Thermo Fisher Scientific · 140 UK roles harvested · live AI workspaces● Updated last week
A 125k-staff giant.
Decomposed from public data.
Thermo Fisher Scientific runs $44.6B in revenue across four segments with ~125,000 staff. We harvested every UK job posting, every Companies House filing, every public SEC record — and built live AI workspaces that do the automatable work. 140 real role pages, 134 role archetypes, 126 extracted skills, 18 named systems.
01 · The work
The Director, Capability — decomposed.
From the UK operations substrate: 140 jobs across 3 site archetypes, 11 functional categories, 126 extracted skills.
| Task | Verdict | Why |
|---|---|---|
| Map capability across 3 UK manufacturing+distribution sites | automate | All 140 JDs decomposed into skills/assets/teams. Spine extraction runs in seconds. |
| Identify skills gaps and training needs | assist | AI surfaces the gaps; human validates against business priorities. |
| Workforce planning for site expansion/contraction | human-only | Strategic judgement, budget ownership, stakeholder alignment. |
The spine extraction (extract_spine.py) processes 141 JDs into the canonical capability model: 140 active jobs → 134 roles → 126 skills → 18 enterprise systems. All quoted, all verified.
02 · The system
An agentic backend pointed at the automatable work.
Harvested, not invented
Every claim traced to a specific JD, SEC filing, or Companies House record. Zero synthetic content.
Spine extraction
Regex + keyword pipeline decomposes JDs into canonical skills/assets/roles with verbatim quote provenance.
Data-hygiene gates
Real Phenom categories, geo-classified, test reqs excluded, false-positive detectors fixed.
Grounded Captain
Live RAG agent over the Fisher knowledge organism — answers from the harvest, never hallucinates.
03 · The headcount
The function doesn't go to zero. It changes shape.
117 UK-located roles across 3 sites. Who remains, and why.
| Role | Heads | Why it survives automation |
|---|---|---|
| Capability Director / Head of | 1 | Owns the capability model; accountable for what the AI surfaces. |
| Site operations leads | 3 | Loughborough, Swindon, Paisley — each site has unique GxP/CDMO/distribution context. |
| Quality / regulatory specialists | Retained | GxP sign-off, MHRA inspections, batch release — these are statutory human gates. |
What shrinks: manual JD analysis, skills-gap spreadsheets, system-inventory audits. The spine runs those in seconds.
What's new
- data-hygiene 2026-06-10 ITS-273: Real Phenom categories, geo classification, test-req exclusion, false-positive detector fixes, filter smoke test — all 5 defects fixed.
Sector context
-
Danaher
$24.6B revenue, ~60,000 staff, DBS lean model — the only near-peer at ~55% of TMO size
-
Sartorius
€3.5B revenue, pure-play bioprocessing — direct competitor to Fisher's bioproduction arm
-
Agilent
$6.95B, analytical instruments — competes with Thermo Scientific instrument division
-
Waters
$3.17B, premium LC/MS for pharma QC — Patheon-relevant competitor signal
-
Revvity
$2.86B, diagnostics + life science tools — adjacent to Specialty Diagnostics segment
The provocation
You run a $44B company.
Here's the capability system that reads your public footprint.
What 'know your organisation' looks like when the AI has read every JD, every filing, every entity record.
Get this for your org
Your competitors are Danaher, Agilent, Sartorius. They could have this first.
The Fisher proof surface is from public data alone — your public data reveals your capability gaps to anyone who looks.
Thermo Fisher is the largest company we've decomposed from public data. 141 job postings, 5 Companies House entities, a full SEC 10-K — all readable, all structured, all queriable. The spine runs in seconds. The Captain answers from the harvest. Your public footprint is your capability intelligence — whether you use it or not.