The audit stops costing person-weeks

The problem

The risk is not doing it badly: it is not being able to prove it.

The real cost of an audit is not the auditor's day: it is the person-weeks of reconstruction beforehand, repeated for every standard and every cycle. Hunting down the quarter's cleaning records, cross-referencing them with batch releases, locating the certificates of analysis of the raw materials involved. And the real risk is not doing things badly: it is not being able to prove they were done well. The auditor raises the non-conformity because there are documentary gaps in the reconstruction, not because the cleaning was not done.

• An exporting plant lives in permanent audit: cosmetics GMP as the backbone, plus quality, environment, health and safety, corporate audit, the national health authority and the destination market regime.
• Each one asks for the same reality ordered differently, and that reorganization is done by hand, again, every time.
• The real cost is not the auditor's day: it is the preceding person-weeks of reconstruction, repeated per standard and per cycle.
• The auditor raises the nonconformity because there are documentation gaps in the reconstruction, not because the cleaning was not done.

How it fits the IRIS system

Edge and Connect capture; Agents reorganizes by standard.

Edge captures the physical evidence as it happens: before-and-after photos of the critical points at shade changeover, line status, mixer panel readings. Connect captures the documentary evidence: orders and weighings, certificates of analysis, laboratory results. And Agents does what no person wants to do: map every piece of evidence against the requirements of every standard, compose the dossier by period and by standard, and — the most valuable part — detect what is missing while it can still be captured. On audit day the auditor logs in and downloads the evidence pack for the period in the format of their standard. Nobody in the plant has spent an hour preparing it.

The documentation gap stops being discovered on the auditor's day and starts being discovered the same day it occurs, which is while it can still be fixed. That change of timing is worth more than the person-weeks saved.

See the full IRIS architecture →

Before and after

Today's preparation versus the permanent evidence pack

Aspect	Today	With iLEAN Agents
Preparation per audit	1-2 person-weeks	Zero
For each additional standard	The whole job repeats	Same evidence, another view
When the documentation gap appears	On the auditor's day	The day it occurs
When evidence is captured	Rebuilt afterwards	At the moment of the event
Cleaning records and releases	Cross-referenced by hand	Cross-referenced and tied per batch
Certificates of analysis	In another folder	Tied to raw material and product

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 6-12 months, depending on audit frequency and the number of applicable standards.
• Direct recovery of person-weeks a year currently spent rebuilding instead of improving.
• Fewer documentation nonconformities, which are the majority of those raised.
• Continuous protection of the certifications that enable exporting — their value is of the order of the revenue of the market they protect.

direct recovery of person-weeks per year, continuous protection of the certifications that enable exports — whose value is of the order of the revenue of the market they protect — and fewer documentary non-conformities. Estimated payback of 6 to 12 months depending on the number of audits per year. *Estimate to validate*.

And the fair question from the production manager

«Isn't this just another document manager?» — the difference is direction. A document manager stores what somebody decides to upload, after the fact, and therefore inherits the same gaps. Here the evidence is generated on its own the instant the event happens, and what Agents does is reorganize it according to whichever standard asks for it. Nobody has to remember to upload anything.

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