Cleaning validation with AI — dynamic MACO and continuous swab/rinse, without a bottleneck at every changeover.

The problem

The changeover between products is where the dead time nobody looks at piles up.

In a multi-product GMP plant, the changeover from product A to product B always follows the same script:

1. CIP/SIP cleaning to a fixed SOP. The CIP system runs the programmed cycle — the same one for every product pair, because sizing it for the worst case is the safe option.
2. Manual swab/rinse sampling by the QA technician. Samples are taken from the equipment's critical points (hard-to-reach areas, gaskets, elbows), labeled and sent to the lab.
3. Waiting on the LIMS. The lab analyzes the swab and the rinse. Results in hours or days, depending on the plant. The line, stopped.
4. Cleaning release. QA reviews, compares against the MACO calculated in a spreadsheet 3 years ago using the worst case for toxicity, signs and releases. The line restarts on product B.

The result: between 8 and 24 hours of cleaning blocking the line at every changeover. In a plant with 12 products sharing equipment and a weekly changeover, that is several shifts a month lost that nobody charges to any product. The MACO is set for the worst case because recalculating it is manual work; the sampling plan is the same for high and low risk because the SOP does not distinguish; the lab is a bottleneck because the wait is linear. And on audit day, someone has to rebuild all of it — because the cleaning dossier lives in a folder of signed spreadsheets.

How it fits the IRIS system

iLEAN does not replace CIP/SIP — it recalculates MACO against reality and closes a living dossier.

The cleaning problem is not a lack of validation: it is frozen validation. The MACO was calculated once for the worst case and never recalculated; the SOP is the same for high and low risk; the dossier lives in a folder. iLEAN acts as the putty over your CIP/SIP, SCADA, LIMS and MES systems — it does not replace them, it reads what each one knows and keeps the MACO recalculated, the sampling proportional to risk and the dossier alive under Part 11.

Tracer reads CIP, SIP, LIMS and MES. Writer keeps the cleaning dossier alive. Edge sees the inspectable surface, and the agent recalculates MACO against the reality of the next batch. The QA person signs — the system does not declare itself clean.

The three iLEAN pieces applied to continuous cleaning validation:

• iLEAN Tracer (Connect) — reads the CIP/SIP cycle parameters (conductivity, TOC, temperature, phase times) from the SCADA or from the CIP's own local PC. If the old machine exposes nothing, it reads the panel by vision. It also reads the swab and rinse results from the LIMS as soon as they land.
• iLEAN Writer — keeps the living cleaning dossier per piece of equipment, product pair and campaign. Traceable master template, Part 11 audit trail, ALCOA+ integrity. The audit reads it directly — nothing gets rebuilt.
• iLEAN Edge — a terminal with vision over the equipment's inspectable surface after the CIP. It visually verifies residue, foam and soil before the technician goes in to take the swab. It works without a network.
• iLEAN Agent — recalculates the dynamic MACO before every changeover, with the real data of the next batch (size, equipment train, updated PDE toxicological figure). It classifies the cleaning by risk level under ICH Q9 and proposes a proportional sampling plan. QA decides and signs.

See the full IRIS architecture →

Before and after

Classic cleaning validation vs. cleaning validation with iLEAN

Aspect	Classic cleaning validation	With iLEAN Tracer + Writer + Edge + Agent
MACO calculation	Once, worst case, in a signed spreadsheet	Dynamic, recalculated for every changeover
Sampling plan	Same SOP for high and low risk	Proportional to risk under ICH Q9
Visual verification	Manual inspection with a flashlight	Edge with vision before the swab
Waiting on the lab	A linear bottleneck	Continuous verification + swab/rinse in parallel
Cleaning dossier	Signed spreadsheet in a folder	Living, Part 11 audit trail, inspection-ready
Changeover time	8-24 h blocking the line	≥30% less in the first 6 months

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• GMP plant making oral solids or sterile products, multi-product equipment (typically 8-15 products sharing critical equipment), weekly or fortnightly changeovers with 8-24 h of cleaning.
• First value expected within a few weeks: the first product pair migrated to dynamic MACO + living dossier + Edge on the inspectable surface.
• Expected reduction in changeover time: ≥30% in the first 6 months, driven by the continuous verification that removes the wait on the lab.
• Indicative payback between 4 and 9 months, dominated by the line capacity recovered during changeovers.
• An adjacent standard already running on continuous verification and dynamic limits: automotive quality demands 25 PPM ^[1]. That is the level multi-product pharma moves toward with continuous cleaning validation.

And the quality manager's reasonable doubt

“What if the AI recalculates MACO wrongly and we let contaminated product through?” — the rule is strict: the dynamic MACO is proposed with the evidence behind the calculation (batch size, updated PDE, equipment train, worst case ruled out and why); QA decides and signs. The cleaning is never released without a human signature. And in anchored tasks (reading a PDE from a document and recalculating a formula from the SOP) the best models brought error below 1.5% ^[2]. The three safety rings exist for exactly this.

^[1] Automotive quality standard ≈ 25 PPM. Source: Symestic.

^[2] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.