What this is
A live signal of what AI-leadership roles in DACH and CEE actually need — derived from the language hiring teams use in their own job descriptions. Each role is decomposed into pain signals mapped to a 7-category taxonomy (Data, Strategy, Execution, Org, Ops, Change, Risk).
How it's built
- Sources: public job listings — Greenhouse, Lever, Ashby JSON APIs, plus LinkedIn search results. No login-walled content, no DOM scraping.
- Extraction: each JD is read against a fixed taxonomy of 27 pain codes (A1–G3,
v1.0). The model returns evidence quotes for explicit pains and a documented inference chain for implicit ones — both visible per card.
- Anti-hallucination: implicit pains are rejected without an inference chain; confidence below 0.5 is dropped. "If not stated, return null. Never invent."
- ICP score: per-job match against a consulting ICP profile (industry, maturity, role intent, pain mix).
Why pain-mining
JDs are an underused public market signal. They reveal what executives are willing to pay to fix — in their own words, dated, attributable to a specific company and stage. Aggregated, they show which problems are spreading and which are isolated.
Reading the cards
ex — explicit pain (quoted from the JD)im — implicit pain (inferred; chain shown after the arrow)- Confidence is per-signal, not per-job
- Archetypes map pain clusters to consulting offer shapes
Disclaimer. Pain interpretations are inferred from public job-description text. They are not statements of fact about the named companies, their strategy, or their operations. Inferences may be wrong; evidence quotes are provided so you can judge for yourself. Snapshot reflects the dataset as of the export date in the footer — JDs evolve, signals decay.
Measured error rate. Extraction quality is measured, not assumed. A hand-audited random sample of 20 JDs (Jul 2026) scored the original extractor at recall 0.89, precision 0.55 — it misses little but over-fires, so headline counts from this corpus are upper bounds, not point estimates. Per-code precision varies widely (best-validated: manual-ops bottleneck, no-central-AI-strategy, regulatory exposure, model-governance gap; weakest: cross-functional silos). The extraction rules were then tightened against the measured failure patterns and re-audited on 20 fresh JDs with independent judges: precision 0.89 (latest batch 0.96), recall 0.87 — new extractions run under these rules; legacy counts keep the upper-bound caveat.