AI-Assisted Requirements Review: Two Years of Production Experience at Airbus
Airbus has published a detailed retrospective on two years of deploying AI-assisted requirements review tools across multiple programs. The system reduces review time by 40% while improving consistency, but requires careful calibration to avoid anchoring bias.
Airbus AI Requirements Review: Two-Year Retrospective
Airbus started deploying AI-assisted requirements review across programs in early 2024. Two years of production experience is enough to draw meaningful conclusions.
What works: Consistency checking (identifying requirements that conflict with or duplicate others), quality checking (identifying ambiguous language, missing acceptance criteria, non-testable requirements), and cross-reference validation (checking that requirements correctly reference applicable standards).
40% time reduction: The headline number covers initial review pass time. Expert review time on flagged issues was not significantly reduced — the AI identifies where to look, humans make the judgment calls.
Anchoring risk: The most important finding for AI tool design. Reviewers tend to accept AI assessments without independent analysis. Mitigation: present AI findings after reviewers complete their independent review, not before.
Where it fails: Novel requirement types with no training analogues, requirements involving customer-specific context not captured in the training data, and requirements whose quality depends on understanding program-specific technical decisions.