The lab spreadsheet reaches the ERP on its own

The problem

The ERP holds the batch release, but not the series behind it.

The file lives in a shared folder the ERP does not look at. Somebody transcribes it later, late and with errors, or does not transcribe it at all and the ERP holds the batch release without the data series behind it. When a complaint or an audit arrives, sheet and batch have to be matched by hand, and by then several versions of the file exist across the network. Connect in listening-inward mode observes the shared folder and reads the file as soon as it is saved. A grounded model parses it, structures it and links it to batch, line and shift. The lab keeps working exactly as it does today: same sheet, same place, same routine.

• Degrees Brix of the finished syrup, CO₂ volumes, pH, fill level, cap torque, residual chlorine and treated water conductivity, washer caustic concentration. These decide whether the product ships.
• That file lives on a shared folder the ERP never looks at.
• Somebody transcribes it later, late and with errors, or does not transcribe it at all.
• When a complaint or an audit arrives, the sheet has to be cross-referenced with the batch by hand — and by then several versions of the file exist across the network.

How it fits the IRIS system

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

Once validated, it enters the ERP through the API. The quality lead reviews the summary and validates. Only then does iLEAN push the data into the ERP via API, already attached to the batch. The two parallel realities end: the process series stops being invisible and becomes cross-checkable against line rejects, complaints and the washer parameters. CFO angle: fewer ERP licences, year after year. The ERP bills per active user. If five people in lab and quality log into the ERP purely to key in process data, that is five quality module licences. With Connect only the person who validates goes in: one. The typical reduction sits between fifty and eighty percent of that module's licences, and it recurs every single year.

Several versions across the network is the symptom that gives away the underlying problem: when the official figure arrives late, everyone keeps their own copy. With automatic push there stops being a reason to.

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 ERP	Days or weeks	Second zero
What the ERP holds	The release, without the series	Both, tied to the batch
Sheet ↔ batch cross-check in an audit	Manual	Already done
Versions of the file	Several across the network	One, the one being read
Manual transcription	Half a day a month	Zero
Quality module licenses	Several	One, the supervisor who validates

Impact estimate

Estimated impact — to validate against your actual 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: the ERP bills per active user. Anyone entering the quality module purely to type results stops needing to: 50-80 % reduction, recurring every year.
• Estimated payback 4-9 months.
• Half a day a month of transcription recovered.
• Manual cross-check time that disappears at every audit.

On top of the license saving there is the half day a month lost to transcription and the manual matching at every audit. Estimated payback between four and nine months. Estimate to be validated against your actual ERP invoice for the last financial year.

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 structure tolerating merged cells, margin notes and mixed units. It is an anchored task, where the best models drop below 1.5% error [1], and the quality supervisor validates before anything crosses into the ERP.

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