Trial feedback keeps finding users who never knew the feature existed. The fix we're testing: 30 to 90 narrated seconds of the product doing real work.
Feedback from our newest users keeps pointing to the same humbling finding: plenty of them never knew the feature existed. Not confusing, not underpowered. Invisible. A colleague made a great observation that for something to be confusing at least needs a baseline to interpret (confusing compared to what?) But if something’s invisible, you don’t need to interpret anything. You can’t form an opinion about a thing you’ve never seen, and you can’t adopt it either.
AI features have it worst, because they’re invisible twice. Once in the product, where an entry point is easy to scroll past, and once in the mind, where an open-ended box offers no clue what to ask it. We keep describing this as the don’t-know-what-to-ask problem, and it explains adoption better than any usability finding we’ve collected this year. Capability without a visible first move is a locked door with no handle.
So here’s the new approach we’re testing: ship a demo with the feature. 30 to 90 seconds of narrated screen capture showing the feature doing real work on a real task, delivered as a dismissible takeover the first time a new user lands. Not a tour, not tooltips. Most users close in-app tips reflexively, which is exactly why the cold-start moment deserves something with a voice and a story instead of another blue dot.
A feature nobody can see has the same adoption curve as a feature you never shipped.
The production bar is deliberately low. We already have recordings from live product walkthroughs; the plan is a title card and a trim, not a marketing cycle. The precedent I keep pointing the team to is the building-in-public launch style that companies like GitLab, Linear, The Browser Company, and Twitch used so well: the PM and the designer who made the thing, narrating what it does, in one take that ships the same week. Polish is optional. Existence is not.
The shelf
And when skeptics shrug at first contact, that’s expected rather than fatal. The pattern we keep citing internally is Figma’s auto layout: plenty of designers bounced off it early, and the same people now can’t work without it. Detractors become converts once they see the value, and the operative word is see. The curve doesn’t start until the seeing does.
One timing note, because distribution is a design problem too. For our users, the morning sign-in is the wrong moment; clinicians go straight into sessions and dismiss anything between them and their day. Right clip, wrong moment, same invisibility. Picking the moment deserves the same care as picking the frames.
The teachable part
Ship a demo with the feature: 30 to 90 narrated seconds of it doing real work. Users can’t adopt what they can’t see, and AI features are invisible twice.