Session: Machine Learning is a new age of Personalized Game Monetization - Kseniia Maiboroda.

[Music] and what we're hearing is we're hearing that personalization is very key in all of this we heard it in the first session and again it brings a 25% increase in Revenue so it makes a lot of sense now to talk about it in depth with our next speaker cassena Mal Mala yes CEO of ele eleatics there we go and uh personalization is a Hot Topic the stage is all yours welcome uh hi I'ma my and uh I Am Co and

co-founder in eleatics and I help mobile games to get their high technology by very easy way uh to grow your revenue and basically happy to be on the Pocket Gamer event again and uh thank you all for joining on this session so today I would like to speak about two Trends about two directions around the mobile games development it's uh user experience and machine learning uh and U maybe you have some thoughts uh in your mind now how it can be connected what's the common ground and why we are talking about these two things uh in the one session but I promise in the 15 minutes you'll get all answers to these questions and I'll share the Practical tips and tricks how you can develop the uh machine Learning System especially for your game how you can use it uh get something from the market on or make it uh like in house solution for your company uh so first of all let's uh speak about the user experience uh in

the Modern Life uh it's not uh connected to some domain anymore we don't split it and uh we have just general habits in online ecosystems and these habits are shaped by the big players on the market uh big uh ecosystems that we have it's social networks it's Ecommerce it's some TV platforms or something like that and everyone you and your players and me uh we use this stuff every day and maybe after hard work day you come back home open your favorite TV platform I don't know what you prefer Netflix Prime video or Apple TV something like that and you get recommendations and basically you are happy that you don't have to waste your time for find something for today and you basically happy to pay for that because someone care about you they can make the personalized content for you based on exactly your preferences not I don't know some advertisement or something like that that's exactly what you want to watch today some great TV

show or good uh uh movie or something like that and uh basically uh we have three things that we get from these big ecosystems that we cannot ignore when we develop the mobile games the first one is personalized content that we talk about now uh the second one is subscriptions models and the second one the ecosystem approach in general uh so what about our players uh the way uh how players Buy has changed for good and they have absolutely different expectations now they don't want generalized content that would like to have personalization and they don't want to have like prescheduled event they would like to have the dynamic content inside of your game inside of your Ina purchases offers inside everything that are around the game people come to the game to play you know they they would like to enjoy your game play they don't want to close a lot of Windows of your offer some popups or some uh very smart features that you uh would like to

monetize through the uh players and uh uh they just expect to have their kind of Netflix experience inside of your game because they are happy to pay for that just imagine if you open again your favorite TV platform and you see there are absolutely not relevant content there your thoughts about that okay the price is too high right I don't want to pay for their system that not care about my preferences I don't want to waste my time to click next next next next and U looking for something that is relevant exactly for me uh and the same your players feel when they open the game and they get five popups of the personal offers on the login and they need click close close close close and they just don't want to buy it they want to play it's not relevant content it's not relevant bundles for them uh so talk about personalization and generalizations what the difference so basically the segmentation is not a rocket science and probably all of you use it in some way right

uh but what's the general generalization it's when we are trying to find some common parameters for the players and make from them groups so we not have really personalization for the player we have some general approach for the groupes of these players and the parameters that we use usually is location platform uh operational system maybe level that displayer uh play right now uh and uh following these parameters we are trying to find some the common approach for the thousands of players in one group and based on this group we set up the offers and based on this group we are trying to make this offers effective and to make that we have to set up a lot of offers and hope that at least one of these offers would be bought uh and U just it's logical because we are trying to just justify our CPU and uh that's the reason why our retention rate can be not so good good as we expected to have

and retention rate it's like very important kpi and uh uh my point of view in eleatics and how I do mobile games and the server service for mobile games that we don't have to choose between LTV and retention rate because it's not there you know only LTV is Main and retention rate is not important we have to make the Smart Balance and to make the Smart Balance uh personalization uh is the thing that will help us to do that what's the uh point of personalization to analyze every player in game Journey to analyze exactly behavior and action of every player inside of the game what's the action after level win what's the action after level lost what's the engagement parameter of exactly this player and in addition to demographical parameters and all the stuff that we use for generalization we can analyze a lot of things about the purchasing Behavior impression for every offer and so on and based on the analysis of these parameters and their combinations we can

build the custom personal offers for everyone in the game what I mean when I'm saying custom offers we have the custom bundle custom price and custom plays inside of the game that are really comfortable and personalized for exactly this player inside of the game and we can set up we can set up the smart timing and smart frequency for every player because someone prefer to buy every session but someone just once a month and uh we have to uh adjust our our user experience for everyone who is playing your game and now we will move to the machine learning and uh look on some examples so you can see here two handsome mens the King Charles and oor born B uh both male born 1948 raised in UK married twice lives in Castle wey in F so we can look on these parameters and say okay uh they're in the same segment right based on these demographical parameters and uh probably

that's true and we can cover only 60% of the players with this approach because uh you know in the Privacy first world a lot of players just doesn't want to consent with gdpr dialogue they don't want to share idfa and so on and so on so if we will lean on the demographical data we can cover only 60% of the players uh and now uh we can uh apply the personalized approach here for the same people and we can we can see that for the King Charles for example uh we have the parameters like place two session per day one purchase per day average bill around $100 uhar love boosters and close the game if uh level lost and I prefer to buy our purchases on the login but for o osbor we have absolutely different metrics you can see it on the slide with this approach we can cover 100% of the players because we don't use their sensitive data to analyze the behavior inside of the game

and based on these parameters you can see that approach for two of these players should be absolutely different because in gaming behavior and their habits are absolutely different but based on the generalized approach as you remember they would be in the segment right um so why I'm talking about the machine learning in combination with the user experience because uh if you would like to analyze tons of parameters and their combination we have to build some powerful enough tool and machine learning that's exactly the key to the personalization door because uh to make personalization we have to act in the real time so we have to analyze tones of data in the real time and provide the right recommendations right now machine learning can uh like can provide this experience and based on the uh my experience with our customers on the market uh I can say that uh using the machine learning for the personalization

of an up offers you can get double average revenue per pain user increase your LTV significantly and very important to grow your conversion rate uh up to 5 7% uh because without conversion we cannot build the long-term relationship with our players uh so we have two options now how we can get the machine learning for the mobile games the first one is to build inh housee solution and uh the second one to integrate third party from the market uh let's see the pros and cons of the both options and let's start from the in-house solution I'll share with you the main points that uh you you need to evaluate before you will decide to build this solution and uh some tip and tips and tricks uh how you can choose the right models for building this personalization system so first of all you have to hire the right experts to your team because the technical stack for the machine Learning System are absolutely different from the technical

stack for the mobile games development you'll need the data scientist analytical team ml uh engineer and uh devops at least uh then the next one you have to choose the technical approach here is two options the first one to build algorithms from the scratch by your team so from A to Z you build the algorithms and ml models the second one to buy existing on the market models for example from AWS they have really powerful models they quite mature they're ready to analyze behavior and make recommendations uh but here we need to understand that the general models that we can buy from AWS for example they are not targeted exactly for the mobile gaming they are targeted for all systems I mean for the uh food delivery for the Uber for the mobile gaming for Ecommerce and so on and so on so the level of accuracy will not be so good as if you build it from the scratch exactly targeted for the mobile games uh the second uh the the next one is choose the

uh type of the model we have four types of the machine learning models and for mobile games basically me uh use and I advise you to use that uh it's uh semisupervised uh machine learning models what does it mean uh it's when the machine learning trained not just automatically by itself but you have the analytical team who review the results of machine learning and give the feedback to machine learning class like this is right this wrong this right this wrong and machine learning model retrain again based on this feedback if you would like to build the system exactly targeted for the mobile games it's the best approach and uh like your team can do it just like in house solution uh the next one you have to ensure the big enough relevant data set because the core uh fact that will ensure the accuracy of your machine uh learning is not just the right model but also the right data said that you will choose to

train this model and as we buil with you the uh machine learning model for exactly the mobile game we have to uh build the data set especially with the behavioral data uh for the players of mobile games just to give you the preliminary numbers if you have the mobile game with du average du like 1 million you'll need the data set for the last last six months to start the train your model and the last but not least it's the training timeline so how long uh after we developed the model we ensured the data set how long it will take to train your model and you can go live and get some result uh again if we are talking about the mobile game with du preliminary 1 million it will take around four five months after everything is done so you run your ml live and it will take four or five months if you have 1 million du to train model and to start to get real results so you can

just evaluate if you would like to make inous solution if you have enough time enough budget to hire this team so in general the timeline of developing the system is around one uh or one and a half year uh so it's good investment but depends on your game life cycle and your goals for now so the second option is to get some third party from the market and uh integrate some Tool uh for example in eleatics we are doing machine learning for mobile games for 3 years we have quite mature models now and we provide the results after just 3 4 days after integration so it's very fast solution and I would like to share with you the points how you can evaluate this short party because we have a lot of liveops tools and probably several liveops plus ml tools on the market and uh I would like to share with you the like um a little bit uh tricky tips how you can evaluate and understand Roi uh if you would like to make partnership with some uh tool from the market so first of all

evaluate time and resources what does it mean how long it will take to integrate this tool and uh how many people you need to operate this system in the future how many people you need to support the technical staff how many developers you need to integrate all your features to this system uh because uh if we are talking about the tri period it can be just one month but if we are talking about the integration it can extend this one month to very long period of time so it have to be clarified before you go with someone and sign some contract uh the next one uh commitment to the longterm machine uh machine learning training timeline as I talked before the timeline of training is really uh core important thing that will ensure the accuracy of machine learning and if you try to make the partnership uh with third party and they required to wait until the first result four five six 6 months because machine

learning have to be trained means that you cannot be sure about result the really mature ml models can give result in a week if you have to wait several months until the first result means that this ml model is not quite mature and it's not trained on the relevant data set and they will use your data set to train their models so it's not the way I I think probably how you would like to make the business the next one review the sensitive data so uh we have to be sure before integrating some third parties uh which exactly sensitive data and privacy data will be used for the machine learning analysis because you know that now it's very strict rules from the both app stores and Google Play and when they review your build with the US SDK and if you will have there the huge list of the personal data of the players that you didn't use before you will have some traveles and every time when you will release your new build you'll get some questions from the stores so just review carefully the

sensitive data that's supposed to be uh taken by the machine learning that you will use the next one uh ass say the data sources so sometimes uh ml system connected directly to the mobile game sometimes they connect it to the uh external data sources kind of Google analytics or something like that and uh if you understand that uh the thir party will take the data uh from the another s party you just have to be sure that you're going to use both of them for the quite long period of time because if ml will be connected to this data source you have to keep this data source for the time when you're going to use the ml system uh model and uh the last but not least is about the billing uh I see that on the market uh very often third parties uh provide the annual contract because like machine learning needs the very long time to be trained and so on and so on and monthly contract will not work because like you have to wait for

the half a year to get some results and so on but basically it's the same that we discussed before if machine learning have to be trained for several months for three uh months means that machine learning model is not uh like mature enough so the monthly billing can be used for machine learning even if someone will talk you that uh no just annual and you have to wait six months to get the first results um so basically that's it uh we'll be happy to answer your questions you can connect me through the LinkedIn ping me just say Hi and um give me feedback about the sessions or answer your questions thank you thank you cassia and you will also be on the panel [Music]