CHAPTER 4 · THE THESIS
Thermo captures. SCIKIQ operationalises.
Two systems · one sentence · four layers · a category nobody else owns.
SITUATION TODAY
Why a 4-layer separation matters
Other vendors blur the line — they bundle a chatbot on top of the LIMS, or stand up a data lake and call it intelligence. Neither works at enterprise scale. The lab needs absolute stability + 21 CFR 11 audit; the enterprise needs flexibility + cross-system reasoning. The two cannot live in the same product without one compromising the other.
Thermo · Labvantage · Benchling
Vendors who own the lab today
All vertical · all lab-only
Palantir · Databricks · SCIKIQ
Vendors who own enterprise AI
All horizontal · few lab-aware
Zero
Vendors who own BOTH well
The category we are creating
Existing tenant
Time SCIKIQ already lives at AVY
No procurement cycle to start
BENEFIT
AI-Native Scientific Intelligence Platform
Not dashboarding · not a data lake · not a chatbot — the category nobody else owns
L4
OUTCOMES
−40% cycle · PFAS-free · +22% patents · −35% complaints
KPI dashboardsExecutive narrative GenAI
Board
↑ signals back ↑
L3
SCIKIQ AI
Copilot · Next Best Experiment · Sustainable substitution · Root-cause
Azure OpenAI gpt-4.1Bayesian opt (BoTorch)Active learningRAG with citations
Decision
↑ signals back ↑
L2
SCIKIQ FABRIC
KG · PfS + ERP + PLM + MES + QMS + supplier + ESG · semantic layer
Knowledge graphOntology / RDFActive metadataEntity resolution
Context
↑ signals back ↑
L1
THERMO PfS
Core LIMS · ELN · SDMS · CDS · Connect (OData v4)
21 CFR 11OData v4Instrument integration
Record
HOW IT WORKS
How the two-way pipe actually runs
1
L1 captures (Thermo)
PfS records every sample, experiment, result, batch with 21 CFR 11 audit. This is unchanged.
▣ Core LIMS · ELN · SDMS · CDS · Connect (OData v4)
→
2
L2 unifies (SCIKIQ Fabric)
PfS + ERP + PLM + MES + QMS + suppliers + ESG join a single property graph. One vocabulary. One lineage.
▣ Active metadata · entity resolution · KG · RDF semantic layer
→
3
L3 reasons (SCIKIQ AI)
Copilot retrieves; Bayesian optimiser proposes; classifiers score; generators draft. Every output cited back to source rows.
▣ Azure OpenAI gpt-4.1 · BoTorch · in-house ML · hybrid RAG
→
4
L4 ships outcomes (Board)
−40% cycle · PFAS-free in time · +22% patents · −35% complaints. The number the board sees.
▣ Narrative GenAI · KPI surface
→
5
Signals flow DOWN to L1
SCIKIQ writes 3 UDFs back to the PfS Sample entity (Similarity · Sustainability · Customer-fit). Scientists see them in their normal LIMS view.
▣ OData v4 PATCH · UDF schema · weekly refresh
BEFORE / AFTER
Why this split wins vs the alternatives
| Dimension | Today | With SCIKIQ + Thermo | Δ |
|---|---|---|---|
| Lab stability + audit | PfS = strong | PfS unchanged · still strong | kept |
| Enterprise reasoning | None | SCIKIQ KG + AI | new capability |
| Vendor lock-in risk | 1 vendor (Thermo) | 2 vendors · 1 standard (OData) | −risk |
| AI capability cap | Vendor roadmap | Open · AVY-owned | +pace |
| Procurement cycle to start | — | Existing SCIKIQ tenant | 0 days |
WHO WINS
"One sentence I can use with the board, the analysts and the customers. Thermo captures the science · SCIKIQ operationalises enterprise intelligence on top. AVY is the reference for AI-native scientific intelligence."
— CEO Deon Stander · Office of the CEO
⚠ IF WE DON'T
What happens if we collapse the layers
- Bundle AI into PfS → vendor sets the pace · AVY's $14M PfS investment becomes a roadmap dependency
- Build a data lake under PfS → 18 months of plumbing · no decisioning · the gap remains
- Drop in a GenAI chatbot → ungrounded · hallucinated · zero provenance · fails the first audit
4layersClean separation of concerns
2vendorsThermo + SCIKIQ · one open door
1sentenceCategory nobody else owns
0lock-inOData v4 is a standard
Two-way. Thermo feeds raw science up. SCIKIQ feeds enriched signals — similarity scores, sustainability scores, customer-fit scores — back down as PfS user-defined fields. The lab keeps its system of record. The enterprise gets its system of intelligence.
Thermo captures. SCIKIQ operationalises. Avery wins.
AI / ML / LLM IN THIS CHAPTER
DATA SOURCES