Analyze and Explore Opportunities on Google Play

Publishing is the beginning. The next question is: are people finding my listing, and are enough of them installing? Guessing does not answer that. Play Console gives you data about how many people see your page, where they came from, how many install, how many stay, and it lets you run experiments to test which icon, screenshots or text works better.

This lesson is Module 2: Analyze and explore opportunities of the Google Play Store Listing Certificate (https://playacademy.withgoogle.com/certificate). As in lesson 50, we teach the skills in our own words and with our own examples; the official course and its exam remain the authoritative source, and we cannot reproduce them.

Honesty note. Report names, metrics and menus in Play Console are renamed and rearranged from time to time. We describe the ideas; use the Console's own help (the ? icon on each report) for the current wording. Nothing here promises installs, rankings or revenue. Data from a small number of users is noisy, and a good analyst says so.

By the end you will be able to:

• Explain the acquisition funnel: store listing visitors → installers → retained users.
• Read the store listing acquisition and user acquisition reports and filter by country, traffic source and date.
• Calculate and interpret store listing conversion rate (CVR).
• Use benchmarks (peer-group comparisons) sensibly.
• Understand how Google Play search and Explore surface apps, and what you can influence.
• Run a sound store listing experiment (A/B test): pick a hypothesis, one change, a sensible audience, a duration, and read the result with its confidence interval.
• Decide what to do next, and what not to conclude.

Big idea: measure, change one thing, measure again. A confident-sounding opinion about "what users like" is not evidence.

• Play Console and the main store listing (lesson 50).
• Percentages and ratios (a conversion rate is installers ÷ visitors).
• Basic idea of an average and a sample size.
• Light Kotlin or Python for the small calculators in this lesson (lessons 6–14).

Imagine a shop in Dugbe market. A hundred people walk past (impressions), thirty step in to look (store listing visitors), and ten buy something (installers). A month later, four of them are still using what they bought (retained users). The shopkeeper can improve the sign (to attract people in), the display (to convince them), or the product (to keep them). Different problems need different fixes.

 Impressions (your icon/title in search, Explore, ads)      100,000
          |   click-through rate (CTR) = visitors / impressions
          v
 Store listing visitors (opened your listing page)           8,000
          |   store listing conversion rate (CVR) = installers / visitors
          v
 Installers (first-time installs)                            1,600   (CVR = 20%)
          |   retention = still active after day 1, 7, 30
          v
 Retained users                                                480   (D30 = 30%)

Notice that a better listing cannot fix a bad app. If installers leave on day 1, improving the screenshots will bring more people to a disappointing experience, and ratings will suffer.

In Play Console you find these under Statistics, Reports (names vary) and the Grow users section.

Store listing acquisition report

This report is about people who visited your listing page and what they did. Typical metrics:

Use filters and comparisons (date ranges, country, traffic source) to see patterns, for example "conversion in Nigeria from Play search" versus "conversion in the UK from referrals".

User acquisition report

This report is about new users over time, including installs that did not come through the listing page (updates, installs from other surfaces) and what they do afterwards: retained installers after 1, 7 or 30 days, cost per install if you advertise, and similar. It helps you see which channels bring users who stay.

Other reports that guide decisions

Privacy thresholds

Console hides data when the numbers are very small (to protect users' privacy), and some figures are modelled or delayed. A new app with 30 visitors will not give clear answers. Be patient.

Let us compute conversion by hand, then in code. Suppose last month's data for Class Register:

Traffic source          Visitors   Installers   CVR
Google Play search         4,200         756   18.0%
Google Play explore        1,100         143   13.0%
Third-party referrals      1,500         420   28.0%
Google Ads                   600          42    7.0%
TOTAL                      7,400       1,361   18.4%

The total CVR is 1,361 ÷ 7,400 = 18.4%. Notice how the weighted total does not equal the plain average of the four rates (which would be 16.5%). Always compute totals from totals. A tiny script that does it for you, in Kotlin:

data class SourceRow(val name: String, val visitors: Int, val installers: Int) {
 val cvr: Double get() = if (visitors == 0) 0.0 else installers * 100.0 / visitors
}
fun main() {
 val rows = listOf(
 SourceRow("Google Play search", 4200, 756),
 SourceRow("Google Play explore", 1100, 143),
 SourceRow("Third-party referrals", 1500, 420),
 SourceRow("Google Ads", 600, 42)
 )
 for (r in rows) println("%-24s %6d %6d %5.1f%%".format(r.name, r.visitors, r.installers, r.cvr))
 val visitors = rows.sumOf { it.visitors }
 val installers = rows.sumOf { it.installers }
 println("%-24s %6d %6d %5.1f%%".format("TOTAL", visitors, installers, installers * 100.0 / visitors))
}

Output:
Google Play search 4200 756 18.0%
Google Play explore 1100 143 13.0%
Third-party referrals 1500 420 28.0%
Google Ads 600 42 7.0%
TOTAL 7400 1361 18.4%

What the table tells you

• Third-party referrals convert best (28%): people sent by a friend, a school website or a WhatsApp group already trust the app. This shows the power of recommendation.
• Explore converts lower (13%) than search (18%): Explore visitors are browsing, search visitors have a need. That is normal.
• Google Ads convert worst (7%): either the ad promise does not match the listing, or the audience is wrong. Do not stop there: ask "why?", and test.
• Volume matters too: referrals convert well but have only 1,500 visitors. A small improvement in search conversion affects far more installs.

Benchmarks

The Console can show a peer set comparison: for example, your conversion rate against the median and top quartile of similar apps in your category and country. Use benchmarks as a direction, not a target. A "lower than peers" rate for a niche school tool is not a failure if the app is intended for a narrow audience. The peer set is chosen by Google and may change.

Seasonality and context

Compare the same weekday patterns and the same season. In Nigeria, installs for an education app will rise when a new school term begins and drop in holidays; an exam-results app will spike in results season. A week-on-week comparison across a public holiday proves nothing. Annotate your own charts with launch dates, app updates and campaigns.

Users reach your listing through:

Tracking your own campaigns

When you share a link to your listing, add UTM parameters so the acquisition report can tell you which campaign worked:

https://play.google.com/store/apps/details?id=com.techbase.classregister&referrer=utm_source%3Dschool_newsletter%26utm_medium%3Demail%26utm_campaign%3Dterm2_launch

The referrer value is URL-encoded (%3D is =, %26 is &). Check Google's current documentation for the Play referrer format before relying on it; the Play Console's campaign URL builder produces correct links.

Quality signals

Google Play considers technical quality (crashes, ANRs, battery, slow start-up measured as Android vitals), ratings and reviews, and engagement and retention. A listing with the best screenshots but a 6% crash rate will not do well. This is why lesson 49's pre-launch report and lesson 57's performance work matter for store success.

A store listing experiment (also called an A/B test) shows different versions of part of your listing to different groups of users at the same time, and measures which version gets more installers. It replaces opinions with evidence.

What you can test

Two kinds of experiments exist in the Console (the names may change):

1. Default graphics experiments: test icon, feature graphic, screenshots and video in your default listing (for all users who see the default).
2. Localised experiments: test text and graphics in one language, up to a number of variants.

You choose a variant count (for example, up to three variants against the current "control"), an audience percentage (how much of your traffic goes to the experiment, split evenly among versions), and the metric to optimise (first-time installers or retained installers).

Designing a good experiment

Step 1: Write a hypothesis. Not "let's try green". Write:

Because user reviews say "I can't see what it does", we believe that adding a caption
to each screenshot explaining the benefit will increase the first-time installer rate
for Play search visitors in Nigeria by at least 1 percentage point.

Step 2: Change one thing. If you change icon, screenshots and text at once and the result improves, you will not know why. If you must test two, run them one after another.

Step 3: Choose the audience and variants. More variants mean fewer visitors per variant, which means longer experiments. With modest traffic, test one variant against the control.

Step 4: Decide the duration in advance. Run long enough to cover at least one full weekly cycle (weekdays and weekends behave differently), often two or more weeks, and until the Console shows a result with enough confidence. Do not stop on day two because one variant is ahead. Early leads often vanish (this is "peeking").

Step 5: Read the results correctly. The Console reports for each variant:

• The installer count (often scaled to the whole audience) and the performance relative to the control.
• A confidence interval (range) for the difference, and often a statement such as "the variant has an X% chance of performing better".

Interpret like this:

Step 6: Learn and record. Whether it won or lost, write it down: what you tried, why, the result, and what you learned. A decision log is one of the most valuable things a small team can keep.

A tiny significance helper

You do not need statistics software to understand the idea. This Python function uses a standard two-proportion z-test to estimate whether the difference between two conversion rates could just be luck:

from math import sqrt, erf
def ab_test(visitors_a, installs_a, visitors_b, installs_b):
p_a = installs_a / visitors_a
p_b = installs_b / visitors_b
pooled = (installs_a + installs_b) / (visitors_a + visitors_b)
se = sqrt(pooled * (1 - pooled) * (1 / visitors_a + 1 / visitors_b))
z = (p_b - p_a) / se
p_value = 2 * (1 - 0.5 * (1 + erf(abs(z) / sqrt(2)))) # two-sided
return p_a, p_b, z, p_value
p_a, p_b, z, p = ab_test(5000, 900, 5000, 960)
print(f"A: {p_a:.1%}  B: {p_b:.1%}  z={z:.2f}  p={p:.3f}")

Output:
A: 18.0% B: 19.2% z=1.54 p=0.123

A p-value of 0.123 means that if there were truly no difference, we would still see a gap this large about 12% of the time. That is not strong evidence for variant B, even though it looks 1.2 points better. With more visitors the same gap could become convincing. (The Console uses its own methods and shows intervals; this helper only builds your intuition. Never claim certainty from one number.)

How many visitors do I need?

A rough rule of thumb: to detect a relative improvement of 5% on a conversion rate of about 20%, you need many thousands of visitors per variant. Small improvements need huge samples. If your traffic is small, test big, bold changes (a very different icon or screenshot style) that can produce large effects, and be patient.

Pitfalls

• Seasonality during the test. A holiday or a marketing push in the middle contaminates results.
• Novelty effect. A new icon may win simply because it is new; check again later.
• Testing the wrong audience. A test run on all countries may hide a result that is true only for Nigeria. Use localised experiments where relevant.
• Optimising installs while hurting quality. A click-bait screenshot can raise installs and lower retention and ratings. Consider optimising for retained installers where available.
• Policy. Variants must obey the same accuracy rules as any listing. You may not test misleading claims.

Put the analysis together as a repeatable monthly routine.

MONTHLY LISTING REVIEW (30 minutes)

1. Funnel: visitors, installers, CVR vs last month and vs same month last year.
2. Sources: which traffic source grew? which converts worst? why?
3. Search terms: new queries? terms with visitors but low conversion?
4. Countries/languages: any new market with traffic but no localised listing?
5. Vitals and reviews: crashes or complaints that explain a drop?
6. Experiments: any running? any finished? log the result.
7. One decision: choose ONE experiment or fix for next month.

A worked example of reasoning from data:

Observation: Google Play search visitors from Nigeria convert at 18%, visitors from Ghana at 9%. Both see the English default listing.
Possible causes: different price expectations, different payment methods, screenshots show naira only, offline use less relevant.
Next step: create a localised experiment for Ghana with screenshots in cedi and a sentence on mobile money, one change at a time.

Notice the language of caution: "possible causes". Data tells you what happened; only tests and user research tell you why.

Mistake #1: Averaging percentages. Symptom: wrong overall conversion. Fix: divide total installers by total visitors.

Mistake #2: Judging an experiment after a day or two. Fix: decide the duration up front and cover at least a full week.

Mistake #3: Changing several things at once. Fix: one change per experiment.

Mistake #4: Ignoring traffic source. Symptom: confusion about "our conversion". Fix: segment by source and country.

Mistake #5: Chasing benchmarks. Fix: use them as direction; compare to your own history first.

Mistake #6: Optimising installs at the cost of quality. Fix: check retention and ratings after a change.

Mistake #7: No decision log. Fix: record each hypothesis and result.

Mistake #8: Not tagging campaign links. Symptom: every referral appears as "third-party referrals". Fix: use campaign parameters.

Mistake #9: Reading noise as signal in small data. Fix: wait, widen the date range, or test bigger changes.

Mistake #10: Fixing the listing when the app is the problem. Fix: check Android vitals and reviews first.

Question 1. What is the store listing conversion rate?

Question 2. Why is referral traffic often higher-converting than Explore?

Question 3. Name three listing elements you can test with store listing experiments.

Question 4. What does a confidence interval that straddles zero mean?

Question 5. Why should an experiment run for at least one full week?

Go deeper: official resources

• Google Play Store Listing Certificate
• Google Play Console Help
• Android Vitals and app quality (developer docs)
• Prepare your app for release