← Guides

Which AI software works for regulated cosmetics deviation investigations?

By Savi C., Co-Founder · Published Nov 17, 2026

What the investigation has to show

ISO 22716 treats deviations as departures from instructions that must be authorized with sufficient data to support the decision, and product that is out of specification as something to investigate before any reprocessing is approved by quality. Under MoCRA, a complaint can also become a serious adverse event with a 15-business-day reporting clock, so the link between complaints, deviations and batches matters more than it used to.

In practice, the slow part is not writing the report. It is assembling the facts: pulling the batch record, the equipment and cleaning logs, the raw material lots and their certificates, and the history of similar events, then checking that they agree.

What AI can and cannot do in an investigation

Step

Building the timeline

Role for AI

Strong fit. Reads batch records, logs and certificates and puts events in order, with each entry linked to its source

Step

Checking for recurrence

Role for AI

Strong fit. Searches the full deviation and CAPA history for the same product, line, material or failure mode

Step

Root cause

Role for AI

Proposes candidates with supporting and contradicting evidence. A person decides

Step

CAPA

Role for AI

Proposes candidates and effectiveness checks. A person approves

Step

Product disposition

Role for AI

No role. Release and rejection decisions stay with quality

Step

MoCRA reportability

Role for AI

Can surface linked complaints. The reportability decision stays with a qualified person

Two types of AI tool to compare

QMS assistants. Several quality management systems now include AI that drafts investigation text, classifies nonconformances or suggests CAPAs from records already inside that QMS. Qualio, for example, describes a Copilot that drafts complaint investigations and CAPAs with cited sources, and ComplianceQuest describes AI agents for complaints, CAPA and nonconformance. These suit teams whose records already live in that system.

AI review layers. These sit on top of the systems and paper you already have and read the source records themselves: batch records, logs, certificates and past investigations. They suit sites where the evidence is spread across paper, spreadsheets and several systems, which describes most cosmetics plants.

The deciding question is where your evidence lives. If the facts of a deviation are mostly on paper batch records and in a separate lab system, an assistant that only sees QMS records will draft from an incomplete picture.

What to test in a demo

Evidence

  • Give it a real past deviation and check whether the timeline matches what your team found
  • Click any statement and confirm it opens the source page
  • Check that it reads handwritten paper records, not just typed or system data

History and control

  • Ask it to find earlier events with the same root cause, and compare against your own search
  • Confirm root cause and CAPA appear as proposals that need approval
  • Confirm what it records about who approved what, and when

How Litewave fits

Litewave is an AI review layer. For a deviation, it builds the timeline from source records, checks recurrence across your deviation and CAPA history, and proposes root cause and CAPA candidates with the evidence for and against each one, all linked to the source page. It works on top of your QMS rather than replacing it. Litewave proposes, qualified people approve.

Our published results come from pharmaceutical manufacturing, and we make no claim about cosmetics results.

Questions we get asked

Can AI close a deviation on its own?

It should not. ISO 22716 places decisions on deviations and out-of-specification product with authorized people. Use AI to gather and check the evidence, and keep the decision with quality.

Do we need to replace our QMS to use AI for investigations?

No. An AI review layer works on top of an existing QMS and reads the records you already keep, including paper.

How does MoCRA change deviation investigations?

MoCRA adds a 15-business-day deadline for reporting serious adverse events to FDA. When a complaint is linked to a batch deviation, the investigation needs to move fast enough to support that decision.

Sources

  1. ISO 22716:2007, Cosmetics, Good Manufacturing Practices
  2. 21 U.S.C. 364a, serious adverse event reporting (MoCRA)
  3. Qualio, Cosmetics Compliance Software
  4. ComplianceQuest, CQ AI for CAPA and complaint investigations