Machine learning (ML) in the gaming industry is most often associated with level testing and dynamic complexity change. Meanwhile, this tool can also be used to increase gaming profits. About this in his material for App2Top.ru Dmitry Shelengovsky, the founder of Playgendary, who is actively introducing ML into his products, told us.
Dmitry Shelengovsky
I identify seven tasks that machine learning allows you to solve when operating games.
1. Expansion of the game’s paying audience
This can be achieved with the help of look-alike targeting based on machine learning technology, which is already used in advertising networks. Its essence is that algorithms study the audience, find common features of people and patterns of their behavior, and then user segments are formed based on this data.
How can this be used to increase revenue in games?
Example: the algorithm determines the common features of people who spend the most time in the game and spend money on it. Then this information can be uploaded to the advertising service (“Yandex.Yandex.Direct, Facebook, Google AdWords and others) as targeting to find people with the same characteristics. Such players are more likely to play for a long time and spend money. Thus, it is possible to optimize advertising costs, increasing its effectiveness. I will make a reservation that in the example above we are talking about ML hidden inside advertising networks, and not directly the game.
2. Retention of players and prediction of their departure
When analyzing a player’s behavior, machine learning algorithms can see the first signs that he is going to leave the game.
For example, the system records that a player is killed too often, he enters the game less and less and spends less and less time in it. This may be a sign that he is disappointed in the project and will leave for good soon. To revive interest, the system can offer the player additional levels, reduce complexity, provide gifts or new functionality. This is important: it is cheaper to retain an existing player than to attract a new one.
There are even special platforms for solving this problem. Japanese Silicon Studio, for example, has created its own YOKOZUNA, which is able to predict the behavior of players, predict how much time they will play, how much money they will spend (and for what), and even what level they will reach. Third-party developers can also use this platform.
3. Dynamic price changes for in-game products
A big misconception is the opinion that a single price for an in-game product is suitable for everyone. Today, many games around the world have millions of users. Obviously, they can’t all have the same income level.
The machine learning algorithm, analyzing a person’s gaming habits, the history of his previous purchases, taking into account the gaming platform and other factors, can determine and offer the optimal price for the player of the goods. Thus, the probability of a sale will increase.
Thanks to this approach, revenue from content sales can grow by 20-40%.
4. Individual offers
As in the case of dynamic pricing, machine learning algorithms can generate individual content offers for players. Having studied the users’ play style, purchase history and other factors, the system will offer the player the product that he wants to purchase more likely right now.
For example, if a player methodically “pumps” the skills of wielding a two-handed sword, then you should not offer him to buy a bow.
5. Turning free players into paid ones
The free-play model in games implies both the presence of paying players and those who play only for free. Machine learning allows you to increase the number of the former at the expense of the latter.
How does this happen?
If the algorithm managed to collect enough information about how the user plays and about his preferences, then the system will better understand his motivation and values. Therefore, it will be able to generate exclusive offers that will have value for this particular player.
By making a unique offer at a comfortable price and at the right time, the system will gently push the user to the first purchase. The psychological barrier of the “first payment” will be overcome, and the probability of the second and third payments increases significantly (compared to the first payment).
6. Optimization of the frequency of advertising
Machine learning algorithms can also be used to optimize the frequency of advertising, if the monetization model implies its presence. For example, to achieve a balance between profit growth and maintaining player loyalty.
Imagine that the algorithm analyzed the behavior of users and noticed that after reaching the fifth level, the probability of their leaving in the next month decreases by 50%. Consequently, the level of loyalty of such players is higher and it is unlikely that a small increase in the volume of advertising will provoke violent discontent. Players who are less tolerant of advertising can be shown less of it.
7. Fighting fraudsters
One of the main problems of promoting any mobile applications is fake installations and “garbage” traffic. As a rule, applications are promoted according to the CPI model, in which the developer pays for each installation. Scammers even build entire “farms” of devices to wind up installations.
All this costs the mobile games industry billions of dollars annually. One study by the World Federation of Advertisers says that by 2025, scammers will cost app developers $50 billion a year. According to the AppsFlyer platform, in the first quarter of this year, their losses have already amounted to $700-800 million (plus 30% compared to last year).
Machine learning algorithms can analyze fraudulent activity, learning to calculate it and block payment for fake installations. Thus, they will save developers millions or even billions of dollars on advertising.
Instead of output
One of the most extensive tests of the impact of machine learning on monetization (of those that were announced publicly) was conducted by the developers of the Game of Whales service, which specializes in ML monetization. At first, their algorithms studied the behavior of 18 million players for 18 months. Then a control group of users was selected, with whom the AI worked. It turned out that in this control group, revenue from free-play games increased by 25%, and the outflow of users decreased by 10%.
What do these figures mean? Monetization of games using machine learning is not just a tool to increase developers’ income, but also a way to make the game more comfortable and enjoyable for the players themselves, giving them exactly what they want without pushing them away with aggressive advertising.
Also on the topic: