App health

4.3 stars can be mostly fives with a non-trivial pile of ones, and those are two entirely different product problems. The histogram shows it at a glance.

What the one- and two-star reviews actually say is collected theme by theme on a separate screen: the work nobody in the category does well shows up there.

An app's standing on the App Store does not fit into one number, but it does fit into a few measurements. The health screen gathers them in one place, because looked at separately the signals that explain one another get missed.

The first is band spread: of the phrases you track, how many are in the top 10, how many at 11-30, how many outside 200. Those three say what your visibility is today, what it will be next month, and how much ground has no relevance yet.

The second is metadata fill: how many countries have a keyword field written, how many characters are being wasted, which fields were never filled at all. An empty country field produces no warning anywhere - this is the only place that produces one.

The third is the rating curve: how the score and count move over time. It does not carry rank directly, but a drop after a release is the fastest signal that something broke in that release - and it weakens your strength across every phrase at once.

The fourth is coverage: how many countries are measured, how many phrases tracked, when the last measurement ran. The measurement is itself something to watch; gaps break comparisons.

The reason these sit on one screen is that movement in one is usually explained by another.

If you lost a few places across every phrase in the same week and did not touch metadata, the first place to look is the rating curve. If the score fell after a release, that is the cause and metadata will not compensate for it.

If your count of top-10 phrases is shrinking while the 11-30 band grows, you are giving back ranks you already won. That usually happens silently, because nobody looks at something already won - and the cause is often the side effect of the last metadata move: a word you removed from the subtitle was holding up the phrases it appeared in.

If the count of phrases outside 200 never falls, relevance is not being built. Either the metadata does not describe what the product actually does, or the phrases you are targeting do not match the product.

Reading band spread while metadata fill is low is misleading: saying "our ranks are poor" while half the fields are empty is judging the result of work that has not been done.

The most necessary section of a health screen is the one listing what it does not show.

No download count. Apple does not give it to third parties; the estimates in circulation are computed backwards from rank and their error bars vary widely by category. No revenue either, for the same reason.

No conversion rate. How many of the people who saw your page installed is visible only in your own developer account. We can tell you that your rank rose while your downloads did not - we cannot tell you how much you lost.

No single health score. "Your health score is 74" does not say what work to do. Instead each measurement stands on its own with what to do about it beside it.

Where a value could not be measured, a short dash sits. Zero is a reading and people decide on readings; writing zero in place of the unknown makes them decide wrongly.

This screen is not something to sit in front of. Once a week, fifteen minutes, and those fifteen minutes have an order.

Band counts first: how did the top 10, the 11-30 band and the outside-200 group move against last week? Then the control group: what did the phrases you did not touch, and your rivals, do - did the movement come from you or the market?

Then the defended band: did one of your top-10 phrases fall? If it did, the cause is usually the last move's side effect, and caught immediately it can be undone.

Metadata fill last: did any characters open this week, is a country still empty? Those four questions produce a one-sentence output, and that sentence decides the week's move.