Foundation color matching with AI — today's N3 the same as tomorrow's N3, batch after batch.

The problem

The consumer buys by shade name — the batch has to keep that promise.

Foundation color matching is not a brand whim: it is a brand promise. The consumer found “her shade” after trying several references, buys it by name and expects the same result on her skin every time. A drift the lab chemist's eye considers within technical tolerance can be perceptible to her — and come back as a complaint, a bad review or a brand switch.

1. The spectrophotometer measures well but arrives late — the start-of-batch sample passes the test, the batch is packed, and the drift appears in an intermediate stretch (pigment settling when agitation dropped, a formula correction that did not blend in, residue of the previous SKU in the piping).
2. Wide ranges multiply the risk — 40+ shades on the same line, frequent SKU changes, residual pigment cross-contamination in dosers. Mid and dark shades are the most sensitive: a small numerical ΔE is more visible to the eye.
3. Multicultural formulas shift the problem — the same pigment in a warmer or cooler base moves the final result; bulk color control does not always capture how the packed product will look under different lighting.

Every drift complaint is hard cost (reverse logistics, replacement, customer service) plus brand cost, which weighs more in the medium term. The classic system works 99% of the time. That 1% is what ends up in a bad review — and every bad review costs more than catching it in time.

How it fits the IRIS system

iLEAN Vision extends your color control to 100% of the flow — the putty between the lab and the filler.

The color matching problem is not a lack of judgment: it is that the color manager's judgment does not scale to the pace of the line. iLEAN acts as the putty that fills the gap between the sample spectrophotometer and the packing head, without asking you to change the LIMS, the ERP or the filler.

Vision looks at the color on every SKU, every batch, every shade change. The agents cross-reference the ΔE with the master standard and the historical trend. If the drift passes threshold, the batch is held. The person signs — never the other way around.

The specific piece for this pain:

• iLEAN Vision — the computer vision layer of iLEAN Edge, colorimetrically calibrated: an industrial camera with controlled lighting over the dosing head or the transfer piping, a CNN trained with your master standards, live ΔE calculation against the SKU's reference. It works with wide ranges (40+ shades) — the network learns your product, not a generic one. It works without a network. If the plant loses WiFi, Vision keeps measuring and holding.
• iLEAN Connect — captures the master standards whether they come from the LIMS, from the color manager's Excel file or from the ERP recipe; and captures the traceability of the previous SKU change (what was packed before, what pipe cleaning was done).
• Color agent — cross-references the Vision image, the master standard from Connect, the historical trend of the same SKU over recent batches and the current formula. If there is a deviation, it holds the batch and alerts the color manager with the data — measured ΔE, comparison against the last N batches, proposed action. The human signature decides to pack, adjust or discard.

See the full IRIS architecture →

Before and after

Classic color control vs. in-line verification with iLEAN Vision

Aspect	Spectrophotometer + lab sample	With iLEAN Vision over the head
Share of the batch inspected	Spot sample at the start (~0.1%)	100% of the flow, before packing
SKU change in a multi-shade range	A single measurement; intermediate drift not captured	Continuous shade-by-shade verification
Moment the drift is detected	Consumer complaint	Before packing, in line
Residual cross-contamination	Risk accepted between SKUs	Detected at the new SKU's startup
History per SKU and per batch	LIMS logbook, rebuilt by hand	Automatic per-batch dossier with photo and ΔE
Decision on an out-of-threshold batch	Verbal, not always traceable	Color manager's signature, on record

Impact estimate

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

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

• Makeup brand with a 30-50 shade range of foundations and bases, 2 packing lines, frequent SKU changes.
• Vision pilot on one dosing head with a calibrated camera and controlled lighting, trained with your master standards from the LIMS. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the annual volume of documented drift complaints and the average reverse-logistics cost per unit.
• Defensible floor for the reduction in drift complaints: ≥ 30%. The hard lever is the hard cost of reverse logistics plus the soft cost of brand and reviews.

And the color manager's reasonable doubt

“What if the AI gets it wrong and holds a batch that is fine?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely compares an image against a master color standard, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN reports the measured ΔE with its history and the color manager 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.