CHAPTER 5 · THE WORKED EXAMPLE
HELIX, in eight beats
Brussels publishes a PFAS restriction Tuesday 09:00. The old way: a quarter. The new way: 7 weeks to A-sample.
SITUATION TODAY
Why HELIX is the right wedge programme
HELIX is Avery's PFAS-free release liner programme. It is currently at-risk on the peel curve, has a hard EU regulatory deadline (ESPR + REACH PFAS restriction), has the right scientist owner (Dr Stefan Koen · Leiden), and a $42M revenue exposure across 3 SKUs / 2 customer programmes. A successful HELIX launch on the new stack is the proof point that funds the next two programmes.
$42M
Revenue at risk
3 SKUs · 2 customer programmes
Q1 2027
Regulatory deadline
EU PFAS restriction enforced
21
Prior PFAS-free experiments in PfS
Cross-site · 2022–2026
Wacker REL-SIL-204
Closest supplier candidate
Already in AD AVL
Borderline @168h/70°C
Current peel-curve status
v1.3 fails ageing
BENEFIT
14 weeks → 7 weeks · $42M secured · 0.78 first-try probability
Same lab. Same scientist. The AI layer is the difference.
PROJECT HELIX · 8 STEPS · 7 WEEKS END-TO-END
1
SCIKIQ
Trigger
00:30
EU PFAS Rule 2026/812 → 3 SKUs · 2 programmes · $42M revenue at risk
▣ Regulatory feed scraper + entity match
→
2
SCIKIQ
Retrieve
Day 1
18 prior PFAS-free experiments surfaced · 1 Wacker datasheet · citations inline
▣ RAG (hybrid: BM25 + dense) over PfS + patents + PDFs
→
3
SCIKIQ
Next Best Experiment
Day 1
95% REL-SIL-204 + 5% biobased rosin · P(meet target) = 0.78
▣ Bayesian opt + Gaussian Process · 21-run prior
→
4
Thermo
PfS execution
Day 9
EXP-HELIX-0188 · 19.4 g/in PASS · 21 CFR 11 audited
▣ Core LIMS + ELN + SDMS (Thermo)
→
5
SCIKIQ
Enrich
Day 9
Joined to supplier ESG + LCA · sustainability 84/100 · written back to PfS
▣ KG update + UDF write-back via OData v4
→
6
SCIKIQ
Cross-check
Day 12
PLM + ERP + QMS + plant runnability · 5/5 green
▣ Multi-system policy engine + LLM reasoning
→
7
Thermo
Scale-up
Week 6
BAT-HELIX-v1.4-LAB-02 · 18 kg · QC pass · A-sample ready
▣ Core Connect Batch entity → SAP/MES
→
8
SCIKIQ
Launch
Week 7
DPP-ready dossier · board narrative · Q1 2027 ship
▣ GenAI briefing generator · DPP schema fill
■ Thermo (lab)
■ SCIKIQ (AI)
HOW IT WORKS
The same 8 beats apply to FALCON · ORION · NOVA · KAIROS · MERIDIAN
1
Trigger (SCIKIQ scans for the signal)
Regulatory feed parser flags PFAS rule + entity-resolves to AVY release-liner SKUs in 30 min — not 3 weeks of legal-to-R&D escalation.
▣ Regulation parser (LLM) + impact graph traversal
→
2
Retrieve (R&D Copilot)
18 prior PFAS-free experiments across Mentor/Leiden/Suzhou returned with inline citations. Dr Koen had personally seen 4 of them.
▣ Hybrid RAG (BM25 + embeddings) over PfS + patents + lab PDFs
→
3
Recommend (NBE engine)
Bayesian optimiser proposes HELIX-v1.4 composition (95% REL-SIL-204 + 5% biobased rosin) · P(meet target)=0.78 from 21-run prior. The single best experiment to run next.
▣ BoTorch Bayesian opt + Gaussian Process surrogate + qNEHVI acquisition
→
4
Execute (PfS does its job)
Thermo materialises the experiment — Sample SMP-2026-09712, procedure in ELN, instrument scheduled, raw traces in SDMS, 19.4 g/in PASS recorded with 21 CFR 11 audit.
▣ Core LIMS + ELN + SDMS — unchanged · Thermo-owned
→
5
Enrich (Fabric joins the result to the world)
Result joined to supplier ESG, customer requirement library, LCA model · Sustainability score 84/100 written back to PfS as UDF.
▣ KG update + OData v4 PATCH back to PfS Sample UDF
→
6
Cross-check (multi-system policy)
Formulation validated against PLM (no duplicate), ERP (supplier on AVL), QMS (control plan exists), plant runnability. 5 of 5 green.
▣ Multi-system policy engine + LLM reasoning + KG
→
7
Scale (Batch entity to plant)
BAT-HELIX-v1.4-LAB-02 · 18 kg · QC pass · A-sample ready. SCIKIQ predicts plant yield + flags raw-material AVL risk for week 18.
▣ Batch entity via Core Connect → SAP/MES · yield prediction ML
→
8
Launch (briefing pack + DPP)
GenAI assembles one-page sustainability + DPP substantiation + spec sheet + board narrative · Q1 2027 ship.
▣ GenAI briefing-pack generator · ESPR DPP schema fill
BEFORE / AFTER
HELIX · old way vs new way
| Dimension | Today | With SCIKIQ + Thermo | Δ |
|---|---|---|---|
| Regulator memo → R&D escalation | ~3 weeks | 30 minutes | −3w |
| Prior-art search | 1–2 weeks · 4 PDFs found | 30 seconds · 18 cited | −2w + 4× |
| Experiments to hit spec | 8 trial-and-error | 2 NBE-guided | −6 runs |
| Sustainability + DPP dossier | 5 days post-hoc | Auto-assembled | −5d |
| Idea → A-sample | ~14 weeks | 7 weeks | −7w |
| Probability of first launch hit | ~25% | 78% (Bayesian) | +53 pts |
WHO WINS
"I get 18 prior experiments I had never seen — 30 seconds. The optimiser tells me which composition to try first. PfS captures the result. SCIKIQ writes the sustainability score back to my Sample. I never leave my LIMS."
— Dr Stefan Koen · Senior Scientist · Coatings · Leiden (NL)
⚠ IF WE DON'T
If HELIX runs the old way
- $42M of revenue exposed to EU PFAS enforcement (Q1 2027 deadline)
- Customer A-sample slips Q3 2027 → loses anchor customer to faster competitor
- Programme costs $4.2M instead of $2.8M (8 trial-and-error vs 2 NBE-guided)
- Sustainability story falls behind Sustainable Advantage™ commitments
7weeksIdea → customer-ready dossier
14→7weeksTime compressed vs. old way
0.78P(meet)Bayesian posterior · first try
$42MsecuredRevenue exposure removed
End-to-end: fourteen weeks compressed to seven. Same eight beats; the difference is the SCIKIQ layer doing 1, 2, 3, 5, 6 and 8 — Thermo PfS still doing 4 and 7.
Same lab. Same scientist. $42M secured. Seven weeks.
AI / ML / LLM IN THIS CHAPTER
DATA SOURCES