The dimensional report stops being retyped three times

The problem

Double and triple keying is structural, not carelessness.

Double and triple transcription is structural. The same number is typed into the spreadsheet, into the ERP and into whatever format each customer demands, each with its own template. Every hop is a chance to introduce an error and a stretch of a highly qualified person doing data entry. And because the ERP updates late, the quality status the office sees is not the shop's real status. In a one-off business where the dimensional report is part of the deliverable, that is paid for twice: in hours and in credibility.

• The same number is typed into the spreadsheet, into the ERP and into the format each customer demands, each with its own template.
• Every hop is an opportunity for error and a stretch of a qualified person acting as a typist.
• Because the ERP updates late, the quality status the office sees is not the shop's real status.
• And on one-off work the dimensional report is part of the deliverable: that gets paid twice, in hours and in credibility.

How it fits the IRIS system

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

Connect "listening inward".

The parser interprets the structure of the spreadsheet even when the format changes between customers. A fixed cell mapping breaks the first time somebody inserts a row.

• iLEAN observes the shared folder and reads the file when it is saved; it does not observe the person.
• A grounded model parses the spreadsheet structure even when the format varies between customers and extracts it dimension by dimension.
• It is stored in central memory tied to the job and its revision.
• After tablet validation, it posts to the ERP through the API and fills the customer template.

See the full IRIS architecture →

Before and after

An isolated spreadsheet vs. an integrated one

Aspect	Today	With iLEAN Connect
Times the same number is typed	2-3	Once, when measuring
Spreadsheet → ERP lag	Days	Immediate
Customer template	Assembled by hand per job	Generated from validated data
Quality status in the office	Days old	The real one
Copying errors	Those of repetitive work	There is no copying
Quality module licenses	4 users	1, the validator

dimensional report retyped 2-3 times with days of lag → posted once, validated and immediately available. Customer report assembled by hand per job → generated from already-validated data.

Impact estimate

Impact estimate — to validate against your 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.

• Shops whose deliverable includes a dimensional report in each customer's format.
• CFO angle: today everybody who transcribes needs a quality module user; with Connect only the validating lead needs a license.
• Typical ratio from 4 licenses to 1: between 50% and 80% less recurring cost on that module, year after year and without renegotiating the contract.
• Indicative payback of 3 to 7 months: among the fastest in the set. To be calibrated against the real ERP invoice.

CFO angle — today everyone who retypes needs a seat in the quality module; with Connect only the supervisor who validates needs a licence. Typical ratio: from 4 licences to 1, between −50% and −80% of that module's recurring cost, year after year, with no contract renegotiation. Estimated payback 3-7 months, one of the fastest in the set. *Typical sector range; to be calibrated against the real ERP invoice. Estimate to be validated.*

And the fair question from the production manager

“What if I change the sheet format for a new customer?” — nothing breaks, and that is the reason for parsing with a language model instead of a fixed cell mapping. If a change leaves something ambiguous it is flagged in the validation rather than entering wrong. In a shop working for customers with different templates, a rigid parser would last a week.

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