Search Unity

Simplify your game testing with our new machine learning product, GameTune

, novembro 8, 2019

One of the most exciting announcements at this year’s Unite conference was the beta release of GameTune, our new machine learning tool for game optimization.

Releasing new features and events is an integral part of running your game. But how do you make sure that features are working optimally for your players? Many developers A/B test before releasing changes more broadly, but each A/B test can take weeks to design, roll out and analyze. Not every test yields useful results, and it can take weeks before you realize that a new feature has had a negative impact on your game.

That’s why we built GameTune. GameTune lets every studio – large or small – harness the power of Unity’s machine learning to test more, learn faster, and act more quickly.

How does it work?

First, create a question you would like GameTune to answer, then choose a metric you want to optimize for. Each question can be used to lead your game toward a different business goal. In the example above, we partnered with Futureplay, a studio out of Helsinki, on Idle Farming Empire. They wanted to test tutorial speeds, and their goal was retention.

Next, give the GameTune decision engine multiple alternatives to choose from. In this example, Futureplay chose four speeds, from very slow to very fast.

GameTune then decides which one to serve to each player of your game, based on what will be the most likely alternative to drive the highest retention for that user.

Unlike traditional A/B testing, which will only reveal what is best for the average user, GameTune finds the best experience for each user.

Any part of your game can be made dynamic. You can test assets, IAP offers, first-time user experience, and so much more.

Learn more about how Futureplay used GameTune to improve revenue and retention.

It’s easy to get started with GameTune

Apply for the beta. Once you’re signed up, you can follow these steps to integrate GameTune into your production process.

To begin, you will first need to integrate the GameTune SDK. The SDK provides an easy way to access the GameTune service and its machine learning capabilities.

Interaction with GameTune works through a Question and Answer model. First, create a name for your Question, which identifies the game’s optimization point (e.g. level_difficulty). Next, you define the alternative answers for your Question (e.g. easy, medium, hard). When a player engages with your game, your Question is asked and gets an optimized Answer back, based on the attributes of that user.

For each Question, you can set a different target to optimize for. You can configure the optimization goal in the GameTune dashboard (Projects – choose a project – Optimization → GameTune). For example, your goal for asking a Question could be to improve the game’s retention, to maximize conversion (having a user perform some specific action), or to maximize for player’s lifetime value.

The final step is to report to GameTune each time a player completes an action related to your optimization goal. For example, if the target is to maximize the number of ad impressions, then a reward should be sent every time a user watches an ad. In the case of retention optimization, the reward is sent automatically when the user comes back to the game. This way, GameTune learns to select the optimal outcome when the next user with similar attributes goes through the system.

When a game with GameTune integration is initially released, GameTune doesn’t have any data about how different alternatives of the Question will perform yet. Until enough data has been collected, GameTune randomly selects an alternative for any given user, functioning much like A/B testing. Once enough data has been collected, a machine learning model is trained automatically, and the optimization begins. As more users go through the system, the model becomes increasingly accurate as it continues to learn and iterate based on your users’ preferences.

Getting started

Sign up for the beta today.

Deixar uma resposta

Você poderá usar estes atributos e marcas HTML: <a href=""> <b> <code> <pre>

  1. Where come from the user attributes? And are they shared with all Unity game ?

    Also, where is stored the AI ? In a server, right? So if the user is offline it won’t work, right ?

    Overall it sounds super interesting, even for some Indie games !

    How many users should we have to get it efficient enough ?

    1. User attributes are only used for optimizing your game. Machine learning model is operated on the server side, if the user is offline they will get the control Answer which is defined by you. Minimum number of users we recommend is thousand per alternative for a question.

  2. Can this be used to optimise a games level order? Can you include an example in the documentation.

    1. That is a possible use case for GameTune yes. You could ask a Question “next_level” with alternatives “level_1”, “level_2”, … Same question could be asked after each level.

  3. Just a thought. Can we use this to teach a simple card game AI to learn user behavior and adapt the gameplay!

    1. Hi Sathyaraj, this product is not suitable for teaching game AI. See for example ML agents https://unity3d.com/machine-learning.