The spreadsheet that rules no longer needs a transcriber

The problem

The spreadsheet decides what is sellable, and somebody retypes it.

This is not side data. It is the 4-CBA that decides whether the acid meets specification. It is the intrinsic viscosity and the acetaldehyde that decide whether the chip is bottle grade or gets downgraded. It is color, diethylene glycol content, moisture, fines. The entire business runs through that sheet. And between that sheet and the ERP sits a person retyping. Half a day a month of a qualified person, plus a permanent gap between the real number and the official number. There is a third cost that almost never gets counted: quality analytics and the thickness measurements from the inspection round live in a format nobody can pull automatic trends from. In a plant whose greatest static risk is corrosion from acetic acid and bromide, that is expensive.

• This is not side data. It is the 4-CBA that decides whether the acid meets specification, the intrinsic viscosity and the acetaldehyde that decide whether the chip is bottle grade or gets downgraded, plus color, diethylene glycol, moisture and fines.
• Between that sheet and the ERP sits a person retyping: half a day a month of a qualified analyst, plus a permanent gap between the real number and the official number.
• The thickness readings from the inspection round live in the same kind of format — one nobody can pull an automatic trend from.
• In a plant whose greatest static risk is corrosion from acetic acid and bromide, a corrosion curve that does not exist is expensive.

How it fits the IRIS system

Connect in listen-inward mode — it reads the file, not the person.

Connect listen-inward mode.

The corrosion curve per measurement point and the specification trend per silo stop being something an engineer builds by hand: they exist the moment the sheet is saved and validated. And the CFO angle is direct — only the person who validates needs an ERP login, and that is a line on next year's license invoice, not a projection.

• It observes the shared folder and reads the file when it is saved — it does not monitor any person: it reads a file.
• A model anchored to the known template extracts each field with its unit and measurement point.
• Central memory.
• Human validation on tablet or desktop.
• Posting to the ERP via API. From then on, the thickness trend per measurement point and the specification trend per silo exist without anyone building them.

See the full IRIS architecture →

Before and after

The lab sheet today versus the connected lab sheet

Aspect	Today	With iLEAN Connect
Gap between real and official data	Weeks	Seconds, after validation
Manual retyping	Half a day a month, minimum	Zero
Intrinsic viscosity trend per silo	Built by hand, when asked	Exists
Thickness readings	On loose sheets	A corrosion curve per measurement point
Sheet format	Has to be respected	Read as it is, even as it changes
ERP quality module licenses	Six	One, the person who validates

from half a day a month of retyping and a weeks-long gap between real and official data, to continuous posting with validation in seconds. From thickness readings on loose sheets to a corrosion curve per measurement point.

Impact estimate

Estimated impact — to validate against your actual license invoice.

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.

• 💰 CFO angle: if six people from the lab and inspection log into the ERP purely to key in results, that is six quality module licenses. With Connect only the validator logs in: from 6 to 1, a −50% to −80% range, recurring every year at the vendor's rate card.
• Estimated payback 3-6 months — one of the fastest in the catalog.
• Half a day a month of a qualified analyst recovered.
• And a usable mechanical integrity database, from the same readings the inspection round already takes.

direct, recurring ERP license savings. If six people from the lab and inspection log into the ERP today exclusively to key in results, that is six quality module licenses; with Connect only the person who validates logs in: from 6 to 1, in the typical −50% to −80% range for that module, with recurring annual savings at the vendor's rate card. Add the qualified staff time recovered and a usable mechanical integrity database. Estimated payback 3-6 months — one of the fastest in the catalog. *Estimate to validate against last year's license invoice.*

And the fair question from the production manager

«Three shifts have been editing that sheet for years — does it hold up?» — that is the anticipated case: parsing does not depend on fixed cell positions, it interprets columns, units, merged cells and margin notes without asking anyone to standardize anything. It is an anchored task, where the best models drop below 1.5% error [1], and a person validates before the push. A value outside the specification window for that grade — an intrinsic viscosity that would be physically impossible for that reactor — is flagged as uncertain, never posted quietly to the ERP.

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