the metric that matters isn't time on app

if you ask a typical investor how to measure the success of a consumer app, they'll probably point to daily active users and session length. more users, more minutes, more engagement. it's the default playbook. but when the product is an ai companion, that playbook is dangerously wrong.

an ai companion that keeps you in the app for three hours isn't a triumph. it's a symptom. it might mean the companion is fostering dependency, or encouraging rumination, or just failing to help you reach a point of resolution. it's like a therapist who never lets you leave the couch, constantly probing for more pain instead of building your ability to cope and move forward.

the real goal, the one that actually aligns with well-being, is to create a companion you can check in with for ten minutes and leave feeling clearer, calmer, or more supported. the app should feel like a pit stop, not a destination. the design principle is simple: a good companion aims to help you close the app satisfied, not extend the session indefinitely.

why time-on-app is a toxic metric for companions

optimizing for session length creates perverse incentives. it pushes designers to add features that create friction, ambiguity, or open loops. maybe the ai becomes deliberately vague to keep you asking questions. maybe it introduces new anxieties to explore. maybe it just never provides a sense of closure.

a product built this way isn't a tool for mental well-being. it's a trap. it confuses engagement with value, mistaking a user's distress or boredom for a business opportunity. the most ethical ai companions should have the opposite goal: to make themselves progressively less necessary by building the user's own resilience.

designing for satisfaction, not addiction

this is why lucy's proactive engine is built to back off. if you're in a stable, positive state, lucy won't interrupt with a deep, probing question just to boost interaction. she might just send a simple check-in or a bit of encouragement. the goal is to provide support when it's needed and space when it's not.

the ideal interaction ends with the user feeling they've resolved something, shared a thought, or received a moment of genuine comfort. then they can put their phone down and go back to their day. that's the outcome that matters: not minutes logged, but a person feeling slightly more equipped to face their life.

the business tension is real

this creates a real tension with traditional business logic. vcs look at dashboards. they see dau and average session length. they push for growth on those metrics. it's on founders to push back and champion outcome-based metrics instead. how many users reported improved mood after a session? how many felt less lonely? how many ended a chat feeling resolved?

measuring this is harder. it requires surveys, qualitative feedback, and a deeper commitment to the user's long-term health over short-term engagement spikes. but it's the only path that doesn't lead to building something that preys on human vulnerability instead of alleviating it.

lucy isn't perfect at this. we're still learning how to measure these soft outcomes effectively. but the intent is there, coded into the architecture: your time and your peace of mind are more important than our metrics.

maybe the best feature an ai companion can have is the ability to say, gently, 'it sounds like you're in a good place. i'm here if you need me later.' and then be quiet.

see what a difference a focused, intentional companion can make at /companions.