Among the metrics AI companies rarely disclose at launch, one would be more useful to prospective users than almost anything currently in a typical product announcement: time-to-first-value, meaning how long it actually takes a new user to get something genuinely useful out of the product, rather than how capable the product is in the hands of an expert who already knows how to use it well.
Why this metric matters more than raw capability
A product can be highly capable in principle and still fail broadly if realizing that capability requires extensive setup, prompt engineering skill, or trial and error most users won't invest before giving up. Time-to-first-value captures the actual user experience of adoption in a way that a benchmark score or feature list doesn't.
Why companies don't disclose this
Unlike a favorable benchmark score, time-to-first-value is a genuinely hard metric to make look good if a product has a steep learning curve, and there's no strong incentive to disclose a number that might discourage prospective users before they've even tried the product.
How to approximate it without company disclosure
Independent reviews and early user reports that specifically describe the onboarding experience - not just eventual capability once mastered - are the closest available substitute, though they require more digging than reading a launch announcement's capability claims at face value.
What this means for evaluating a new launch
A product announcement heavy on capability claims and silent on onboarding experience is giving you half the picture needed to predict whether the product will actually get adopted by the people it's aimed at, versus impress a narrower audience of power users willing to invest in learning it.