Review miningA return analysis: the impact of rankcusp's ASO platform and the value of the cusp band
What is in this resource:
How to build your ASO strategy from end to end
Learn to pick target keywords, write metadata, follow rank and base every decision on data.
From data to action
Simplify the keyword data and stay in line with App Store rules.
Tools that lift rank
Carry the keywords drawn from competitor reviews into your metadata and take your app out of the cusp band into the top 10.
What the analysis answers
Which kind of move works for your app? That answer is in no general guide - it is in your own records. This document is about the method for getting it out of them.
The most frequently asked question in ASO is "how much is it earning us?" The honest answer to that gets stuck in the unmeasurable part of the chain. The second most frequently asked question is entirely answerable: which move works?
That second question is more useful, because it directly decides the next move. "Adding a phrase to the subtitle was tried five times here and won three" tells you what to do next week. "ASO earned us this percentage" does not.
The analysis takes a few months of move records as input. Its output is a win rate per kind of move. The work in between is subtracting the control group every time.
This produces knowledge specific to your app, and that knowledge exists in no general guide. Depending on the category, the product's maturity, your rating count and the shape of the competition, the answer comes out differently for every app.
Splitting moves into types
The analysis starts by grouping the moves. Five types is enough; a finer split becomes statistically meaningless because the number of examples per group falls.
- Adding a phrase to the app nameThe strongest and most expensive move. Changing a field that carries brand value also affects non-search traffic, so it is done rarely and the sample stays small.
- Rewriting the subtitleThe most common move. Visible, indexed, and easy to change. Used to carry phrases from the cusp band.
- Cleaning the keyword fieldRemoving repeats, spaces and stem duplicates and putting a new phrase in the space that opens. The easiest type to measure, because the characters opened can be counted.
- Filling a new country fieldWriting the keyword field for a storefront that was left empty. This is the only type that builds relevance from nothing, and usually the highest-return one.
- Changing categoryThe largest move and it affects every rank. It has to be measured alone; no other change should ship in the same release.
Computing a move's net effect
A move's effect is not the raw difference in rank. Inside that raw difference is the market's movement too, and it has to come out.
Step one: the week before the move, record the rank of every phrase on your tracking list - not just the one you are about to touch.
Step two: make the move and wait for the index to settle. Usually a few days; a week in some countries.
Step three: a week later, measure them all again. Compute the average movement of the phrases you did not touch - that is the market's movement for the week.
Step four: subtract the market's movement from the movement of the phrase you touched. What is left is the move's net effect.
Step five: account for the side effect. If you removed a word from the subtitle, look at how the phrases relying on that word moved too. The net effect is the gain minus the loss.
Those five steps repeat for every move. It looks laborious, and the alternative is never knowing what any move did.
Working out the win rate
After a few months you hold a list of net effects per move. Now the summary per type comes out.
Three numbers for each type: how many times it was tried, in how many the net effect was positive, and the average net effect in places.
If a type was tried fewer than three times you cannot say anything about it. That means the answer is "unknown", and writing unknown is better than writing zero.
When computing the average net effect, keep aside the phrases that left or entered the first 200; their rank is unknown and an invented number distorts the average.
The three numbers are read together. A type with a high win rate but a small average effect is reliable and slow. One with a low rate but a large effect is risky and valuable. Those need different strategies.
The type at the bottom of the table is the one to drop. That is the analysis's highest-return output and usually the hardest part to accept.
| Pattern | What to do |
|---|---|
| High rate, small effect | Reliable and slow; the backbone of the weekly loop. |
| Low rate, large effect | Risky and valuable; try monthly, choose the phrase well. |
| Low rate and low effect | Drop it. The analysis's highest-return decision. |
| Tried fewer than three times | Unknown. Do not write zero; keep trying. |
Five things that break the analysis
This analysis produces meaningless results when the measurement discipline slips. There are five spoilers.
The first is more than one change at once. If you changed both the app name and the subtitle in one release, that move cannot be attributed to any type and has to leave the record.
The second is a missing starting point. Without a measurement from before the move the net effect cannot be computed, and that row leaves too.
The third is a missing control group. If the phrases you did not touch were not measured, the market's movement is unknown and the raw difference gets taken for the net effect.
The fourth is gaps in measurement. If a week was missed, the interval between two measurements grows and the difference looks sharper than it was.
The fifth is external events. If the score fell after a release, or a rival was pulled, every measurement that week is affected. Those events have to be marked on the record.
All five share one consequence: as more rows drop out of the record, the analysis loses power. If ten rows fall out of a twenty-move record, the remaining ten cannot produce a meaningful rate per type.
- One field per release; otherwise the move belongs to no type.
- Always measure before a move; a move with no starting point does not count.
- Measure the phrases you did not touch; the market's movement comes from there.
- Mark the measurement gaps; a gap sharpens the difference.
- Releases, rating changes and rival movements all have to be on the record.
Return by country
The same analysis runs per country too, and the result is usually surprising.
The storefront where you get the most downloads may not be the one that returns the most. Competition in large markets is dense and the same move wins fewer places.
The first move in a country whose keyword field was left empty is usually the highest-return move there is - because it builds relevance from nothing. That is the finding a per-country analysis surfaces most often.
The time the index takes to settle also varies by country. An effect visible in three days in one storefront takes a week in another. Without adjusting the measurement window per country, a move gets written off early.
Visibility has to be read against market size as well. Being first in a small market can reach fewer people than being fiftieth in a large one.
Once a per-country return table exists, where the time should go becomes clear. Without that table, time usually goes to the most familiar country - and familiar is not the same as productive.
What to do with the result
The analysis produces a table. That table has one use: changing next month's plan.
The first change is making the highest-rate type the backbone of the weekly loop. One move of that type every week produces a reliable accumulation.
The second is trying the low-rate, high-effect type once a month and choosing the phrase carefully. That type is saved for the most popular phrases at the top of the band.
The third is dropping the type at the bottom. That transfers the time saved to the other two and is usually the decision that raises total return the most.
The fourth is redistributing the countries. More time to the storefront that returned more, less to the one that returned less.
And there is one thing not to do: turning the table into a performance indicator. A win rate is a learning instrument; made into a target, it starts favouring moves that are easy to measure and ineffective.
The analysis's limits
The answer this analysis gives is real but narrow. Knowing its limits stops the answer being stretched too far.
The first limit is sample size. Twenty or thirty moves means four to six examples per type. That shows a tendency; it does not give statistical certainty.
The second is time. A type that worked a year ago may not work today - the category may have matured, your rating count may have grown, or something may have changed on Apple's side. The analysis should be repeated quarterly.
The third is non-transferability. The type that works for your app may not work for another. This result is not shareable knowledge; every team gets it out of its own record.
The fourth is the chain. The analysis ends at visibility. It does not say how many installs a move produced, because conversion cannot be measured from outside.
The fifth is the product. When the strength threshold becomes unreachable no type works, and the analysis shows that as "every rate fell". At that point the conclusion to read is not about move selection: the work has moved to the product.
Frequently asked
- How many moves before I can run this analysis?
- At least three per type, which for five types means about twenty moves. At one move a week, roughly five months.
- How is the net effect computed?
- By subtracting the average movement of the phrases you did not touch from the movement of the one you did, then deducting any loss the same move caused on other phrases.
- Can I apply my results to another app?
- No. The answer depends on the category, the product's maturity, the rating count and the shape of the competition. Every team gets it out of its own record - which is exactly where this analysis's value lies.
- What does it mean if every type has a low rate?
- Usually that you have hit the strength threshold: there is no distance left to take with metadata for the phrases you are targeting. The conclusion to read is not about move selection; the work has moved to the product.
- How often should I repeat it?
- Quarterly. A type that worked a year ago may not work now; the category may have matured or your rating count may have grown.
Explore our resources
Keyword competition in the games category: from obligation to competitive advantage
Struggling to get past the App Store's metadata rules and the limits of the 100-character keyword field? rankcusp answers with a platform designed to solve what we call the "100-character lock.
See resource
Retail apps and App Store search: getting ahead
Grasp clearly the difficulties retail apps face in App Store search and identify the most effective keyword opportunities. With category data, country-level comparisons, applicable suggestions and cusp band examples, this guide helps you measure your rank, define your keyword strategy and put into practice the concrete steps that will raise your visibility.
See resource
rankcusp Pro agency partnerships
Open new doors with rankcusp Pro: become an app partner, strengthen your team's ASO expertise and run your clients' App Store visibility yourself. Download now and grow your impact!
See resource