The lab spreadsheet no longer lives cut off from the ERP

The problem

The value that decides whether a unit ships lives in a file the ERP never sees.

Two parallel worlds — the real spreadsheet, where live test results actually are, and the official ERP, with data transcribed days late by someone who loses time every month copy-pasting and sometimes gets a value wrong that determines whether the unit can ship.

• Every shift the test lab fills in a spreadsheet with each unit's results before release: dielectric withstand, insulation resistance, contact resistance, ground torque. It is fast, flexible and the lab trusts it, which is why it never went away.
• That spreadsheet lives in a shared folder. The ERP, which is the official record the customer and the auditor rely on, receives the same values days later, retyped by hand.
• Someone loses hours every month copy-pasting rows, and sometimes types one value wrong — the very value that determines whether the unit can ship.
• So there are two parallel truths, and when a customer or an auditor asks for a unit's results, nobody is completely sure which one to show. The spreadsheet is more current; the ERP is more official; neither is fully trusted.

How it fits the IRIS system

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

Connect's "listen inward" mode — watches the shared folder (not the person), reads the delta when the file is saved, parses every cell with a grounded language model, validates it on screen, and pushes it to the ERP via API.

The lab keeps its spreadsheet, its columns and its habits. iLEAN reads the change each time the file is saved, shows it for validation and pushes it to the ERP, so the official record stops trailing reality by days. And the people who only logged in to type results no longer need a seat.

See the full IRIS architecture →

Before and after

The test lab's spreadsheet today versus the spreadsheet wired to the ERP

Aspect	Today	With iLEAN Connect
Dielectric and insulation results in the ERP	Days late	On save, after validation
Retyping results into the ERP	Hours every month	Zero
Wrong value on a release-critical test	Possible, and hard to spot	Caught against the test limits
Lab people with an ERP seat	4-5, just to enter results	Only the validator
Lab spreadsheet layout	—	Kept exactly as it is
Results for a customer query	Which file is right?	One record per unit serial

A days-long gap between real test and ERP data → zero latency. Lab transcription time → zero.

Impact estimate

Lab spreadsheet connection: estimated impact, to validate against your 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.

• Estimated payback 4-9 months, counting license savings and the transcription hours recovered in the lab.
• CFO angle: if 4-5 lab people log into the ERP only to enter results, that is 4-5 licenses; with Connect only the validator needs one, typically 50-80% fewer licenses in the quality and lab module.
• The gap between real test and ERP data goes from days to zero latency, and lab transcription time to zero.
• And a mistyped dielectric or insulation value stops being a silent risk on a unit headed for a data center, because the value in the ERP is the one the lab measured, checked against its limit.

*CFO angle*: if 4-5 lab people log into the ERP today just to enter results, that's 4-5 licenses; with Connect, only the validator needs one → typically 50-80% fewer licenses in the quality/lab module, estimated payback 4-9 months. *Estimate to validate.*

And the fair question from the production manager

“That spreadsheet has been edited by many hands for years — will it really be read correctly?” — reading cells from a sheet whose columns are known is an anchored task, where the best models drop below 1.5% error [1]. And each result is checked against its test limit and shown to the validator before it reaches the ERP, so an odd cell is questioned, not copied.

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