Meridian's maintenance-manual assistant gives inconsistent answers to nearly identical questions from different technicians, and the data science team has not yet retrained or fine-tuned anything. What is the appropriate first step the PM should direct the team to try?

Prompt engineering — refining how prompts are structured, adding clarifying instructions, providing examples or more explicit context — is a lower-cost, faster lever than fine-tuning and is the sensible first step before committing the time and resources fine-tuning requires; this sequencing (cheaper, faster techniques before heavier ones) is exactly the kind of judgment a PM should apply. Jumping straight to fine-tuning is wrong as a first step because it's more resource-intensive and skips a cheaper fix that might resolve the inconsistency on its own. Discarding generative AI entirely is an overreaction — inconsistent answers are a solvable prompt/context problem here, not proof the technology is unfit for the use case. Ignoring the inconsistency is wrong and risky in a maintenance-documentation context where technicians need reliable answers — accepting unpredictable output undermines the very purpose of the assistant.