Separate the observation, interpretation, and next step
An observation describes the source data: six collected reviews mention trouble sharing a list. An interpretation explains why that might matter: a core task may be failing for some reviewers. A next step proposes a check: reproduce sharing with two accounts on the reported version.
Those statements have different levels of certainty. An exact count can be verified against a sample even when the explanation is uncertain. A recommendation should stay open to revision when you inspect the original sources or learn more about the app.
Know what public store data can tell you
Appfox’s current beta focuses on public App Store and Google Play evidence: listings, collected reviews, competitors, and tracked rankings. Research briefs and daily findings organize that evidence around an idea or an existing app.
A listing describes what a publisher presents publicly. A review describes one person’s account. A rank is an observation for a query and market at a time. None of these alone gives you a complete view of sessions, retention, revenue, or every potential customer.
The product preview shows the dashboard interface. Treat figures in marketing illustrations as examples of presentation, not audited revenue figures, evidence about the pictured apps, customer endorsements, or measured results from using Appfox.
Read the scope before the conclusion
When assessing a finding, check the store, country, language, dates, number of collected records, and freshness of the collection. Missing scope is a reason to investigate further. Do not fill in an unknown field from an assumption.
A collected sample can miss older reviews, other languages, countries, or records unavailable from the source. Collection limits can also change which observations appear. Compare equivalent scopes before describing a trend.
Google Play documents several review filters and notes that some reviews include device or version details. That is a useful reminder to distinguish available fields from missing ones; it is not a claim that Appfox has access to every field in an owner’s console.
- Same store, country, language, and collection rule in both periods.
- A source date and a collection date, where known.
- A visible numerator and denominator for each percentage.
- An explicit “unknown” when data was not observed.
What a theme count means
A theme is a grouping of reviews that discuss a similar subject. A mention count should count reviews, not repetitions of a word. If one review repeats “sync” four times, it is still one review mentioning sync.
The denominator matters: 6 of 12 collected reviews is 50% of that sample. It is not 50% of active users, paying customers, or installs. If reviews can belong to more than one theme, theme shares can add up to more than 100%.
AI classification can misread sarcasm, translations, mixed sentiment, or two similar-sounding problems. Inspect the original text before prioritizing a theme. The worked example uses hand-labeled fictional records so every label and calculation can be checked.
A before-and-after change is not a causal result
Compare periods with matching durations and collection rules. Record releases, campaigns, outages, pricing changes, and other events that may affect the result. A small sample can be useful for finding a concrete problem while remaining too weak for a broad conclusion.
A review theme rising after a release does not establish that the release caused it. A decline after a fix does not isolate the effect of that fix. Look for reproduction steps and independent evidence; use an appropriate experiment when you need a causal answer.
Rating summaries need their own context. Apple permits resetting an overview rating when releasing a new version, while written reviews continue to display. A comparison that ignores that reset can mix incompatible measurements.
What a research brief cannot establish
Reviews can reveal recurring frustrations, appreciated features, and language customers use. Listings can show positioning and visible offers. Together, they can help you choose a hypothesis worth testing.
They cannot by themselves establish market size, willingness to pay, competitor profit, or whether a business will succeed. An absent complaint does not prove that a problem is solved. A feature absent from a listing may still exist inside the app.
Take the strongest open question into an interview, task test, or small experiment. Record what evidence would change your mind before you invest in the full implementation.
What is confirmed in the beta
Invited users can use research briefs, daily findings, review themes, and competitor and rank tracking. RevenueCat, session replay, reply drafts, and API access are planned. This page explains how to assess evidence; it is not an API specification or a guarantee of refresh intervals, retention, or coverage in every market.
Ask about the specific store and country you need when discussing beta access. Prices and allowances shown on the pricing page are proposed, and no release dates are announced for planned capabilities.