ThinkingAI
Three colleagues at a desk monitor talk through a test result in a bright office

Test the change.Ship with evidence.

Plan product experiments by chatting with our agents. They help you pick the cohort, run the test and read the result, then plan the next round.

Habby
“ThinkingAI gave us the visibility we needed to act fast, test smarter, and grow in new markets.”
HabbyMobile game studio behind Archero and Capybara Go
Read Habby's story
Capybara Go game art, a capybara with an orange on its head relaxing in a hot spring
$109M+
in-game spending on Capybara Go in its first 90 days
2B+
lifetime downloads across Habby’s games

One place to plan, runand read your experiments.

Design, run, report and repeat your product experiments in one workspace, on the same data as your analytics.

  • Design

    Pick the users, the version each group gets and the metrics that decide the test.

  • Execute

    Config Center gives each group its version with no app release. Traffic layers run tests side by side.

  • Report

    Each report tracks key, secondary and guardrail metrics, with P values and confidence intervals.

  • Repeat

    Open the result in Analytics and let the agent draft the next test from what you learned.

Test the experience inside your app.

Our SDKs give each group its own version of a feature, layout or price, inside your product.

SDKs for

  • iOS
  • Android
  • Unity
  • JavaScript
See SDK setup in the docs

Test any change.Trust what the result says.

Onboarding, prices, messages and features, each tested on the same metrics as your analytics.

  1. Feature and UI tests

    Test tutorial length.

    • Vary the step count by group
    • Use Config Center to skip an app release
    • Pin QA devices to one version

    Ask

    Which tutorial has the best completion rate?

  2. Price tests

    Price the starter pack.

    • Try a different price in each group
    • Compare payer rate and revenue per user
    • Monitor the refund rate

    Ask

    Which price earns the most per user?

  3. Push message tests

    Test winback copy.

    • Compare up to four messages with a control
    • Keep a no-message control group
    • Let triggered tasks auto-send the winner

    Ask

    Which push brought the most users back?

  4. Parallel tests

    Keep test groups apart.

    • Limit each user to one test per layer
    • Put competing tests in the same layer
    • Reuse traffic when a test ends

    Ask

    Which layer has room for a shop test?

  5. Reading results

    Is the lift real?

    • Track key, secondary and guardrail metrics
    • Compare P values and confidence intervals
    • Results freeze when the test ends

    Ask

    Did the shorter tutorial lift completion?

    Experiments are an add-on suite. Talk to sales to add them to your plan.

Test results shape the next experiment.

Every test feeds the next one. The agent helps at each step, and you approve what goes live.

The next test starts from what you learned

The result opened in Analytics by channel: organic 4.4% to 5.1%, plus 16%; search 4.2% to 4.7%, plus 12%; video ads 3.6% to 3.7%, plus 3%. The agent suggests a different starter offer for users from video ads.

Ship what works.Know what each change earns.

See how two studios used user data to decide what to change in their games.

  • Polaris: Crown Legacy, a knight defending castle walls

    UNDERDOX

    Same day
    answers to data questions, down from one to two weeks
    3 days
    from problem found to final design

    When players left after losing troops, UNDERDOX designed a way to bring those troops back.

    Read the UNDERDOX story
  • Whiteout Survival

    Century Games

    $2B+
    Whiteout Survival revenue in two years and one month
    #1
    grossing mobile game worldwide, March 2025

    Century Games used player behavior to adjust when offers appeared and how quickly Whiteout Survival got harder.

    Read the Century Games story

Explore Product Analytics