The plant's most expensive data lives in a shared folder

The problem

This is not a discipline problem. It is a latency problem.

The laboratory produces the data, saves it in its spreadsheet, and then somebody moves it into the business system and statistical control. It gets transcribed late, transcribed with typos, and often only the part that "has to be uploaded" gets transcribed at all, losing the rest. Half a day a month of qualified staff dedicated purely to copying numbers from one screen to another is the visible cost. The invisible one is worse: when statistical control fires a trend signal, the data that would have anticipated it had been sitting for three weeks in a spreadsheet nobody read. This is not a laboratory problem or a discipline problem. It is a latency problem, and it is solved by removing the manual step, not by demanding it be done faster.

• The laboratory generates the data, saves it in its sheet, and then someone moves it to the business system and to statistical control. It is transcribed late, with typos, and often only the part that "has to be uploaded".
• Half a day a month of qualified staff copying numbers from one screen to another is the visible cost.
• The invisible one is worse: when statistical control raises a trend signal, the data that would have anticipated it had been sitting for three weeks in a spreadsheet nobody read.
• And this is the plant's most expensive data: film thickness, adhesion, salt spray, radial and lateral runout, imbalance and the verdict of the tests that qualify the part number.

How it fits the IRIS system

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

Connect in listen-inward mode.

iLEAN feeds the statistical control the plant has already deployed. It does not compete with it: it removes the data-entry bottleneck that today makes it arrive late to its own signals.

• iLEAN observes the shared folder and reads the file the moment it is saved. It does not observe the person: it reads the file.
• A grounded model parses the sheet tolerating its real format — merged cells, margin comments, mixed units — and structures the records in central memory.
• The lab manager validates on a summary screen.
• After validation, Connect posts to the business system and to statistical process control through their interfaces.
• An important point for anyone who has already invested: iLEAN feeds the statistical control the plant already runs. It does not compete with it; it removes its data entry bottleneck.

See the full IRIS architecture →

Before and after

Today's laboratory versus the connected laboratory

Aspect	Today	With iLEAN Connect
Gap between real data and system data	Weeks	Minutes
Manual transcription	Half a day a month	Zero
What reaches the SPC	Whatever there is time to upload	The complete series
Trend signal	Fires late	Fires on fresh data
Sheet format	Has to be respected, nobody touches it	Tolerated as is: merged cells, mixed units
Quality module licences	4-6	1, the supervisor who validates

Impact estimate

Estimated impact — to validate with your own 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 4-8 months.
• 💰 Financial angle: if 4 to 6 laboratory and quality people log into the business system today purely to enter results, that is 4-6 quality module licences. With Connect only the validating supervisor needs one: from 4-6 down to 1, a recurring 50-80 % reduction each year depending on your vendor's rate.
• Half a day a month of qualified staff recovered.
• The complete series available for root cause analysis, instead of partial records.

estimated payback 4-8 months. 💰 Financial angle: if 4 to 6 laboratory and quality people log into the business system today purely to enter results, that is 4-6 quality module licences. With Connect only the validating manager needs one: from 4-6 down to 1, a 50-80 % reduction recurring every year at the vendor's list rate. *Estimate to validate against last year's actual invoice.*

And the fair question from the production manager

«What if the technician changes the sheet layout?» — that always happens, which is why parsing does not depend on fixed positions: the model interprets the sheet tolerating merged cells, margin comments and mixed units. It is an anchored task, where the best models drop below 1.5% error [1], and the laboratory supervisor validates on a summary screen before anything crosses into the system.

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