AI Agents for Manufacturing

QualityPilot Pilot, prototype, early rollout or time-limited trial; results are limited or not yet measured.

AI-assisted deviation investigations, root cause and CAPA

Added 2026-10-13 · Last updated 2026-10-14

Role

Quality · Typical users: quality manager

The problem

Every deviation or nonconformance needs an investigation: assess product impact, find the root cause, look for trends and write a CAPA that will hold up with customers and auditors. The knowledge to do it well exists in years of past investigations, but investigators cannot search it quickly, classifications differ between sites, and many sites write in languages other than the customer's. Catalent, a CDMO with nearly 40 sites, describes investigations as among its most critical and resource-intensive processes.

Step by step

  1. Export 2–3 years of closed deviations, NCRs and CAPAs with their root causes and effectiveness checks; clean obvious duplicates.
  2. Give the assistant read-only search over that history (retrieval), so every suggestion cites past records.
  3. For each new event, have it propose: similar past events, a severity classification with reasons, possible root causes and data to check.
  4. The investigator runs the real investigation (5 Whys, fishbone, data) and uses the assistant to draft the report and CAPA text.
  5. A quality leader reviews and signs; record that review in the QMS.
  6. Use the same history to prepare audits: list open CAPAs, overdue effectiveness checks and repeat issues by clause (ISO 9001 / IATF 16949).
  7. Measure investigation cycle time, repeat deviations and CAPA effectiveness before and after.

Tools

  • Catalent Qai on Microsoft Azure OpenAI / Foundry, Azure App Service, Microsoft Fabric
  • QMS platforms adding AI: ETQ, MasterControl, Veeva, Sparta TrackWise, Intellect, Qualio
  • Any enterprise LLM with access to your past 8D/CAPA reports for a pilot

Result

Catalent (contract drug manufacturer, nearly 40 sites on four continents), Qai built on Azure OpenAI (GPT-4 and GPT-4o) with Microsoft (Sept 2026): retrieval over the network's deviation and CAPA history, consistent minor/major/critical classification across sites, first drafts in English for non-English sites, human validation documented for every output, read-only and no creation or modification of GMP records. Rollout was nearly complete, with early signs of productivity gains in deviation management; Catalent says the figures are still being measured, and audit management is next. No percentage has been published yet (Microsoft customer story).

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

  • No figures yet: Catalent has not published measured savings.
  • In regulated industries (GMP, medical devices, IATF), AI tools that touch quality records need validation; keep them read-only at first.
  • The assistant suggests causes; it does not prove them. The root cause must come from evidence.
  • Your history is only as good as past investigations. Poor root causes in the archive will be repeated.

Try it yourself

You are a quality engineer (ISO 9001 / IATF 16949).

Inputs: (1) the new nonconformance description, (2) up to 10 similar past NCR/CAPA records with root cause and action [pasted].

Tasks:
1. Suggest a severity (minor/major/critical) with reasons.
2. Compare with past records: which look similar and what fixed them; is this a repeat?
3. List possible root causes (people, method, machine, material, measurement, environment) and the data needed to confirm or eliminate each.
4. Draft containment actions and a CAPA plan with owners and an effectiveness check.
5. Draft the report in [language].
Output tables; quote the past record behind each comparison; mark everything as hypotheses to be verified; write 'not found' when the records do not show something.
Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Want more like this? Join the newsletter

Related cases

Sources

Weekly: 1–2 practical AI use cases for your plant

Newsletter in

Free, one email a week. By subscribing you agree to receive the weekly newsletter (and its sponsor message) from AI Agents for Manufacturing. Every email includes an unsubscribe link, and you can unsubscribe anytime. See our Privacy Policy.