RESOURCES

Useful even
if you never
buy from us.

Practical material for plants evaluating manufacturing autonomy: how to compute your own giveaway number, what to record before any automation project, how to write actuation envelopes, and what to ask every AI vendor including this one.

0GATED PDFs
6RESOURCES
REUSABLETEMPLATES
VENDORNEUTRAL
FILL DISTRIBUTIONFILLIX
TARGETGIVEAWAY 49.0 g53.0 g MEAN 51.3 g · VARIANCE IS THE COST, NOT GENEROSITY

Shrink the variance and the mean can follow it down.

GIVEAWAY MATHSBASELINE CHECKLISTENVELOPE TEMPLATEVENDOR QUESTIONSARCHITECTURE BRIEFROI MODEL

THE LIBRARY

Six things worth taking

CALCULATOR

GIVEAWAY MATHS

From fill target, observed mean and variance to an annual number, with the compliance constraint modelled properly rather than assumed away.

CHECKLIST

BASELINE CHECKLIST

What to record before any automation project so that afterwards you can prove what changed and defend it.

TEMPLATE

ENVELOPE TEMPLATE

A structure for defining actuation limits per parameter, per formula and per vessel, with approval fields built in.

CHECKLIST

VENDOR QUESTIONS

Twenty questions to ask any vendor proposing AI in a GMP workflow. Ask us all of them first.

BRIEFING

ARCHITECTURE BRIEF

The Serumon deployment architecture and data flows, written for IT, OT and security reviewers rather than for buyers.

MODEL

ROI MODEL

The spreadsheet logic we use to size a pilot business case, including the ramp period where savings are honestly zero.

GIVEAWAY MATHS

Run the number on your own line

Take your target fill, your observed mean fill and your unit volume. The difference between mean and target, multiplied by volume, is product you are giving away.

The interesting part is the safety margin: it exists because of variance, not generosity. Reduce variance and the margin can shrink without touching the compliance floor. That is the whole thesis of fill-weight autonomy.

  • Giveaway % = (mean fill − target) ÷ target.
  • Annual loss = giveaway % × annual volume × product cost.
  • Safety margin is set by variance, not by policy.
  • Reducing variance is the only honest way to lower the mean.

WORKED EXAMPLE

TARGET
50.0 g
MEAN
51.3 g
GIVEAWAY
2.6%
UNITS / YEAR
18,000,000
PRODUCT GIVEN
23.4 TONNES
VALUE (ILLUSTRATIVE)
$0.4M–$1.6M

BASELINE CHECKLIST

Record this before you automate anything

QUALITY EVENTS

Shade holds, re-shading passes, broken emulsions, scrapped batches and quality escapes, per SKU family, for at least six months.

FILL PERFORMANCE

Target, mean and standard deviation per head and per format, plus reject rates and underfill incidents.

TIME

Changeover hours, clean cycles, vessel occupancy and hold days from production to release.

COST

Raw-material value per batch, re-work labour, and the cost of a late failure after distribution.

DATA AVAILABILITY

Which tags exist, at what resolution, retained for how long, and whether anyone has ever queried them.

DECISION OWNERS

Who currently makes each call, on what basis, and what would have to be true for them to delegate it.

VENDOR QUESTIONS

Four questions that separate real from theatre

  1. 01

    SHOW ME A DECISION RECORD

    Ask to see one full record: observation, prediction, confidence, proposal, envelope, approver, model version and outcome. Vagueness here is disqualifying.

  2. 02

    WHAT HAPPENS WHEN IT IS WRONG?

    Ask for the escalation behaviour, the confidence threshold and the safe-state path. "It has high accuracy" is not an answer.

  3. 03

    WHAT IS ASPIRATIONAL?

    Ask which capabilities are validated in production versus in simulation versus on a slide. Any honest vendor has all three categories.

  4. 04

    HOW DO I REPLAY A BATCH?

    Ask them to replay a specific batch against the exact model version. If the answer is a log export, there is no audit trail.

BRIEFINGS

For each audience

PLANT DIRECTOR

The economics: which losses are addressable, what a pilot costs, and what payback would have to look like to be credible.

PROCESS ENGINEER

The control detail: state estimation, envelopes, MPC behaviour and what happens when a sensor degrades.

QUALITY MANAGER

Validation approach, audit design, ISO 22716 alignment and the evidence pack an inspector would see.

IT / OT SECURITY

Network segmentation, egress policy, credential scoping, break-glass access and incident commitments.

GLOSSARY

Terms used across this site

Terms used across this site
TERMMEANING
ΔE (Delta E)A numeric measure of perceived colour difference between a batch and its brand standard
GiveawayProduct filled above target weight to guarantee the legal minimum, given away for free at scale
EnvelopeThe approved minimum, maximum and rate limits within which an agent may move a parameter
Right-first-timeThe share of batches that meet specification without re-work or correction
Shadow modeThe agent predicts and records but writes nothing to equipment
Bounded autonomyThe agent acts alone inside validated envelopes and escalates anything outside them
ISO 22716The cosmetics Good Manufacturing Practice standard
Digital twinA simulated model of the make-and-fill process used to predict outcomes before a batch runs

RESOURCE POLICY

How we publish

NOEMAIL WALLS
NOTRACKING PIXELS
YESREUSE ALLOWED
YESVENDOR-NEUTRAL

THE PACK

What arrives when you ask

Request the resource pack and you get the calculators, checklists, templates and the architecture briefing as plain files, sent by a human, with no automated sequence following you around the internet afterwards.

If you want the deeper deployment documentation, that goes under NDA because it contains equipment and security detail.

  • Giveaway and ROI models as spreadsheets.
  • Baseline and vendor-question checklists.
  • Envelope template.
  • Architecture briefing for IT, OT and security.
  • No marketing automation attached.

PACK CONTENTS

CALCULATORS
2
CHECKLISTS
2
TEMPLATES
1
BRIEFINGS
4
GATING
NONE
DELIVERY
BY A HUMAN

RESOURCES FAQ

Using this material

TAKE IT

Get the resource pack

No form-fill sequence, no drip campaign. Tell us who you are and what you are trying to fix, and a human sends the files.