AI can do substantial work leading up to a quality decision, but qualified people should keep final judgment, authorization, and regulated accountability. Ovada gathers evidence, follows approved steps, flags missing or conflicting information, and prepares recommendations for human review.
What should remain with qualified people
- Interpreting ambiguous or conflicting requirements
- Assessing product, patient, or consumer risk
- Judging incomplete evidence
- Investigating unexpected conditions
- Accepting residual risk
- Authorizing and signing a regulated outcome
No generic confidence score should hide these decisions or imply that model output replaces the designated authority.
Deterministic checks and contextual interpretation
Explicit checks, like comparing a result with a numeric limit or confirming a required field, can use deterministic rules. Contextual work, like deciding whether two records conflict, can use model intelligence, with uncertainty surfaced for review.
Agent · incoming material · lot MG-26-0412
COA-2026-0671Rules handled the repeatable checks. AI drafted the interpretation. A qualified person makes the disposition.
What a reviewer needs to see
- The source evidence used
- The approved requirements applied
- The steps completed
- Missing or conflicting information
- The proposed recommendation
- Open matters that need judgment
- Who is responsible for authorization
Traceability is worth more than a claim that the AI is “confident.”
Two examples
In incoming material disposition, the agent extracts CoA results, compares them with the active specification, flags missing tests, and prepares a recommendation. The qualified reviewer decides whether the evidence is enough and signs the disposition.
In deviation management, the agent rebuilds the event timeline, gathers related records, and drafts the deviation record, including whether CAPA should be considered. A qualified person decides the conclusion and signs.
Questions
Can an AI agent sign a quality record?
Not in Ovada. The agent prepares and routes the work. Only a qualified person can apply the electronic signature that authorizes a regulated outcome.
What happens when the AI is uncertain?
Uncertainty, missing evidence, and conflicting requirements are escalated to a qualified person with the source evidence attached, instead of being resolved silently.