External signals · impact on AVY
Loading external signals…
1 2 3 4 5 6 7 8
CHAPTER 1 · THE QUESTION

The eleven weeks

…is what Avery loses on every new-product launch — and the largest fixable lever in R&D.

SITUATION TODAY
What happens today on every launch
A new launch follows the same waterfall. Dr Aiyana Chen's team spends 3 weeks framing the problem (searching prior trials in PfS + lab PDFs + memory), 5 weeks running trial-and-error formulations (one design-of-experiment at a time, instrument-bound), 2 weeks validating sustainability (REACH + ESPR + LCA hand-checks), and 1 week handing off to the plant. That 28-week cycle is the gap to category-leading materials companies who launch in 16.
16w
Materials industry leader
Top quartile launch cycle
28w
AD today (avg of 6 active programmes)
Source: PfS programme history 2022-26
6
Programmes per year
Across MG · SG · Embelex
~$60k
Loaded R&D cost / week
Scientist + instrument-hours + materials
BENEFIT
$3.2M per programme · $19M / year
R&D cost out · 11w cycle back · +25 pts first-pass success · 4w earlier revenue
WEEKS
11
lost per launch · 28w today → 17w possible
FORMULATION → LAUNCH CYCLE (weeks)
Today · 28w
Target · 17w
Saved · 11w
0w14w28w
HOW IT WORKS
How SCIKIQ takes the 11 weeks back — three moves
1
Replace search with retrieval
R&D Copilot returns 18 prior similar PFAS-free experiments + 1 supplier datasheet in 30 seconds — not 3 weeks of manual hunt.
▣ Hybrid RAG (BM25 + OpenAI embeddings) over PfS + patents + lab PDFs
2
Replace trial-and-error with active learning
Bayesian optimiser proposes the next experiment most likely to hit spec; P(meet target)=0.78 from 21 prior runs — not 8 trial-and-error attempts.
▣ Gaussian Process surrogate + multi-objective Bayesian opt
3
Replace post-hoc checks with pre-lab scoring
When the scientist saves a draft formulation, sustainability + regulatory + customer-fit scores are written back to the PfS Sample UDF in real-time.
▣ Hazard classifier · LCA inference · OData v4 UDF write-back
BREAKDOWN
Where the 11 weeks (and $3.2M / programme) actually go
Trial-and-error formulation
5w · $1.4M · 45%
Bayesian NBE collapses to 2w
Problem framing & prior-art search
3w · $1.1M · 27%
RAG copilot collapses to 1w
Sustainability + regulatory check
2w · $0.4M · 18%
Pre-lab scoring, near-zero
Earlier revenue capture
· $0.3M · 10%
4w earlier ship × launch margin
BEFORE / AFTER
Before / After · per programme
DimensionTodayWith SCIKIQ + ThermoΔ
Cycle time (Idea → A-sample) 28 weeks 17 weeks −11w / −40%
First-pass formulation hit rate 22% 47% +25 pts
Lab cost per programme $4.2M $2.8M −$1.4M
Failed runs per launch 11 4 −7 runs
Revenue captured earlier +4w × programme margin +$0.7M
WHO WINS
"I get six weeks back per programme. My team runs four targeted trials in the time we run one trial-and-error sequence today."
Dr Aiyana Chen · Principal Scientist · Adhesives · Mentor (OH)
⚠ IF WE DON'T
Cost of inaction (annualised)
  • $19M of R&D spend trapped in cycle time (6 programmes × $3.2M)
  • 4 launches/yr arrive after the regulator (PFAS, ESPR, Prop 65 windows missed)
  • 2 customer programmes ceded to faster competitors (industry observation)
11wback per launch28w → 17w · same lab · same scientist
$19Mtrapped / yearAcross 6 active programmes
+25pts first-passFormulation hit rate
0.78P(meet target)Bayesian posterior · first try
Every great industrial company faces the same question. The factory works. The labs work. The customers pay. But the next product — the one the regulator hasn't named, the customer hasn't asked for — is taking too long. At Avery Dennison that gap is eleven weeks per launch. Across six active programmes that compounds into $19M of trapped R&D spend per year.
28 weeks today. 17 weeks possible. The delta is the prize.
AI / ML / LLM IN THIS CHAPTER
Hybrid RAG · BM25 + Azure OpenAI text-embedding-3-largeBayesian active learning · BoTorch-styleGaussian Process surrogate modelMulti-objective acquisition (qNEHVI)
DATA SOURCES
Thermo PfS · samples · experiments · results · batchesAD R&D programme history 2022-26 (6 programmes)Industry launch-cycle benchmarks (specialty chemicals + materials)AD loaded-cost model (scientist + instrument-hr + materials)