What Apple's analytics reports look like on a real nine-app account
Two weeks of building an App Store Connect client, and what the reports actually contain.
We spent a fortnight building AppStato, an iPhone app that reads your own App Store Connect numbers, and most of that fortnight went on one thing: learning what Apple’s reports actually contain, as opposed to what you assume they contain the first time you parse one.
Here is the short version. Apple’s analytics reports are not a ledger you append to. They are snapshots that replace each other. If you add them up, your numbers will be two to three times too big, and they will look plausible the whole way. That is the mistake we made, and Apple documents the rule we had missed.
Everything below was run against one developer’s real account of nine apps in November 2026. The figures are deliberately given as ratios and percentages rather than counts, because they are somebody’s real business; where a worked example needs numbers, they are invented and labelled.
Reports restate, and you must not merge them
Request an ongoing report and Apple generates a new instance each day. Each instance carries not just yesterday, but the days before it as well, restated. One day in our test account was restated twice: it read one value in the first instance, higher in the second, higher again in the third. None of those readings was wrong. The later one replaced the earlier one.
We had been summing. Summing inflated traffic by two to three times.
Apple’s own documentation is unambiguous once you find it, on the Data Completeness and Corrections page:
Instances from a more recent
processingDateoverwrite instances with an earlierprocessingDate. […] Don’t merge records for a Date of2024-03-30across two report instances to avoid double counting. Use the latest instance.
So the rule is: key your rows on (app, report family, data date) and keep only the row from the
instance with the newest processingDate. Not the newest file you downloaded, and not the newest
row you read — the newest processing date, which is a field on the instance.
How far back a report restates is per report, and Apple states it on each report’s page.
| Report | Documented completeness | What we saw |
|---|---|---|
| App Store Downloads | Within two days | Instances carried the two days before the processing date |
| App Store Discovery and Engagement | Within three days | Instances carried the three days before |
That matched exactly, which was reassuring: the behaviour that cost us two days of debugging is written down, with the right numbers, on the page for each report.
One more line from the same page, worth knowing before you cache anything as final:
In rare cases of data corrections, you might receive new record sets for a date in the past, beyond the completeness threshold.
What matched App Store Connect, and what did not
Once the restatement rule was in, we reconciled the API against the App Store Connect web interface for the same dates. Two metrics matched exactly, every day: first-time downloads and product page views. Impressions did not.
| Metric | API against App Store Connect |
|---|---|
| First-time downloads | Matched exactly, every day, on both apps compared |
| Product page views | Matched exactly, every day |
| Impressions, steady iOS app | About 1.4% below App Store Connect |
| Impressions, low-volume macOS app | About 39% below |
The last row is the interesting one. A 1.4% gap looks like rounding. A 39% gap looks like a bug, and we went looking for a missing report — the App Store Web Preview report has no impressions, so that was not it either.
It is not a bug. Apple says so on the help page for downloading analytics reports: rows are omitted where the data comes from fewer than five users or five unique devices, some reports are not generated at all if minimum thresholds are not met, and “a small amount of statistical noise is added to metrics”.
An app with a handful of impressions a day spread across territories, devices and sources is mostly made of rows below that threshold. A busy app is not. The gap is not an error; it is privacy protection, and it costs small apps far more than large ones.
We show impressions anyway, and say this in the definition rather than quietly presenting a number we know runs low.
Product page views are a choice, not a field
There is no “product page views” column. There is an Event column holding Impression,
Page view and Tap, and a Page Type column with ten or so values. We count page views where
the page type is Product page or Store sheet, and exclude App privacy, Developer page, App
version history, In-app event, Media view and the rest. That is what matches App Store Connect day
by day.
It is a reconstruction, though, not a definition Apple publishes. It is the first thing we would re-check if Apple adds a page type.
Two systems, two time zones
App Analytics and Sales and Trends do not agree about what a day is.
App Analytics data is based on Coordinated Universal Time (UTC). By default, Sales and Trends data is shown in UTC, but you can change the timezone to Pacific Time (PT).
And on the sales side, about matching the dashboard to what you can download:
To match the timezone used by downloaded reports, select Pacific Time (PT).
So the downloaded sales reports — the ones an app like ours actually reads — are Pacific, and the analytics reports are UTC. A “yesterday” built from both sources is two different yesterdays, eight hours apart.
We resolve each window against its own source and label it, and we do not draw a funnel from impressions through to revenue, because the two ends are measured on different clocks. Apple also notes that transactions “may not be reported on the same dates in App Analytics and Sales and Trends due to processing time differences between the two systems”, which is a second reason not to join them.
A quiet day and a missing day are not the same thing
A day covered by a report with no matching rows is a zero. A day no report covers is unknown. Collapsing the second into the first invents a flat line, and flat lines are what people screenshot.
So we carry three states. A window where every day has a report shows a value and a change against the previous window. A window with gaps or suppressed rows shows the value with “Limited data” and no percentage change, because a change measured against an incomplete baseline is a made-up number. A window no report covers says “Not available”.
Example data, to show the shape of it:
| Day | Report covers it? | Rows | Shown as |
|---|---|---|---|
| Monday | Yes | 12 | 12 |
| Tuesday | Yes | none | 0 |
| Wednesday | No | — | not counted; the window is marked “Limited data” |
Those three numbers are invented. Every figure on this page that comes from the real account is a ratio, a percentage or a count of readings.
Two things that cost us an afternoon each
An HTTP 500 means you typed your vendor number wrong. The sales endpoint answers a wrong vendor number with a 500, not a 400 and not a 404. We read it as an Apple outage and waited. The app now says, on a 500 from the first sales request, that the vendor number is the likely cause.
The first request for a report type needs an Admin key. Apple’s help page is explicit: “An Admin role is required to request a new Analytics Report type for the first time. Once a report type has been requested, API keys with Sales and Reports or Finance roles can download generated reports.” The first report then arrives roughly 24 to 48 hours after the request, not immediately — so a new user’s first run genuinely has nothing to show, and an app that cannot say so clearly will look broken.
Related: an ongoing request starts from the day you make it. History comes from a one-time snapshot, and on request Apple has analytics data from 1 January 2024. We read both, and prefer the latest instance from either.
Eleven currencies
Sales rows arrive in the currency the customer actually paid in. In this account that was eleven currencies, most of them carrying free downloads with no proceeds at all. Apple’s own note: “All downloadable reports display amounts in local currencies as transacted by customers”, while the Sales and Trends dashboard “estimates the USD amount of sales and proceeds […] based on a rolling average of the previous month’s exchange rates”.
We do not convert. A single in-app purchase in our reconciliation appeared as one figure in euro in the downloaded report and a different figure in US dollars in the dashboard — the same transaction, two currencies, and an exchange rate in between that Apple describes as an estimate. Converting would have made a tidy total out of an estimate of an estimate, so we show per-currency subtotals and let the number stay true.
Apple’s own caution, from the same page, is a good one to end on:
The Sales and Trends dashboard shows near-real-time estimates, while downloaded reports contain finalized data. […] Avoid comparing totals directly between the dashboard and downloaded reports.
If you are writing one of these
- Key on
(app, report family, data date)and keep the newestprocessingDate. Never sum instances. - Store the instance and processing date as provenance. When a number changes, you will want to know which reading it came from.
- Keep a zero and a gap apart, and refuse to show a percentage change over an incomplete window.
- Expect small apps’ impressions to read low, and say so in the interface rather than in a footnote.
- Resolve each window in its own time zone, and label it.
- Treat an HTTP 500 on the first sales call as a wrong vendor number until proved otherwise.
All of this went into AppStato, an iPhone and iPad app that reads your App Store Connect account with your own API key and keeps everything on the device. The longer how-to versions of several of these are in the AppStato guides.
Sources, all read on 17 November 2026: Apple’s Data Completeness and Corrections, App Store Downloads and App Store Discovery and Engagement report pages, and the App Store Connect help pages on downloading analytics reports, differences in reporting tools and units, proceeds and sales.
