Lifetime value forecast – How to predict your game's revenue

[Music] hello everyone my name is Vera I'm an analyst at dev - dev we help developers analyze their games and improve behavior and financial metrics and one of the most important metric that we work with is lifetime value and today I would like to share my experience in lifetime value calculation with you but let's start from the description LTV or lifetime value were still the customer lifetime value it's a revenue that average user bring to the project during his lifetime in connection between a lifetime value and very new is very close before because if you know how much money we get from one user we

can assume how much revenue you will get from 1,000 of users or any other amount of users and lifetime value is often used for traffic analysis important that users should bring to the project revenue that necessary there should be more than attraction quotes and it's considered that LTV should be greater than three customers customer acquisition costs it's tell us that project is successful from financial point of view that's why it's very important for segment LTV by different traffic sources to define most profitable sources and get rid of useful sources so let's talk about ways of calculating lifetime value actually it's not so easy because this metric is about huge about huge paintings and we should calculate it right now but we don't know how users will behave in they of how will they pay so we need to make some

forecasts of this metric and now I'm going to tell you about several ways that we used in our company for calculating elsevier first way it's ideal method actually we didn't use it but this method gets us a real value of a lifetime so how to do that you can close registration it's very simple but it's unreal method because nobody wants to close the project so another way is to select a cohort of users until all users stop using the app and after that after after they do it we can just divide total revenue by total audience that will be our lifetime value advantage of this method is that we get real lifetime value so it's not a forecast it's a fact value so we can use it and apply to different analyze it for different cohorts but disadvantages make this

approach useless because at first we need to wait for really long time well users stop using the app it could be one month or half a year because users do behave differently so we can wait for really long time and another disadvantages are that while we were waiting this users inactivity we can make some changes in the app launch some experiments where in external factors can influence on users behavior like for example if our product will be published on iTunes or Google Play homepage it will increase our audience but it was the Jiva start using a web that's why we can't apply this lifetime value calculated for the old coger to these new users which completely different that's why it's method is useless and we have two more methods both of them calculated by the

same formula like daily air PU or multiplied by lifetime and difference between these methods in lifetime calculation but let's start from daily air for you how to calculate it you just need to take revenue for specific day and divide by audience XF users at that day and that will be our air PU what about life time one way to calculate it is make expert estimation for example if developer know his project well you can define that if user were inactive for the last seven for example days we can mark him as short because probably he will not come back to the app this period can be not only seven days but any other period of time it can be fourteen days or even months so period of victory we will mark users as short

and after that we should calculate lifetime value for these users let our I'm sorry lifetime not lifetime value and lifetime it's the difference between install date and date of last visit so it's number of days that users spent in the app and here is an example let's imagine that today is 17th of January so we calculate lifetime we need to select users who are inactive for the last 7 days it means that they'll last activity today is 10th of January and after that for each users we need to calculate his life time so difference between these two days last activity build and install date that will be left and value of specific users and then we need to find the average value of all of these our lifetimes that will be our lifetime which we can use in formula to calculate a lifetime value advantage of this

approach is that it's really easy you can use just Microsoft Excel and some basic data like sessions and payments and you can calculate it without any developers by yourself but disadvantages are that we use only one value of daily air PU but it could change for each users during some time and the second disadvantage is about life time because we need to make summation which can be not very accurate we can assume that user will not come back to the app but really he come back in the man's world example but he already marked humans charmed and that will be our mistake so it's not very accurate and that's why we have one more way and one more approach to calculate life time I think you know that a lifetime it's an integral of retention or area on the retention curve to make this calculation easier we can

replace integral to sum of retention values so we can just sum up all the retention values for each day's and that will be integral of retention and our life time how to calculate it to make this calculation we need to know several retention values for example retention on the first day on sevens for teens then we need to find mathematical function which will describe retention curve more accurate here you can see on this example in the legend of this chart in our company which writes several functions to find the bet which of these function will describe will get us closer values to original so after you define this function you need to use solver Microsoft Excel or Google spreadsheet and using all of this data you can extend this line for a longer

period of time why do you even need to do that because usually we know retention for a short period of time for example four months and calculate lifetime for this short period is useless so we need to know retention for longer period of time to sum up all of these values so we need to make a forecast of retention and again using solver several retention values and mathematical functions we can prolong this cure for a longer time for example for coffee here and common question here is how to for how many days to prolong need this tension curve and there are two approaches how to define the moment where well to stop first approach is to again make an expert estimation for example if you know that your users spend don't spend more than two months

in Europe it makes no sense to prolong get this function for a year because lifetime of your users is much smaller another approach is to find the point when these curve stop to change our attention well it's tough to change and this curve became like a gorizontal line the each day you have the same retention values it's a signal to stop extending this line and after that when you create the forecast of retention you can sum up all the values as I told before and that will be lifetime of your users disadvantage of this method are the same is the same as a previous one so we use only a one value of air PU which can change but advantage is that we we get additional retention forecast which can

also be used for our experiments user analysis and any other activities and now I would like to say couple of words about calculation of CD for products with subscription actually LTV meaning here stays the same it's still revenue that average subscriber brings to the project but a lifetime and chor this description became a little bit different recently this it's all that lifetime its difference between install date and date of last visit but here lifetime it's number of subscription periods so if users was free subscription free monthly so his lifetime is free what about Swart again previously we need to make some expert estimation where to mark user as short and we use days of inactivity but here

we surely know that if users didn't problem get his subscription we can mark his as Jordan because he decided still to not use the app further and here you can see several formulas how to calculate the value for subscriptions let's start from the last one lifetime value its subscription price multiplied by a lifetime you can see that this formula looks like the previous one like when we daily air pu x lifetime but here we know self you can be different for users and for different periods of time but subscription price usually don't change only if you make some experiments it could change so these factors make this calculation more accurate because this these digits usually don't change so a problem here in this formula again

in lifetime calculation and here we can see three formulas how we can calculate lifetime for subscribers first one is vo method that we discussed before we need to wait while users stop elongate their subscriptions and then divide number of subscriptions to a number of subscribers that will be like time but again we should wait for really long time to calculate this lifetime did by this way the next formula is very popular you can face with it in the internet it's 1 divided by Chandra but if we see if you look at the example on the left side we can see that for example if you have 100 subscribers on the first period then 70 subscribers on the second one so it means that we lose 30 subscribers and to calculate churn rate we just need to divide such a subscribers on 100 subscribers in the first period that

will be charted on the first period and here we can see that churn rate of any each of this period are different and the reasonable question here is what which of these charge rate is to use in this formula actually whatever rate we use this calculation will not be accurate that's why we have one more formula is sort and we're weak it looks scary but actually it's easy here we need to use short term Qian rate it could be charge rate of the first period then long term chan rate it could be churn rate of the last period you know in this example it could be a churn rate of the sixth period and that's it only two values and you can calculate a lifetime using this formula and it will be more accurate than later life time calculated by two previous formulas and what is important here when we calculate

life time for this for products with subscription usually in such kind of products there are several subscription plans and it's very important to compare lifetime when you for each of this plan for example a subscription price usually it is known for us also we can calculate lifetime and lifetime value for each plan and then define which of all plans bring us more revenue and after that after we define the most profitable plan we can highlight it and push users by it and also it's very important if you have unlimited plan and um because usually it's one-time purchase that gives unlimited access to the app and it's really important to understand what is more profitable to let users pay once or push them make small small payments but

regularly that's why we can be helpful here so that's it about calculations and now I would like to a couple of words about how to optimize lifetime value actually it's complicated formula which includes several metrics like retention pain share RP u RP p you and to increase lifetime value it's better to work with all of these metrics if you improve them step by step it will influence on lifetime value so I'm sure that you know how to work with all of these metrics that's why it's it's better way how to work with lifetime value and that's it great thank you very much for that so I'm going to it's opportunity to just check if there's any questions looks like you're you captivated so I

mean I'm assuming this is an amazingly difficult area because it seems so easy here's a bunch of clients they're gonna spend as much money surely we can predict it and it's obviously not very easy and but you know in terms of like there's obviously these very different methods and some of them are involved a lot more publication and can be found any kind of rules of thumb that of emerge in terms of you know like ways of getting to a kind of a number easily or is it always worth going down to the the actual Matt's actually I described several ways how to calculate lifetime that are really not very hard it's not it could make any person but now machine learning is developing in this area and it's really a mess and it's very complicated that's why developers higher specific people so to do that but this

calculation really is there it should be fair I think you're right I mean I mean I've tended to go for like 90 day as kind of a predictive kind of drop-off point and and because the numbers are often so low after 90 day for a lot of games it's a rule of thumb it's a kind of a nice easy way to have a very immediate number if somebody's asking you today but I think you're right these this price this is so much better so I'm gonna steal some of that and on that note thank you very much thank you [Music] you