AI can search data and past investigations far faster than people, suggesting likely causes for quality problems. Where AI genuinely helps root cause analysis, where it misleads, and how to keep investigations grounded in evidence.
A quality team investigating cracked housings asked an AI assistant to review the problem description, process data and two years of previous investigations. Within minutes, it suggested three likely causes, ranked by how often similar problems had occurred, and pointed to an investigation from three years earlier involving the same material and a similar crack pattern. That earlier report described exactly the cause the team eventually confirmed: moisture in the material combined with high moulded-in stress. The team still had to prove it. But they started in the right place, much as the AI-assisted NCR workflow did.
AI is becoming a useful assistant in root cause analysis. It is not a substitute for evidence.
Where AI helps
- Searching previous investigations, 8Ds, NCRs and lessons learned for similar problems.
- Analysing large volumes of process, quality and machine data to find correlations and patterns.
- Suggesting possible causes, organised in categories, as a starting point for a fishbone diagram.
- Drafting investigation reports and summaries from notes and data.
- Highlighting gaps: missing data, untested hypotheses or incomplete actions.
Where it misleads
- Correlation is not causation. AI finds patterns, but patterns can be coincidental or caused by a hidden third factor.
- Confident but wrong suggestions. Generative AI can propose plausible causes with no evidence, a risk we highlighted in ChatGPT for engineers.
- Bias towards common causes. Unusual problems may be missed if the AI favours what happened before.
- Poor data quality. Incomplete or inconsistent records lead to poor conclusions.
- Missing context. AI does not see the gemba: the condition of the tool, the operator's technique or the change that was not recorded.
Keeping investigations grounded
- Treat AI suggestions as hypotheses to test, not conclusions.
- Go to the gemba: observe the process, examine parts and talk to people.
- Prove causes with evidence: turn the problem on and off, compare good and bad conditions, use data properly, as described in root cause analysis beyond the 5 Whys.
- Keep a qualified person accountable for conclusions and actions.
- Record how AI was used in the investigation.
Preparing your data
AI-assisted analysis works best with well-organised data: consistent defect codes, process data linked to parts and batches, and investigation reports written in a structured format. Improving data quality pays off whether or not AI is used.
The right balance
AI can make investigations faster by finding relevant history and patterns quickly. People make investigations right by testing hypotheses, understanding the process and implementing effective actions. Together, they solve problems faster and more thoroughly than either alone.
