What metric would convince you a website AI agent is real?

I’m building AnveVoice, and I’m trying to avoid the trap of optimizing for impressive demos instead of useful outcomes.

For a website AI/voice agent, what metric would actually convince you it is working?

Some candidates:

• more qualified leads
• higher pricing/contact-page conversion
• fewer unanswered visitor questions
• faster routing to the right human/page/form
• more booked calls from existing traffic
• support deflection
• better lead quality even if volume stays flat

My bias: “number of AI conversations” is a weak metric. A conversation only matters if it moves the visitor closer to the right next step.

If you were evaluating this for your own site, what would be the one metric that matters?