The raw sugar certificate of analysis, turned into an operational signal

The problem

The best predictor of the next three days is being used as a filing document.

The certificate of analysis is the best available predictor of the next three days of quality problems, and today it is used as a filing document rather than as an operational signal. By the time the deviation is discovered, the liquor is already in process. Correcting it means reprocessing, extra activated carbon consumption and refining train hours. The same applies to logistics alerts: a discharge delay read three hours late is a train waiting for raw material. The structural problem is that these signals reach specific people, through different channels, and never cross-reference with live planning data.

• The certificate of analysis is the best available predictor of the next three days of quality problems, and today it is used as a filing document rather than as an operational signal.
• By the time the deviation is discovered, the liquor is already in process.
• Correcting it means reprocessing, extra activated carbon consumption and refining train hours.
• The same applies to logistics alerts: a discharge delay read three hours late is a melt plan that no longer holds.

How it fits the IRIS system

Connect in chaotic-sources mode — and it returns an action, not a notification.

Connect in "chaotic sources" mode:

iLEAN proposes; it does not decide. What arrives at the right role is a concrete action against the melt plan, not a note saying that mail has arrived.

• A dedicated mailbox and messaging number, with iLEAN on silent copy 24 hours a day.
• The model extracts intent (raw sugar arriving, discharge delayed, analytical deviation) and entities (shipment, origin mill, polarization, color, ash, dates).
• Cross-reference against the melt plan and the ERP's active orders.
• Return of a concrete action to the right role, not a notification that mail has arrived. iLEAN proposes; it does not execute. The decision stays with the person in charge.

See the full IRIS architecture →

Before and after

The incoming certificate, before and after

Aspect	Today	With iLEAN Connect
What the certificate is	A filing document	An operational signal
When the deviation is seen	With the liquor already in process	Before opening the feed
Cross-reference with the melt plan	Mental, if there is time	Automatic
Polarization, colour, ash	Read and filed	Extracted and compared against plan
A discharge delay	Read three hours late	Minutes
What reaches the shift manager	The email	The recommended action

Impact estimate

Impact estimate — to validate against your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 3-7 months, depending on shipment frequency.
• Counting avoided reprocessing, extra carbon and steam consumption, and train stoppages caused by logistics blind spots.
• From hours of latency between the external alert and the plant's reaction, to minutes.
• Estimate to be validated against your own shipment history.

estimated payback of 3 to 7 months depending on shipment frequency, counting avoided reprocessing, extra carbon and steam consumption, and train stoppages caused by logistics blind spots. *Estimate to be validated*.

And the fair question from the production manager

"Is the AI going to reply to the mill or the shipper?" — no. It does not write, does not answer and confirms nothing outwards. Extracting intent and entities from a document that is already in front of it is an anchored task [1]; what it does with that is prepare the decision and put it in front of the right role.

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