After early testing shows Meridian's customer-service virtual assistant performing well on simple rebooking questions, an executive tells the board the chatbot will 'handle nearly all customer inquiries without human involvement.' The evaluation team's own criteria show it reliably handles only about 60% of real inquiry volume. What should the AI project manager do?

CPMAI's foundational task on why AI projects differ from traditional software explicitly includes managing stakeholder expectations to prevent overpromising and avoiding vendor/leadership hype outpacing what evaluation data actually shows. When an executive's public claim ('nearly all inquiries, no human involvement') diverges sharply from the measured evaluation result (about 60% coverage), the project manager's responsibility is to correct the record using the actual evaluation criteria and data, resetting expectations before the gap causes a credibility or operational problem — for instance, understaffing human agents based on a false coverage number. Letting the overstatement stand for morale reasons trades a short-term feel-good moment for a longer-term credibility and staffing risk when reality shows through. Narrowing the evaluation criteria to retroactively justify the claim is a form of gaming the evaluation and directly violates CPMAI's insistence on criteria tied to genuine business questions, not to whatever makes an existing claim look true. Canceling the project outright is an overreaction to a communication problem, not a technical one — the underlying 60% result may still represent solid, useful progress worth continuing to iterate on. The exam point: expectation management against hype is an ongoing PM duty across the CPMAI lifecycle, evaluation included, not just a Business Understanding-phase concern.