A cosmetic cream batch is a tree — and hand-drawn trees break at the first complaint.

The problem

A paper batch record is not traceability — it is a promise that traceability exists.

In a cosmetics plant, the tree behind a finished batch has many branches:

1. Master mix in the mixing vessel (reactor or blender) with several raw materials, each carrying its supplier CoA.
2. Adjustments made on the fly (pH, viscosity, colour), decided by the technical manager and written down in the paper batch record — or in a spreadsheet.
3. Bulk split across several packaging lines and several SKUs (50 ml, 100 ml, gift edition, private label format).
4. Packaging that stamps the final batch code on the jar, but loses the upstream genealogy if the data was never cross-referenced in real time.

On a day when the plant runs without surprises, the paper batch record is enough. On the day of the complaint — a consumer with an adverse reaction, a raw material CoA arriving late with a deviation, a US market alert under MoCRA — the manual rebuild begins:

• The mixing room spreadsheet in a shared folder.
• The packaging delivery note in another.
• The paper batch record in the physical archive.
• The supplier CoA in the purchasing inbox.
• And the question "how many jars of the 50 ml SKU shipped to the Spanish market share this master mix?" takes days to answer — when the authority and the retailer wanted an answer in hours.

How it fits the IRIS system

iLEAN Tracer is the filler between the mixing room, the bulk and every SKU.

The problem is not a lack of data — it is that every branch of the tree lives on its own island, and the tree is never assembled in real time. iLEAN Tracer sits on top of what you already have (ERP, MES if there is one, mixing room, packaging line, batch record) and connects the dots around the master batch ID.

Connect captures CoAs, the mixing room, the batch record and the adjustments. Edge reads the label on every SKU. The agent assembles the genealogy tree. The person signs — never the other way round.

The iLEAN pieces applied to cosmetic genealogy:

• Connect in the mixing room — captures the CoA of every raw material whether it arrives by supplier email, the charge sheet from the mixing room industrial PC (modern or old and isolated), the adjustment dictated by voice by the technical manager ("add 0.5% NaOH to the batch"), the paper batch record photographed. Everything enters at second zero, anchored to the master batch ID.
• Edge on the packaging line — a vision terminal that reads the label of every packaged SKU and the unit counter, linking the final batch to the master batch and to the fraction of bulk that went to that line. If the line loses the network, it keeps counting and reading locally.
• Genealogy agent — builds the tree forward (from raw material to jar) and backward (from complaint to raw material). When the question "which jars share this master mix?" arrives, the answer is immediate. And if product has to be withdrawn, the agent proposes the exact list of affected SKUs and markets; the person validates and signs.

See the full IRIS architecture →

Before and after

Manual genealogy vs. cross-referenced genealogy with iLEAN Tracer

Aspect	Manual batch record + spreadsheet	With iLEAN Tracer
Raw material CoA	Supplier email, filed in purchasing	Captured and linked to the master batch at second zero
On-the-fly adjustments	Note on paper or a parallel spreadsheet	Dictated by voice into the system, timestamped and signed
Bulk split across SKUs	Packaging delivery note, rebuilt by hand	Every SKU carries its recorded fraction of bulk
Market complaint	Days spent finding which jars share an origin	Exact list in minutes, ready for a selective withdrawal
CPSR / GMP / MoCRA audit	Manual rebuild of the dossier	Complete dossier per batch, automatic
Behaviour if the network drops	n/a	Edge keeps reading and counting locally

Impact estimate

Impact estimate for your plant — to be validated against your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized cosmetics plant, 1-2 mixing vessels and 2-3 packaging lines, multi-SKU with own brand + private label. Connect capture in the mixing room + Edge on the priority packaging line as a pilot.
• First expected value within a few weeks: complete genealogy running for the pilot SKU batches, demonstrable to auditors and to demanding B2B customers.
• Indicative payback between 4 and 9 months, depending on the historical average cost of handling complaints and the weight of the B2B / export channel in your mix.
• Hard levers: shifting mass withdrawals to selective withdrawals, cutting ≥ 30% of past scope; hours saved rebuilding the dossier by hand; premium B2B / private label channel paying for demonstrable traceability.

And the fair doubt of the cosmetic technical manager

"What if the AI gets an adjustment wrong, or links a CoA to the wrong batch?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualises an existing piece of data (reading a CoA, linking it to a master batch, matching a fraction of bulk to an SKU), the best models brought the error below 1.5% ^[1]. And even so, nothing critical is decided alone: the agent proposes, the person signs. The three safety rings exist precisely for this.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.