Against engagement

What makes a person open an astrology product tomorrow morning? In the ordinary case, whatever this morning did not settle. A question that closes produces no second visit. A question left slightly open produces several. Which is the trouble with the number almost every product is built to raise: how many people open it on a given day.

The trade writes it DAU, for daily active users. Divide the daily number by how many people opened it at least once that month, and you get what dashboards call stickiness. Engagement is the family name for all of it. We will count opens, because nobody runs a website while counting nothing. We will not make the count a target.

The refusal list gave this one line: we are not designing for time spent. The argument was owed to you at length.

Grant the metric its due, because almost everywhere else it earns it. A maps application opened daily is being used. Frequency and usefulness move together, so watching the number is watching a rough picture of whether the work landed. It is cheap, it arrives inside the hour, and it is usually not lying.

Astrology is the exception. Here the frequency number reads out how much unfinished business we sent people home with, and raising it means sending them home with more. A calendar you check hourly is not doing a calendar's work. It is a fear rented by the day.

A reader who comes back is not a failure. Somebody reading two essays a week because they are worth reading is what I want. The trouble starts when the return is engineered rather than earned, because the cheapest lever for engineering it is the one that leaves you unsettled.

Nobody sits in a room and decides on that. A roadmap is a list of things to try. Whatever makes the number go up survives the next review; whatever does not gets cut. Two years of honest reviews by decent people, and the product has quietly been shaped to leave readers with their questions half open. Nobody chose it.

The failure has an old name. Charles Goodhart, writing about British monetary policy in 1975, pointed out that once you start using a measurement to control something, the measurement stops behaving the way it used to. The short version most people quote came from Marilyn Strathern in the European Review in 1997, writing about how British universities were assessed, and she credited the phrasing to Keith Hoskin: once a measure becomes a target, it stops being a good measure.

The second problem is counting, not incentives. An open is countable: it carries a timestamp, costs nothing to record, and by Tuesday it is a line on a chart. Whether a reader closed the page with their question answered is not countable at all. One side of the argument brings a number to the meeting and the other brings a paragraph. The number wins, in every company, without anybody being dishonest. It fits on a slide.

What we count instead

Resolution: the reader leaving with the thing they came for and no reason to come back this week. A decision they can now make. A fear returned to its right size. A sentence they can repeat to the relative who frightened them.

I would like to tell you we measure that. We cannot, and I will not invent a number to sound rigorous. What we have are four weak proxies, worth naming as weak. Replies showing that somebody worked something out, which can also be politeness. Questions that stop coming back, which might mean an essay worked and might mean the reader gave up on us. Movement between an essay and its plain-English gist, which says only that somebody wanted both registers. Refunds, the one of the four that costs us money to be honest about.

Four weak numbers pointed at the right thing beat one strong number pointed at the wrong one. That is the trade I am making, and it is not a comfortable one.

What the refusal takes out

There is no streak here, and nothing else that counts your visits back at you. A streak makes a run of days feel like something you own and could lose, a small penalty for staying away, and the day you skip us should not feel like a mistake. Push notifications go the same way, all of them: we already refuse to alert you when a planet arrives somewhere, and the wider rule is that we will not interrupt your day to tell you anything, including that we have published something. Nor is there a red dot or a counter of unread items, because a badge is an errand somebody else set for you, and clearing it becomes the entire product.

When an essay ends, the page ends. Nothing plays next, no queue choosing your next twenty minutes, no list that refills as you reach the bottom.

Posts appear about twice a week, slow enough that there is nothing to check hourly. The way to follow today is the feed, which asks nothing of you. When the email list opens it will carry one mail per new essay, and the archive is never mailed.

The form all of this borrows from is the pañcāṅga (पञ्चाङ्ग), the almanac sheet for a day. It reports the day and stops. It has never once asked after you.

What a design settles before anybody believes anything is Dhruv's subject: a sun sign on a dating profile is a display setting before it is a belief, which he takes apart in the sign in your bio. Mine is the builder's side: which of these get built, and what not building them costs.

What it costs

Advertising is the one revenue line where opens turn straight into money, and we have postponed it rather than refused it, which makes this refusal cheap for us today. A company funded by attention could not publish this essay, and I would rather say so than take credit for a discipline I am not being tested on.

The expensive part is churn, meaning the rate at which paying customers stop paying. Our plan needs about 190 people paying a monthly fee and 20 guides sold in a month; that arithmetic is already published in full. Frequency appears nowhere in the sum. Churn appears everywhere in it, and churn is where a product built to be opened rarely gets punished.

Here is the shape of that, in a rule I can state exactly. Since a Reserve Bank of India rule of August 2019, a recurring card payment has had to warn the customer at least 24 hours before each charge, and the customer can cancel at any time, after which nothing further can be taken. So when the membership opens, every member will get 12 reminders a year that they are paying us, for a site we have built them to open less often than that. Twelve prompts to reconsider, from a product that will never prompt them at all. I know which way that runs.

Take the simplest thing bhagol sells or ever will: a written guide. You buy it once, it arrives by email, and it never needs you to come back. The monthly fee is the hard case.

One company has published what a decision like this costs, at a size that could absorb it. Reporting its results for the last quarter of 2017, Facebook said changes made to favour meaningful interaction over passive consumption had cut time spent on the service by roughly 50 million hours a day. It could print that because it had 1.40 billion people opening the product daily. We are looking for 190 paying readers.

There is one number I want and will never have. How many people read one thing here, settled the question they arrived with, and had no reason to come back for a year. No analytics package reports it, and every report that does arrive will read as a bad month.

A flat line on that chart, with the essays still being read and the guides still selling, is the best news this company could get. Nobody has drawn that chart yet.