Comment & Opinion

4 ways games companies can leverage machine learning

How games makers could be more effectively leveraging the petabytes of big data they’re accruing

4 ways games companies can leverage machine learning

David Drai is CEO and co-founder at Anodot.

Gaming is one of the most data-intensive domains on the planet. In the river of big data that games vendors produce, there are undiscovered nuggets of gold.

Yet unlike online commerce, gaming is still not making the most of this big data treasure piling up in its data centres.

Games developers should be using this data to more effectively enhance the overall player experience, eliminate behaviour toxicity, maintain better application health, manage game balance and lower churn. They should be, but they’re not. Why is this and how can it change?

Here are four ways gaming could be doing more with big data.


Click here to view the list »
  • 1 Improve player sportsmanship

    Multiplayer online gaming is often contingent on cooperation between players. To cooperate, players need to communicate, and where there is online communication, there is inevitably some abuse.

    It should be noted that the vast majority of players are sportsmanlike and courteous in their interactions.

    According to Riot Games, only one per cent of the League of Legends community is considered “consistently unsportsmanlike”. Yet, just like in the physical world, one bad apple can ruin an entire bushel.

    Games vendors understand that toxicity, abuse and bullying markedly degrade the gaming experience and negatively impact revenues. Players who are made uncomfortable by an abusive game environment will simply find a competing environment.

    The problem is to accurately identify abusive behaviour within the terabytes of chat logs accumulated from hundreds of millions of players, in a dynamic gaming chat environment notorious for rapidly-evolving slang and seemingly-endless abbreviations that morph into acronyms, and in a reasonable timeframe.

    To effectively interpret cross-player interaction, companies like Riot Games have begun leveraging machine learning and other advanced technology to parse massive volumes of chats, understand the unique semantics of gamer slang and acronyms, and craft automated yet contextually appropriate responses to abusers.

    But the field is still in relative infancy, and abuse, in gaming and other arenas like social media, continues to plague the online world.


  • 2 Maintaining application health

    The cornerstone of modern gaming is engagement - keeping players in an immersive experience for as long as possible.

    App crashes, glitches and performance problems - which all fall under the umbrella of “application health” - endanger engagement, brand equity and revenue.

    As a market focused on player engagement, gaming is also highly dynamic from a development perspective, and truly complex from a platform perspective.

    This explosive combination makes gaming massively susceptible to the crashes and micro-glitches that can kill performance and revenue.

    When you look at games giants like Outfit7 - which has over eight billion downloads and 350 million active monthly users generating 2.5TB of usage data a day – the sheer scale of potential problems is absurd.

    Today, games vendors can use application performance tools to sort through day-old data, identify glitches, and decide how critical a given error or crash should be considered.

    Alternatively, they can rely on user complaints to raise issues. Either way, games engineers will only become aware of performance degradation days after the fact.

    Factoring in the time it takes to pinpoint the root cause of a given problem, and the development effort required to rectify it, glitches that cause serious revenue or reputational damage can persist within game code for weeks.

    The key to identifying and resolving these glitches is hiding in the reams of big data that is gathered every minute by games vendors.

    Indeed, that data is meaningless until the industry addresses the nature and power of the tools and methodologies they’re using to extract answers from this data.


  • 3 Ensuring a balanced gaming experience

    In complex, interconnected multiplayer online gaming ecosystems, maintaining game balance is a constant challenge.

    Game balance broadly refers to how the game is tuned, whether rules are applied consistently across all game scenarios, whether players subjectively feel that the game is fair and potentially winnable, or whether strategies that the game purports to support are viable.

    Accordingly, it is a key concept in game design and development and is crucial to the user experience.

    The problem is the nearly infinite scope of permutations and highly-subjective perceptions involved in game balancing.

    In dynamic and fast-moving games development environments, game designers are challenged to effectively predict how changes to one game parameter may cascade and affect other parameters.

    The use of quantitative analysis and anomaly detection for the purposes of game balancing is one way game vendors could better put their big data to work.

    Since systemic changes to game environments can have cascading effects, real-time monitoring of data relating to predefined KPIs would enable developers to gain better insights into how each change affects the way players play and perceive the game.


  • 4 Minimising churn

    Gaming suffers from the high cost of user acquisition and, like any online service, it's harder to get people to come back than to keep them in. Thus, lowering churn is something that keeps game execs up at night, and remains a hot topic in the gaming world.

    Today, most games vendors can accurately identify when a player stops playing and activate an outreach campaign to attempt to lure them back.

    However, this is essentially the digital equivalent of closing the barn door after the cows have left. Because the what of churn is the easy part: the player left. What needs to be addressed is the why of churn.

    The trick - as companies like Yokozuna Data have figured out - is to be able to leverage big data to effectively predict churn in order to proactively prevent it.

    Data scientists are already working on ways to leverage historical data to create a model for prediction of when a player will stop playing.

    A simple example of how this could work? If game-monitoring tools noticed that a new player was racking up a lot of deaths, and was thus potentially getting frustrated, a well-tuned predictive analytics algorithm could proactively offer tips for staying alive or maybe even a free premium upgrade of some relevant parameter that would help ensure continued engagement.

    The bottom line

    Gaming is uniquely user experience-centric, with a fiercely loyal player base. But it’s also a space in which many players are surprisingly fickle, churn frequently and are not afraid to express their negative opinions.

    It’s a revenue monster, with over $36 billion in revenue last year alone, in which it still costs games makers more than $4 to acquire a user, over $50 to convert a player into a first-time purchaser - and in which only eight per cent will ever spend any money at all.

    An effective application of machine learning to big data can help quickly identify the changes in playing trends and user behaviour that affect the overall experience – and ultimately translate into revenues.

    The insights are right there, in the data. The question is, how well are games vendors responding to them?


PocketGamer.biz regularly posts content from a variety of guest writers across the games industry. These encompass a wide range of topics and people from different backgrounds and diversities, sharing their opinion on the hottest trending topics, undiscovered gems and what the future of the business holds.