Your 90s packaging line still runs on MS-DOS — and your BI has a hole the size of its counter.

The problem

The old machine is still there — and nobody wants to touch it because it works.

The plant manager puts it plainly: "that line delivers OEE, it runs on its own, I am not stopping it for a retrofit". And they are right. The line produces, the shift operator reads the counters on the CRT screen and writes them down in a spreadsheet at the workstation. The corporate BI has a hole exactly the size of that line — and the IT director has been asking for budget to integrate it for years.

1. The PLC retrofit costs more than it is worth. Not because the vendor is being difficult: because touching old OT without stopping production is a delicate, expensive operation.
2. The physical spare part is no longer manufactured. Replacing the whole machine is not an option while it keeps making good product.
3. The counter arrives late and by hand. The reality of that line reaches the BI in next day's spreadsheet, with typos and gaps.

The classic excuse — "I am not integrating that old machine because it costs more than it is worth" — was true for decades. That expired. It is not a moral flag: it is arithmetic. AI dropped the cost of recontextualizing formats through the floor. Today the screen is read the way a person reads it, without touching the PLC, without opening the OT ring.

How it fits the IRIS system

iLEAN does not ask you to scrap the machine — it puts a camera in front of it and leaves the line alone.

Connect is the filler between the 90s line and the corporate BI. It does not tear out the tiles: it fills what was left hollow. In its manual/intermediate gradation — the one iLEAN's foundation designed precisely for machines that will not let themselves be touched — a fixed camera sees the screen the way a person sees it and the AI extracts the data.

The camera looks at the CRT. Connect reads the counters. The agent ties them to the batch and to the BI. The person signs — the PLC never even notices.

The iLEAN pieces applied to the MS-DOS line:

• Connect (manual/intermediate gradation) — fixed camera in front of the CRT screen, capture on a cycle (every X minutes or on event). It detects the expected tab; if the camera cannot see it, it does not invent: it flags the capture as uncertain. The gradation is decided line by line — there is no single recipe.
• Connect (listening outward) — it captures whatever falls around the line: the shift supervisor's message reporting a jam, the manufacturer's email with the shutdown procedure, the maintenance note from the last intervention. All of it enters the system at second zero.
• Agents — they cross the counter that was read with the production order, normalize it to the BI/MES format and deliver the data as if the line had been integrated all along. If they have doubts (an unexpected tab change, a counter stuck at zero for too long), they propose a verification to the shift supervisor, they do not invent. The person signs; the line does not restart on its own.

See the full IRIS architecture →

Before and after

Counter written down by hand vs. counter read by camera with Connect

Aspect	Shift spreadsheet + manual entry	With iLEAN Connect + Agents
Counter capture	The operator writes it in a spreadsheet	Fixed camera + reading by vision
Cost of touching the PLC	Expensive retrofit, line stoppage	Nothing is touched, neither PLC nor HMI
Arrival at the corporate BI	Next day, with errors and gaps	At second zero, normalized
Traceability of the capture	Trust in whoever wrote it down	Literal photo + timestamp + batch
OT ring isolation	Respected because nothing is integrated	Respected — a camera is not an OT network
Ability to compute real OEE	Approximate, delayed	Per shift, with a verifiable source

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 lay it out so the committee has an order of magnitude; we refine it during the diagnostic.

• Packaging line with an MS-DOS HMI, counters written down by hand by the shift today, corporate BI with a hole on that line.
• Connect pilot (fixed camera + reading by vision + normalization agent into the BI). First expected value in a few weeks: the first shift enters the BI with the figure read off the screen.
• Indicative payback between 4 and 9 months, depending on how much that line weighs in the consolidated OEE and on the hours spent today rebuilding the figure by hand.
• Reduction of the latency between line event and consolidated BI data of ≥ 30%. The underlying lever is not administrative efficiency: it is seeing the real OEE of that line for the first time.

And the IT manager's reasonable doubt

"What if the AI misreads the counter and injects noise into the corporate BI?" — hallucination is a problem of free generation, not of anchored tasks. Reading a counter on a CRT screen is an anchored task par excellence: the best models pushed the error below 1.5%^[1]. And even so, what is critical is never decided alone: uncertain readings are not written to the BI — they are proposed to the shift supervisor for verification, and the person signs.

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