At Coda we strive to provide simple yet powerful tools to our game developers. The Coda SDK is a prime example. It includes all you need to bring your game to market including measurement and attribution tools, monetisation solutions and a game analytics platform to name a few.
We are excited to now be able to launch one new capability of our SDK: machine learning. In the third of our series on this field of AI we look at the role it will play directly within our games.
As ever Coda’s machine learning expert Muhammed Miah is on hand to explain the technical side of the process.
Personalisation
Machine learning will largely be used within the SDK to personalise the gaming experience for each player. As you will discover it enables us to work out which gamers are likely to play regularly, and provides insight into their gaming habits. We can determine when they will next play the game as well as what their response is likely to be towards advertising. The data the algorithm accrues can even highlight the gamers that are likely to make in-app purchases.
“For me, the key role of machine learning in the SDK is to deliver the best experience possible to each of our gamers,” explains Muhammed. “This starts with understanding more about them, predicting their preferences, and easing any possible pain points. We currently aim to optimise the gaming experience in two primary ways. Firstly, we keep people playing the game for as long as possible, and secondly, we monetise them in the best and most effective way possible.”
“For example, say the machine learning model thinks that a person is finding the game too easy. It can then adjust the difficulty of the game itself, making it more difficult and engaging for the player. This also naturally results in that player becoming more sticky.”

“One other type of user that we try to predict early on are what we term ‘hardcore users’. This group of people enjoy our games the most and we want to know who they are as early as possible. You can imagine that some of them will wish to thank the developers and opt for the ad-free version of the game, and so we may try to promote that appropriately to them.”
Machine learning can also analyse data to tell us more about a person’s gaming habits. For example, how often a person plays a game and what time they enjoy playing. This is very useful data as it enables Coda to optimise our communication with the gamer.
“If a person, for example, only plays a game at night, it’s better to not send them a notification reminding them to play in the morning. So with data that we have collated we can communicate with them in the way that is most relevant to them,” suggests Muhammed.
“I find it exciting that we are able to utilise this very powerful realm of AI, right from within the client’s device itself”, he mentions. “Our algorithms are even able to run offline, without an internet connection, meaning that communication with our backend servers is not necessary at all. These algorithms first observe the player’s behaviour and the details of how they interact with the game. They then surface key pieces of information, what we call features, to be used. Finally, at the appropriate time, the machine learning model is run with these features to produce the desired prediction. In short, the SDK collects data and computes features before running the machine learning model.”
Data to help monetisation
“We also use machine learning to gauge a gamer’s response to ads,” adds Muhammed.
“From the data we get we understand how each individual user feels about ads. Do they prefer to receive rewards for watching one? Do they tend to click-through on them? Or is the next ad likely to make them quit the game altogether? Using this data we tweak the game to ensure that they keep playing while seeing the number of ads that they are happy with.”
“One of the most powerful signals that we have is if a person stops playing. In relation to ads, we aim to show them in a sustainable way that both keeps each user playing while allowing our developers to continue making more games.”

Talk of monetisation is incomplete without mentioning in-app purchases. Here machine learning will help Coda pinpoint the best possible moment to promote IAPs and which gamers are most likely to respond to them.
“We're getting significant amounts from in-app purchases,” adds Muhammed. “We aim to use machine learning to discover whether a gamer is the type of person who wants to make an in-app purchase and if they are we want to reduce friction to make it easy for them to do so.”
“In the future we will be even more effective at optimising our games,” says Muhammed. “The more games we publish the more data we have and that means the more themes and trends that the machine learning algorithm will be able to spot. Eventually much of what we do will be automated by machine learning.”