Agents organizes the evidence pack by certification standard

The problem

Before every audit, someone rebuilds months of changes from paper.

When the BRCGS or ISO 22000 auditor arrives, someone today spends weeks reconstructing the traceability of the period's mold, color and resin changes. The plant shows loose papers; the auditor raises a non-conformity for documentation gaps.

• When the BRCGS or ISO 22000 auditor arrives, someone spends weeks reconstructing the traceability of every mold, color and resin change in the period.
• The evidence exists, but scattered: startup sheets in folders, lab results in Excel, receiving data in the ERP and the rest in people's memory. Each source has its own format and its own person who knows where things are.
• The plant shows loose papers, the gaps show, and the auditor raises a non-conformity for documentation gaps — not for a bad closure, but for missing proof. The closures may be perfect; the file is what fails.
• And bottlers audit too. Each one with audit rights means the same reconstruction repeated with a different checklist and a different date.

How it fits the IRIS system

Edge and Connect capture each change; Agents files it by standard.

Every critical mold or color change already generates before/after photos of the critical points, a decision log and a timestamped signature. Agents automatically classifies that material by certification standard — BRCGS, ISO 22000, ISO 9001 — with zero human work afterward.

Every critical change already leaves before/after photos of the critical points, a decision log and a timestamped signature. Agents classifies that material by standard — BRCGS, ISO 22000, ISO 9001 — as it happens, so the dossier is closed the moment the change is, with zero work afterward. Quality stops being the team that compiles and becomes the team that reviews. The auditor's visit stops being a project and becomes a review of what is already filed. Bottlers notice the difference in the first audit after go-live.

See the full IRIS architecture →

Before and after

Audit preparation today versus the dossier filed as it happens

Aspect	Today	With iLEAN Agents
Rebuilding changes before an audit	Weeks	None
Mold, color and resin change history	Loose papers	Photos, log and signature per change
Indexing by certification standard	Done by hand per audit	BRCGS, ISO 22000, ISO 9001 automatically
Missing evidence for a change	Found by the auditor	Flagged when it should exist
Each bottler's own audit	Rebuilt again	Same evidence, filtered view
Dossier delivery	Days of compiling	Downloadable instantly

Weeks spent rebuilding traceability before every audit → dossier downloadable instantly.

Impact estimate

Impact estimate — to be validated with your BRCGS and bottler audit calendar.

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 how many audits you face per year. Plants audited by several bottlers recover it fastest.
• Pre-audit reconstruction goes from weeks to zero: the dossier is built as each change happens. Nobody has to stop the quality team for weeks before the visit.
• Documentation non-conformities lose their main cause, because missing evidence is flagged when it should have been generated, not on audit day. The audit becomes a review of evidence, not a search for it.
• Each bottler with audit rights gets the same evidence filtered for their checklist, instead of a new round of compiling by the quality team. The same evidence answers every checklist.

Payback 6-12 months · weeks of rebuilding reduced to zero. From weeks of pre-audit reconstruction to zero. Estimated payback of 6 to 12 months depending on audit frequency. Estimate to validate.

And the fair question from the production manager

“Will a BRCGS auditor accept a dossier filed by AI?” — the AI files, it does not judge: the images come from the cameras and the signature is the quality supervisor's. Assigning a known evidence type to the clause it supports is an anchored task, where the best models drop below 1.5% error [1], and the classification rules are reviewed by your quality manager before go-live. Any rule can be adjusted later without touching the evidence itself. The auditor sees original evidence, organized. Nothing is rewritten after the fact, which is exactly what an auditor wants to see.

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