CASE STUDIES // ILLUSTRATIVE
Modelled,
not invented.
Serumon is pre-revenue, so these are engineering scenarios built from real plant economics and real process behaviour — not customer results. Every number on this page is a model, and every model is labelled. When real deployments produce real numbers, they will replace these with a named customer attached.
Shrink the variance and the mean can follow it down.
SCENARIO 01 // SHADERA
Colour cosmetics: the shade-hold tax
A foundation line runs 34 shades across three substrates. Shade is judged against physical standards by four matchers covering the whole site. Roughly one batch in nine is held for ΔE, and around a third of those need two or more re-shading passes.
Shadera reads inline, projects ΔE forward to end of batch, and prescribes the pigment trim to Emulson while the vessel is still open. The hold becomes a correction made three hours earlier.
- Assumption: 11% of batches held for ΔE at baseline.
- Assumption: $5k–$40k pigment cost per correction pass.
- Modelled effect: drift detected before end of batch on most holds.
- Metric contracted: holds per 100 batches, measured against shadow baseline.
SCENARIO 01 / MODELLED
- BATCHES / YEAR
- 1,150
- HELD AT BASELINE
- ~127
- DETECTED EARLY (MODEL)
- MAJORITY
- RE-SHADE PASSES SAVED
- MODELLED
- METRIC
- HOLDS / 100 BATCHES
- STATUS
- ILLUSTRATIVE
SCENARIO 02 // FILLIX
Personal care: 1.3 grams per bottle
A 50 ml jar line targets 50.0 g with a 49.0 g minimum. To protect compliance the line runs a mean of 51.3 g. That is 2.6% giveaway across 18 million units a year.
Fillix models the per-head distribution, tightens head balance, and walks the mean down while keeping underfill probability under the quality limit. The compliance floor never moves; the safety margin gets smarter.
- Assumption: 18M units/year, 2.6% giveaway at baseline.
- Assumption: existing filler supports fine setpoint resolution.
- Modelled effect: mean fill reduced as variance is characterised.
- Metric contracted: giveaway %, with zero underfill tolerance.
SCENARIO 02 / MODELLED
- TARGET FILL
- 50.0 g
- BASELINE MEAN
- 51.3 g
- MODELLED MEAN
- 50.4 g
- GIVEAWAY BEFORE
- 2.6%
- GIVEAWAY AFTER (MODEL)
- 0.8%
- STATUS
- ILLUSTRATIVE
SCENARIO 03 // EMULSON
Skincare: the batch that broke at 3am
A high-value SPF emulsion breaks intermittently. Root cause is never conclusively found because the process trace lives in a historian nobody queries and the lab result arrives eight hours after the decision point.
Emulson estimates emulsion state during the batch, predicts the fault, and trims shear and temperature inside the approved envelope. The chemist gets an alert with the reason and the option to say no.
- Assumption: $18k–$90k raw material per held batch.
- Assumption: 6–14 vessel hours lost per event.
- Modelled effect: fault predicted before end of homogenisation.
- Metric contracted: right-first-time rate and scrap volume.
SCENARIO 03 / MODELLED
- FAULT EVENTS / YEAR
- ~30
- MATERIAL AT RISK
- $18k–$90k
- VESSEL HOURS / EVENT
- 6–14
- PREDICTION HORIZON
- PRE-EOB
- METRIC
- RIGHT-FIRST-TIME
- STATUS
- ILLUSTRATIVE
SCENARIO 04 // TWYNEX
CDMO: the week that scheduled itself badly
A contract site runs 34 SKUs a week across six vessels and four fillers, sequenced by an experienced planner in a spreadsheet under allergen, colour and CIP-validity constraints.
Twynex solves the same problem with cuOpt, respecting every constraint the planner respects, and hands back a sequence with materially fewer clean cycles and better filler balance. The planner keeps the veto.
- Assumption: 61.5 changeover hours/week at baseline.
- Assumption: 22 clean cycles per week.
- Modelled effect: sequence optimisation reduces both.
- Metric contracted: changeover hours per week and filler utilisation.
SCENARIO 04 / MODELLED
- SKUs / WEEK
- 34
- CHANGEOVER BEFORE
- 61.5 h
- CHANGEOVER AFTER (MODEL)
- 44.2 h
- CLEAN CYCLES
- 22 → 17
- FILLER UTILISATION
- 71% → 84%
- STATUS
- ILLUSTRATIVE
SCENARIO 05 // STABION
Release: inventory sitting on a lab queue
Finished goods wait three to ten days for micro and stability clearance. Occasionally a late failure appears after distribution has already started, and the cost multiplies.
Stabion scores risk from process signatures at make time, flags the batches that deserve scrutiny, and assembles the cited evidence pack so the qualified person signs faster with more information, not less.
- Assumption: 3–10 day average hold at baseline.
- Assumption: late failures cost 10–40x an early catch.
- Modelled effect: risk visible at make time, not at release time.
- Metric contracted: hold days and late-failure count.
SCENARIO 05 / MODELLED
- AVG HOLD
- 3–10 DAYS
- RE-TESTS
- 1–3
- LATE FAILURE MULTIPLIER
- 10–40x
- RISK VISIBLE AT
- MAKE TIME
- EVIDENCE PACK
- AUTO-ASSEMBLED
- STATUS
- ILLUSTRATIVE
METHOD
How these models were built
PLANT ECONOMICS
Unit volumes, fill targets, batch values and hold durations drawn from published industry ranges and design-partner conversations.
PROCESS PHYSICS
Emulsification, rheology and fill behaviour modelled with the same surrogates the product uses, not with arbitrary improvement percentages.
STATED ASSUMPTIONS
Every scenario lists its assumptions on this page. If you disagree with one, the conclusion changes, and that is the point.
ACROSS SCENARIOS
The metrics we are willing to be judged on
ALL FIGURES ON THIS PAGE ARE MODELLED SCENARIOS, NOT CUSTOMER RESULTS.
Pre-revenue claims should be stated as planned motion, not completed traction, until contracts, telemetry and case-study evidence exist.
CASE STUDY FAQ
About these numbers
Because the alternative is either an empty page or invented results. Scenarios let a plant engineer check our assumptions against their own reality and decide whether the maths could work for them.
Yes, as design-partner deployments produce measured results and partners consent to be named. The scenarios will be archived rather than quietly deleted.
Yes. Give us your volumes, fill targets, hold rates and changeover hours and we will build the same model against your numbers, including the case where it says do not buy this yet.
Then the modelled upside is larger and the risk of a disappointing pilot is smaller. Sites with messy baselines are usually the best pilots, provided the data to measure them exists.
YOUR NUMBERS
Model your plant, not ours
Send volumes, fill targets, hold rates and changeover hours. We will build the scenario against your data and show you the assumptions we used.