company announcements, earnings-call statements, workforce filings, deployment reports, and industry research. Every evidence entry records its source. What the database deliberately excludes is vibes — anonymous predictions, survey sentiment, and unsourced "X% of jobs will vanish" claims.
How often is the evidence base updated?
In curated waves rather than a continuous crawl. Each wave adds and re-scores companies for a segment, and the dataset ships with its generation date so you can see exactly how fresh the numbers you are reading are. The current public generation is from 2026-04-05. When a new wave lands, every page computed from the dataset — including this one — is rebuilt from it.
Can the database tell me if my own job is at risk?
Not by itself — and we think honesty about that matters. The database scores companies and sectors from documented evidence; your personal exposure depends on your specific task mix, tools, and role, which no industry average captures. That is what the free 7-question assessment measures, and the Oracle reading then maps your result against the sector evidence.