How to Leverage Predictive Analytics to Enhance iOS Marketing Performance - Keren Gritzman, Pecan.ai

[Music] [Music] [Music] so I would like to welcome to the stage the uh last solo session of the day which is how to leverage Predictive Analytics to enhance iOS marketing performance by Karen from peekin coming up the stage so hi everyone and I'm really excited to be here can you hear me well yeah cool

perfect thank you so I'm Karen and I'm the gaming customer success manager at pan Ai and I'll be speaking to you about the wonderful world of scan which as I assume you know took away our user level attribution to channels and campaigns I will share with you how can you still regain your iOS visibility it's okay to put it there that's fine is it thank you ah okay thanks H so how can you regain your iOS business visibility by using Predictive Analytics thanks okay so I've been working with gaming customers for the past 10 years and in the last two years since I started at pan I've been working mainly with marketing and user acquisition teams helping them enhance their marketing performance through AI based predictions which is basically what pean does so just in a few words pan is a predictive analytic

platform focusing on the business value and the business kpis so we actually work with different um business teams in companies helping them use predictions as part of their decision making without having to rely on data science or data engineering teams we provide user level predictions H for various use cases such as campaign optimization and management H churn prevention increas in conversions uh mmm and more and we work with various companies non-gaming and gaming such as s playay dream games Bagel code holy water and many more so let's start from where it hurts and kind of climb up from there bottom line user acquisition managers you're not alone at the end of the day I assume that you like everyone are expected to scale and you need to find the best way to do it and that's not an easy task always

so after a few years of amazing growth in the gaming industry that I'm sure you're all aware of H we have experienced H some kind of decline in the last two years post pandemic it seems that the gaming consumers are simply spending less that means that the competition is increasing and the prices of acquiring new high value users is getting more and more expensive in addition the marketing landscape is getting more and more complex you need to find the best blend for your specific company you need to find how to invest your marketing budget the best way and if that wasn't enough privacy is getting stricter so we have apple that actually launched their app tracking transparency policy in 2021 and their scan framework and of course privacy sand books coming up soon and we are all kind of preparing for it

so that basically means that you need to get more out of Less in a kind of a jungle which your eyes have closed and that's not easy so the best way to do it is by actually maximizing your data using long-term H kpis in decision making being proactive rather than reactive and use H Predictive Analytics okay cheers so H you might be familiar with some type of versions of the same example I will share we still see it with customers today so one of the most common metrics used for campaign H management and optimization is LTV day7 the Assumption behind it is that roas day7 is actually a good proxy for long-term uh performance of campaigns

now in many cases it's completely correct it is a good indicator however not always and not only as we see campaigns are not linear we have some campaigns that actually acquire users to tend to convert very early on we have companes that actually acquire users who tend to convert very late and might have higher retention rates so you want to keep a good blend of both here's an example h of two campaigns H that share the same Target H we have the purple and the grein compaign A A and B so as you can see here we have on the xaxis the days since install and on the Y we have the roas of these two campaigns and we can see in the kind of zoom in on your left you can see that the green campaign actually did not meet the target Ross of day seven that means that the company uses LTV day7 as their

main metric will probably decide to lower the budget of this campaign but as we can see when time progresses this campaign is actually overperforming and reached at the break even Point much earlier this is what we call El Late Bloomer so we would like to keep both type of campaigns and especially the green campaign as we can see and the best way to do it is using Predictive Analytics and and why you could actually predict on day seven or even earlier what will be the uh day 180 or any day you choose LTV of your campaign you can calculate the Ross and then make make decisions based on that so you'll make decisions very early on but based on long-term kpis okay so I assume some of you are actually already doing it today and using it today in your decision making

and if not I encourage you to try uh but it's kind of straightforward when it comes to Android it gets much more complicated when it comes to iOS and the scan framework and I will share so first just to kind of align in a few words about scan so as you know um um scan is Apple's um framework to actually help companies measure their ad activity while actually protecting the users's Privacy that means that we actually get input about our users engagement however only on the first 24 hours we get it in delay in an aggregated level on a campaign level and also partial that means it makes every calculation much more more complicated than it used to be

okay we are unable to actually connect the user to the campaign or to attribute the user to The Campaign which is the Baseline to pretty much every metric we used until now now Roi basically became unknown that means that the only way we can make budgeting decisions is based on assumptions and early indications which as we just saw in the example before my actually be misleading so just to go through the kind of life cycle before scan so app developers used to spend their marketing spend on different marketing channels to acquire new users they can later monitor the performance of these users the revenue they generated and they could have calculated exactly how much revenue and what is the Ross of this specific campaign for long-term metrics like uh

roas day one day 90 now after scan we lack the ability to actually attribute as I mentioned the user to the campaign so we can monitor the performance and the only thing that we can do is take a look at Ross day one and even that is not completely accurate and full so I assume that your LTV overtime looks more or less like this the idea is that revenue and the LTV revenue is generated by long-term engagement and retention of users so if we can see only the 24 hours it is simply not good enough to make good decisions regarding our campaigns and the budgeting okay that is why we actually saw a clear

impact on the market after Apple introduced scan so we saw a lot of marketeers kind of shift their budget from iOS to Android which is a kind of a shame because I assume you are all aware of the fact that the arpu the average revenue per user of iOS users in general is higher than the Android users and in addition Apple was just the first to do it so we will have sandbox and pretty much need to deal with the Privacy anyway so we can kind of shift our budget we need to face it and find the best solution of how to work with it so having said all of that how can you actually still regain your iOS visibility I will share a glimpse to pean's solution and we can continue from there so the first thing thing that we do when

we H start working with a new customer is we run a deep analysis of our customers user level data based on this analysis we actually build predictive models that predict for each user what is the predicted LTV that he will generate on day 30 60 90 180 okay so basically we take into account all the data that you've collected on that user and in the gaming industry we collect a lot of data so we take everything into account the demographics uh his engagement the revenue he generated so far everything taken into account when we actually predict what is the LTV for this specific user in addition we actually help our customers optimize their conversion schema so we help them find the best way to build a conversion schema and also

Implement some predictive events inside their conversion schema to be used uh later on so once where everything is done H we actually changed the schema and went live this is the process and how it looks like we are generating a prediction for each user H regarding his future LTV we send this prediction ER to scan through the MMP and later receive postbacks those postbacks actually are being H analyzed and prepared to be used in the process that will come actually and then we distribute the revenue that we've predicted to begin with for this specific cohort we distribute this Revenue to the campaigns based on the postbacks using a machine learning algorithm that means that at the end of the day you actually have predicted LTV

on a campaign level for all of your campaigns that are on the scan of course and you can use it for campaign management and campaign optimization that means that if we said before that the only thing that you have with scan is the one revenue or day one Ras now you can actually regain your visibility and I can actually make decisions based on quantifiable uh results on the uh Ross of day again 30 60 depending what is the model that you choose and then you can actually really compare the compains and decide how to allocate your budget H the best way so we send our predictions to our customers so they can implement it into to their bi dashboard so they can pretty

much use it easily as they're used to be H to work until now we also H share our um dashboards of course but anyway at the end of the day what they have is for for each campaign on each time frame that you choose you can see exactly what was the cost how many installs did the campaign generate the CPI the day one Ross and the predicted LTV or predicted roas they can also see which type of users this campaign actually acquired okay the quality of the users they can monitor the performance and the installs the revenue over time Etc so they can make smart decisions based on that so to kind of sum things up H I'd like to share two example of two customer of ours uh one of them is a bagel code and H and you can see here a quote

from near their VP marketing I will give you a second to read and then I can drink a little bit both customers and not only them of course changed completely the way they made decisions in scan after they started working with pan and we can see here another example of a customer that before work working with us actually felt very uncomfortable making budgeting decisions in iOS and started lowering his spend decreased his user acquisition in iOS after the solution was implemented once he felt comfortable with it got his visibility back had tools to actually make smarter decisions he could then scale and increase his H iOS share H while actually meeting his Target Ross so you can see it here and um and the light purple line is when we actually

implemented our solution so no matter how you choose to tackle it if you want to choose a company if you want to do it on your own here are three takeaways I would love for you to kind of take from this session first don't settle for outdated descriptive app metrics like day one or day seven LTV or roas use Predictive Analytics use use your data wisely be proactive don't be reactive second build your conversion schema wisely it's amazing to consult with peers it's great to get inspiration from other companies and the way they build their conversion schema if they want to share it but at the end of the day you need to build something that is relevant for your specific app for your users's behavior and also pref ably include predictive signals in it and

last thing be be flexible and experiment now is the time we have scan for coming and we have sandbox these things change constantly use this time to kind of try experiment find what works well for you and then you can add additional layers on top of that that's it thank you very much and if you want us to help in any way you are more than welcome to visit and meet us in the stand thank you very much thank you that's great Karen thank you very much thank you now didn't leave a lot of time no no um we are coming close to the end but do we have time for one question maybe when we so um do we have a a question from the from the audience one down the front here please we'll get you a mic so that the people at the back can hear thank you and then uh hello G from multiplayer I don't know if it's working yes yes the predictive analysis what's when when you go 180 days from the first day that you

predicted what's the confidence Spectrum like how how much can you hit what you guys predict yeah well at the end of the you mean about the accuracy of the prediction yeah okay so it changes based on the data at the end of the day if you have a a large enough amount of data and also data about things that happened before depending how table your app Etc so we can reach even very high accuracy rates uh like 90 even up to 90 or or more than that but it's very much dependent on the specific data so it's not something like a metric that I can uh that I can say h but I can say that it will probably or not probably surely be much better than any multiplier or kind of Benchmark that a lot of companies bi Benchmark use today hope it answers your question cool thank you well we we'll we'll leave it there if anyone does have another question of course you can look car up on meet to match and come and find you

around the venue so thank you very much for your time that was great thank [Applause] you