Against
the feed.

Almost every app on your phone is built to be opened again in an hour. Not as a side effect — as the objective. Somebody is measured on it, and the design follows the measurement, and that is how you end up with a product that is very good at being opened and not especially good at anything else.

We built one that is trying to be closed. This is what that means in practice, and what it cost.

The metric does not describe the product.
It decides it.

What the metric does

Pick daily active users and you will build streaks, because a streak is the cheapest way to make somebody return on a day they had no reason to. Pick session length and you will build an infinite list, because a list with an end is a place where people stop. Pick notification click-through and your alerts will start being about nothing in particular, at the hour when people are most likely to be bored.

Nobody in that story is cynical. Each decision is locally correct. The direction is set by what gets counted, and after two years of locally correct decisions you have a product whose actual purpose is to interrupt someone.

Why that is worse here than elsewhere

An app about two people in a room has a specific failure mode, and it is that the app becomes the thing being attended to.

If OOHLA is good at pulling you back, then at some point one of you is on it while the other is in the room, and the product has replaced the exact thing it exists to produce. There is no version of that we could count as a success. A relationship product that wins your attention has taken it from the only other party in the relationship.

So the goal is inverted. We want short visits, at the right moment, that end in the phone going down.

What we did not build

No feed. Nothing anywhere in the product refreshes into new content because you pulled it. Every room has an end, and you reach it.

No streaks, and no counter that resets. Missing a Tuesday is not a failure and the interface will never imply that it was.

No notification that we invented. Push says a count and a room — two things in Tease — and nothing more, and it only fires because the other person actually did something. There is no reactivation campaign, because a message engineered to make you open an app at half nine on a Sunday is a message about our numbers rather than about your evening.

No score. There is a room called Us that shows what the two of you have actually done, and it deliberately has no rating, no trend and no monthly comparison, because a number invites you to be behind on it.

The things we took out again

Some of this we got wrong first and had to remove, which is more instructive than the parts we got right.

There was a points balance. It was called sparks, it had a number, and it was spendable. We could not write a sentence explaining what a points balance in a relationship is for that any of us believed, so it went.

There was a pulse score and a five-month trend chart. The bar heights were an array in a file — the same for every couple, including the two years of history it generously credited you with on your first day. That is not a metric, it is a decoration shaped like one, and the honest version turned out to be better anyway: you have answered forty-one questions to each other is a line somebody remembers doing.

An invented number is worse than no number.
It is a lie with a chart.

What we did not conclude is that marking something is wrong. A thing the two of you genuinely did, recorded because it happened, is not a streak — it does not reset, it does not expire, and it cannot make you feel behind. The objection was never to noticing. It was to manufacturing.

What it costs

All of it, honestly. No engagement loop means no engagement, in the sense the word is normally used. There is no growth mechanism inside the product — nothing to share, nobody to follow, no reason for a third person to ever hear about it from the interface.

It also means we cannot tell you that OOHLA is used every day, because it is not supposed to be. Some weeks the right number of times to open it is zero.

The number we would rather have

Not visits. Not minutes. Whether the evening was different because of it — which is not instrumentable, cannot be put on a dashboard, and is the only thing that would actually mean the product worked.

So we are left doing the unfashionable thing: building for an outcome we cannot measure, and refusing the proxies, because every available proxy points somewhere we do not want to go.