Agent skill for growing revenue

The growth-advisor skill answers “what should I test next?” from your own subscription numbers. Your AI coding tool reads your analytics, prices, placements, flows, and past A/B tests, scores each part of your business against the market, and returns a short list of hypotheses ranked by the dollars each one puts at stake. When you pick one, the skill builds what the A/B test needs and sets up the test.

Note

growth-advisor changes nothing in your account without your yes. Every read is read-only, and each write — a flow copy, a segment, a product, a price, an A/B test, a placement change — waits for you to confirm it.

What you can ask

Ask the way you’d brief a colleague. You don’t need to name the skill:

Audit my app. What should I test next to grow revenue?

Revenue dropped last month. Why?

Is our annual plan too cheap in Germany?

My A/B test on the onboarding paywall ended. What now?

Where do users leave our onboarding?

I have three ideas: a weekly plan, a trial on annual, and a second placement.
Which one should I test first?

Which country refunds the most, and do people who start a trial renew better?

Here's our App Store link. Where are we losing money?

A question about your numbers gets an answer from your numbers, with no audit attached. An app with no data yet gets a starting plan built from market medians in Adapty’s State of Paywalls 2026 report, with every number labeled as an estimate.

growth-advisor
Revenue dropped last month. Why?
Ran 4 read-only commands
  • adapty analytics chart --chart revenue --compare-from 2026-08-01 --compare-to 2026-08-31
  • adapty analytics chart --chart revenue --group-by store
  • adapty analytics chart --chart subscriptions_new --group-by store
  • adapty analytics chart --chart subscriptions_expired --group-by cancellation_reason

Your revenue dropped because fewer people renewed on Google Play, not because fewer people bought.

New subscriptions held steady in September. Renewals on Google Play fell by a third, while the App Store didn’t change. Most of the lost renewals are monthly subscribers whose payment failed.

Next steps

  1. You: did anything change in your Google Play billing setup in early September?
  2. Me: suggest hypotheses to win back the monthly subscribers who left. Want me to?
Example conversation with a sample app. Your numbers will differ.

Before you start

  • An Adapty account with data: The skill reads the same analytics your dashboard shows. Without data, it gives you a starting plan. Without an account, it works from what you send instead (see Without an Adapty account).
  • An AI tool that reaches your Adapty account: The skill works through the Adapty CLI or the Adapty MCP server. The CLI is the better fit, because the audit makes its reads through it. Use Claude Code, Codex, Copilot CLI, Gemini CLI, or the Claude app after you set up the CLI there. With MCP only, the skill offers to set up the CLI and, if you decline, names the reads that couldn’t run.

Install the skill

growth-advisor is part of the adapty-skills plugin.

To install the plugin, in your terminal, run:

npx adapty skills install

Restart your agent to load the skills. For other installation methods, see the plugin README.

Claude Code doesn’t update the plugin on its own. To get new skills as they ship, in Claude Code, run /plugin, open Marketplaces, select adapty, and choose Enable auto-update. For the update command in other agents, see the plugin README.

How the audit works

When you ask what to do without a specific question, the skill runs an audit. It looks at the whole business over the last six months, not only at what changed last: a part that has been weak all along costs more than a recent dip.

  1. How your app earns: The skill finds the store that brings the money, your main market, your main product by revenue, and what each placement shows now.
  2. What you already tried: The skill lists your past paywall and flow A/B tests with their dates, audiences, and results, and spots paywalls that ran side by side with no test. You don’t need to type your test history; the reply states it as facts, so you only correct what’s wrong.
  3. A scorecard: The skill grades each part of the business as strong, typical, or weak: where the money comes from, the funnel from install to paid, the plan mix, renewals, refunds, and your placement setup. Conversions are graded against the market grid for your plan, region, and category from Adapty’s State of Paywalls 2026, and renewals and refunds against Adapty’s State of In-App Subscriptions 2026. The weak parts are ranked by dollars a month at stake.
  4. The cause: The skill follows the biggest signal down by store, attribution channel, plan length, country, or week until one group explains it. Before it blames traffic or a specific change, it verifies the other possible causes.
  5. The screens: Before it says anything about a screen, the skill renders the live flow and, for a flow, reads its funnel to find the screen where users leave. A paywall your app draws itself needs a screenshot. The audit lists the screenshots it needs under Send me screenshots and says which hypothesis each one unlocks.

What you get back

The skill replies with a short answer and the scorecard, then a list of hypotheses grouped by placement, in the order to test them. Each hypothesis says why your app needs it, what result to expect and where that estimate comes from, what the change is worth in dollars a month, how long the A/B test would take at your purchase volume, and who does what. A hypothesis can target everyone at a placement or one segment, such as former subscribers.

growth-advisor
Audit my app. What should I test next to grow revenue?
Ran the audit reads
  • adapty apps list
  • adapty analytics chart --chart revenue --group-by country
  • adapty analytics chart --chart revenue --group-by duration
  • adapty analytics conversion --conversion trial-paid --granularity month
  • adapty ab-tests list --type flow
  • adapty placements get <placement-id>
  • adapty product-prices list --product <product-id> --csv
  • adapty flows config get <flow-id>

Your app converts like a typical US app; a cheap annual plan and lost renewals cost you most.

PartYoursAgainst
Paywall view → paid3.1%2.75% US mediantypical
Trial → paid29%26.2% US medianstrong
Annual price$29.9910th percentile, US photo appsweak
First renewal, monthly41%58% six months agoweak

The annual plan is cheaper than 90% of US photo apps, while your conversion sits above the median.

From your data: you tested two onboarding flows in July and kept “Main”.

Suggested hypotheses

On the onboarding placement:

1. Replace ‘Annual’ at $29.99 with a copy at $33.99 on onboarding, for App Store users in the US

  • Why: Your annual plan is in the cheapest tenth of US photo apps while your paywall converts above the median, so the A/B test shows whether $33.99 earns more per user.
  • Expected: ARPU up 15–25%, judged by revenue per user, refunds included. Price raises in Tier-1 markets tend to win in Adapty’s library of real subscription A/B tests.
  • Time: About 4 weeks, at about 110 purchases a week on this placement. Renewals won’t show yet, so the test is judged on first-year revenue and refunds.
  • Who: Me: create the product copy, push it to the App Store, build the flow copy, and set up the A/B test. You: approve the new product ID and publish the flow copy.

Next steps

  1. You: which one do you want to start with? I’d pick 1, because every new user sees it.
  2. Me: create “Annual 2” at $33.99. Want me to?
Go with 1.
Create a product copy

Create Annual 2, a copy of “Annual” at $33.99 in the US, with the store product ID com.app.annual2? The store product ID is permanent. The live product and its subscribers stay as they are.

  1. Yes
  2. No, change something first
Example conversation with a sample app. Your numbers and hypotheses will differ.

You pick which hypothesis to start with. If you’re not sure, the skill names its own pick and why. The skill keeps no backlog, so save the reply if you want to come back to the other hypotheses later.

Prices

The skill reads what each product costs in every country and store, as the store charges it, in local currency. It then places each price against Adapty’s price index from the State of In-App Subscriptions report: the 10th to 90th percentile for your category and plan, in 22 countries. For other countries, the skill compares with the region or with the country’s own past.

From that, the skill can suggest these price hypotheses in the same list as the rest:

  • Raise or lower the core price of your main plan in your main market.
  • Raise or lower the price in one country whose price is out of line with your main market. For a gap in one market, the skill first verifies that the paywall is in that market’s language.
  • Change the plan ladder when annual costs too much or too little against weekly or monthly.

The skill suggests a price raise only in Tier-1 markets, such as the US, the UK, Germany, or Japan: in Adapty’s library of A/B tests, the same increase lost elsewhere. The new price starts from what the store charges now and lands on a price point buyers expect, such as €38.99. The skill doesn’t read competitors’ prices.

The A/B test runs on a copy of the product, not on the live one. The skill creates the copy, pushes it to the store, sets its prices, builds the flow copy that sells it, and sets up a country segment and the A/B test. A winning price then moves to the original product. Existing subscribers keep their price unless you ask to move them.

To change a live price without a test, ask for that. The skill saves the current prices to a file so you can restore them, previews the new prices, and tells you when each store applies them.

Warning

Prices can only be read for a store that is connected to Adapty. On the App Store, that needs the App Store Connect API key, not only the In-App Purchase key.

From hypothesis to A/B test

For every hypothesis, the skill suggests an A/B test and offers to set it up. Before it does, it verifies that everything the test needs exists, and names what is missing and who makes it:

The test needsWho makes it
A changed flow: a new design, copy, or product setThe skill builds a copy of the flow through the flow-generator skill, with one change per hypothesis, so the test shows what that change did. Before you publish the copy, the skill offers to review the copy with the flow-audit skill. A flow joins a test only once it is published.
A new plan lengthThe skill creates the product and pushes it to App Store Connect and Google Play Console, after you confirm its store product IDs and USD price.
A price test’s copy of a productThe skill creates the copy, pushes it, and sets its prices. You approve the new store product ID.
A trial or offerYou create it in App Store Connect or Google Play Console. On the App Store, a trial needs a new product; on Google Play, it’s an offer on the existing plan. Once you give the skill the store offer ID, it adds the offer to the product in Adapty.
A paywall from products already in AdaptyThe skill creates it.
A paywall with a new designYou make it in the Paywall Builder.
A segmentThe skill creates it through the audiences skill.
A new placementThe skill creates it after you approve its developer ID. Your app needs a code change and a release to fetch it.

Each piece the skill makes waits for your yes. When a piece is yours to make, the skill’s Next steps list says so, and the skill sets up the test once you tell it the piece is ready. It never sets up a test with a piece that doesn’t exist yet.

When every piece exists, the ab-tests skill creates the test as a draft. Before it starts the test, it shows you exactly who sees what. An A/B test is the default, and the only way to measure what the change did.

To show the change without a test, ask for that. The skill can show it on a new placement, or to some or all users of an existing placement. Before the skill changes an existing placement, it saves the placement to a file so you can restore it. It also names any A/B test that runs on the placement. The change doesn’t affect that test.

When an A/B test ends

Tell the skill that a test ended. It finds the test, shows each variant’s revenue and probability to be best, and asks you which variant won. The leading variant isn’t the winner until you say so: with fewer than about 200 purchases per variant, the test hasn’t settled yet. If the test is still live, the skill offers to stop it with your winner, then suggests the next hypothesis for that placement.

While a test runs, the skill says how far each variant is from about 200 purchases, never which one leads.

Without an Adapty account

Send an App Store or Google Play link, the app’s code, or screenshots of your onboarding and paywall, and ask how to make more money. The skill answers in full from what you sent, the market medians, and Adapty’s library of A/B tests:

  • A store link: The skill reads the category, the subscriptions in the in-app purchase list, the languages, and what reviews say about paying.
  • Screenshots: The skill reads the plan lengths, the preselected plan, the trial, and the purchase button.
  • The app’s code: The skill reads the product IDs, the paywall and onboarding screens, their copy in every locale, and where else the paywall opens.

The skill reviews the screens with the paywall-teardown and onboarding-teardown skills. Prices and conversion need your numbers, so the last Next step offers to set up Adapty in the app. Once it’s set up, the next answer runs on your own data.

Limits

  • No ad spend: The skill can split installs, trials, and revenue by attribution channel, but not spend or ROAS. For those, use the adapty-attribution skill.
  • No results for some test types: The skill reads the results of regular paywall and flow A/B tests. For tests across several placements and onboarding tests, it names the test, and you read its results in the dashboard.
  • No category of its own: Your app’s category isn’t in your Adapty data. The skill infers it from the app’s name and paywall and says which one it used, so you can correct it.

To query your analytics yourself, without the skill, see the CLI analytics reference. To pull analytics from a server or a scheduled job, where no one can log in through a browser, use the Export Analytics API with a secret API key.