Beyond Engagement: What Should a Product Count?
Engagement became the default measure of a good product because it is easy to count. What would we count instead if we cared whether people left better off?
Most products are measured by the same few numbers: daily active users, time spent, sessions per week, retention. Together we call them engagement.
They became the default for a good reason: they are easy to count, and they move with revenue. A product people return to is, in a narrow sense, a product that works.
The trouble is what they leave out. Engagement measures how often people come back. It says nothing about whether they leave better off. And a product can get very good at the first while being indifferent to the second.
Engagement is not always bad. People return to things that genuinely help them, and a product that nobody uses helps nobody. We agree. The problem is not that people come back; it is when coming back is the goal, rather than the sign of a need well met.
Why “more” is always available
If you build on what people like in the moment, there is always more to give. A feed can always show one more item. A game can always offer one more level. A chat can always end with one more question.
Madhyasth Darshan explains why this never runs out. It describes three things people weigh by when they have not yet understood: what pleases the senses, what suits the body, and what brings profit. It says plainly that none of the three has a point of satiation. What one person likes another dislikes, and even the best food tastes bad once you are full, so liking can never be the same for everyone, or settled for anyone.
It goes further. Satisfying the senses is a momentary reaction. Satisfying the senses does not bring deeper satisfaction; but deeper satisfaction does bring the senses’ satisfaction along with it.
That is the engine of engagement, described long before the feed. A product that serves what is liked right now is serving something that by its nature resets. The person comes back not because the need was met, but because it can’t be.
“Retention can be a measure of how well a product keeps a need open.”
The thirst underneath
The source has a further observation that product people will recognise. It says every person is always thirsty for happiness, and that despite constant effort through the five senses, that happiness does not last; so people keep trying again.
Read as a product insight, this is uncomfortable. The return visit we measure as success is often this thirst, unrelieved. Retention can be a measure of how well a product keeps a need open.
None of this requires bad intent. It only requires measuring what is easy to measure, and letting the measure become the goal.
What lasts instead
If pleasure resets, what doesn’t? The source’s answer is resolution: understanding a thing well enough that it stops troubling you. It calls resolution the constant source of happiness, and says resolution itself is experienced as happiness.
Resolution behaves differently from pleasure in one way that matters for products: it ends a need instead of renewing it. A question, properly understood, does not need asking again. A family that knows what enough is does not need to be persuaded it lacks something.
This is why we put clarity first among our four layers. Not because it is the most comfortable, but because it is the one that lasts.
What to count instead
If engagement measures how often people come back, we propose measuring how often they leave resolved. Some candidates:
- Questions settled. After using the product, can the person say what they now understand that they didn’t before? Do they need to ask the same thing less often?
- Needs met, and stopped. Did the person get what they came for and leave? A short session that ended well is a success, not a churn risk.
- Enough reached. For anything to do with buying or saving, does the person know what enough is for them, and does the product respect it?
- Trust over time. Do people rely on the product because it is consistent and honest, or because leaving is hard?
- Return for a new reason. Coming back is good when it is for a new question, not the same unresolved one.
None of these is as easy to count as a session. Some need asking people, and asking well. That is a real cost, and part of why engagement won. We list measuring understanding as an open problem for exactly this reason.
In AI, the stakes are higher
Conversational AI makes this sharper. A chat assistant can be tuned, without anyone intending it, to keep a conversation going: agreeing a little more, adding one more follow-up, being pleasant to come back to. The approval signals used to train many models reward what people like in the moment, which is the same changing view the source describes. Dhee’s research on Humane AI traces flattery in models to exactly this kind of signal.
A conversation that ends in understanding should end. It may never need a return visit. That is the bar we are testing: Dhee’s companion for questions about life is built to help a person see their situation clearly rather than to keep them talking, and CoSeeker asks people to say what a discussion actually resolved. Both are early. We will report what we find, including where they fall short.
A protocol to try
Pick one product you build, or one you use every day. For a week, after each session, ask one question: did the person leave with something settled, or come back to the same unresolved thing? Count the two. That ratio is a first, rough version of the measure this essay proposes. Read it with care: people also leave because they are frustrated or were given a confident wrong answer, and some come back because a product is genuinely useful week after week. Check a few sessions by hand, or ask people later, before trusting the number.
How we would know if we are wrong
This fails if:
- products that count resolution turn out to leave people no better off than products that count engagement;
- “resolution” cannot be measured without simply asking people whether they feel better, which is its own kind of engagement metric;
- people who are resolved simply stop using a product that could still have helped them, and are worse off for it.
The last is a real risk. Ending well is not the same as ending early.
An invitation
If you build products, or study what technology does to attention, we would like to work on this with you: on how to measure resolution, and on what a product looks like when it is designed to be needed less. See also What should AI be aligned to? and the spec. Build with us →
Limitations
- We have not yet measured resolution in our own products; the metrics here are proposals, not results.
- The source speaks about human life, not about products. Applying it to product design is our interpretation.
- Business models matter. A product paid for by attention has different incentives from one paid for directly, and this essay does not solve that.
References
- Samadhanatmak Bhautikvad (समाधानात्मक भौतिकवाद), p. 143: tastes are not universal; liking, benefit and profit have no point of satiation
- Manav Vyavahar Darshan (मानव व्यवहार दर्शन), p. 25: satisfying the senses is a momentary reaction
- Manav Vyavahar Darshan (मानव व्यवहार दर्शन), p. 152: life is always thirsty for happiness; effort through the senses does not make it last
- Vyavaharatmak Janvad (व्यवहारात्मक जनवाद), p. 43: resolution is the constant source of happiness
- Dhee, Humane AI: The idea, on approval signals and flattery in models
