Optimizing UA Budgets by D7 ROAS | UA Coffee Talk

hey guys welcome to you a coffee talk today we're going to be talking about a blog post from Nebo Nebo you create a blog post that was titled yd7 Rao's might be the worst KPI to optimize towards since CPI yeah so I thought we could just open by having you provide a little bit of context in terms of you know why did you write this blog post what's the kind of story behind you know your motivation to write about this topic sure so basically what and when I joined Network my my role changed quite a bit because it was a significantly smaller company yeah so I was not just responsible for like my campaigns I was also responsible for the money yeah the company's money and when it's gonna pay back right I mean that's what like every single marketing president should be that's right but that's not really happening okay so over time you know when you start looking to like the data and like trying to understand what's

happening with the money you're spending you like see these patterns and you see that most people are using certain metrics or KPI so optimize their campaigns and they're not really reflecting they're not really the best cake guys to look at or they cannot be looked at in an isolated way so that's something I've been looking I've been looking for a while thinking about it for a while and then when vo started really taking off and when basically most are spent Facebook's win spend moved to vo I started noticing these patterns in like pay attention and and cost prepare that were somewhat limiting to us and luxury and us to scale think the further and I felt like I need to need to you know just write about this and share more about this and explain how you know different optimization models impact the count of a certain business and just for our audience out there when we talk about what do you mean by optimizing towards a KPI and so you're talking

about when you're trying to determine your you a budget or how much you're spending again you know it gets a specific channel or something like that then use that KPI to allocate how much budget to spend right yeah exactly so so initially it was just CPI we will find sources with the lowest possible CPI and then spend more money bigger right then people were like oh my god this is all fraud I'm [ __ ] I should like you know but dude this more yeah you know in a better way so people moved away from CPI to Ross and you know the campaign's that have higher d1 or these seminars get more budget and as a result you know your overall Ross should improve but it's not always the case because it's just your short-term or improves but it doesn't really it doesn't really mean that the money's being back in a row as we mean return on adspend exactly you know you actually had a session there was a an event you a master summit by the dalgo where we have like 25 you know you a experts from San Francisco and you ask the question what

KPI do you guys optimize against I think with it 24 out of 25 or 25 out of 25 were d7 rats right yeah it's easy to work with you senator s I mean it makes a lot of sense the cohorts are mature enough so you can make some conclusions based on after seven days based on the quality of that audience but at the same time if you just look at these seven rows it's not it's not good enough so you have to kind understand what the retention is what optimization models you're using whether it's Facebook Google is it like instant or a rewarded video so basically I just try to like explain how these different levers impact the performance and and that if people want to spend money in scale they really need to like look beyond just and just kind of like even stepping back like why do you think people use d7r ads it's like so you know one seven days because you want something that's quick so you know seven days kind of gives you a four week of data but you know I I guess that the

issue I have is that you know for it feels like a lot of you a marketers just want to like they don't understand the why they just want to know what what do I do what's 37 her last target and then they don't really question it the short answer is we're lazy like we I just want the most simple KPI if I can drive can execute and this go boss the wall to it yeah and and that that's fundamental issue with our space is that that we can't think of just taking the the most easiest KPI because it doesn't fit and that's why I really like your article because I remember when I was at play studio is like when when the action was a view or a you I can remember it was when Facebook recently launched a new kind of bit outdoor mid optimization model and and then we're like oh man they said Ross is great there's still a ton of money out of it then we look at like day 60 then we're like oh that's

this interesting because like what we saw was like on day seven there was a reward of eel Network where the robots was lower but the curve of that that was much more consistent over a long period of time whereas like Facebook would fly now and so if I look at day 60 of anything that your water video network actually did back yeah right so in that particular case you're like yeah that's because of the cost basically with like these that's rusticated optimization walls you pay significantly more very user and then your risk profile increases right and that's what most people don't really understand and that's why I like the article that you wrote actually talked about like the cost the driving of the cost of acquisition yeah what like yes and for me for me actually getting that user that is paying money I think Facebook does a really great job yeah because it's so expensive to get that user on that platform it actually makes sense and and not to say like you know don't spend money on Facebook it's just like you have to what

we we realize is that oh maybe you really should have a different KPI by Channel yeah obviously if you're working with a hundred different channels it's very very hard to do that but let's say like even if you dilute it to something you know and in between between like ultra granular first is like the laziest thing is like even like oh for rewarded video like this is going to be my day to save much more focus on top end channel bye-bye yeah yeah yeah and break it out like that because you understand like different placement types different channel types have different payback windows and you have a different goal for each right when it's the issue arises I don't think there's a fundamental issue with data Munroe ask the issue is when you consolidate all your channels into a single metric instead of separating it out yeah another reason just to get back to your original question why these seven Rossi's is popular it's easy to report in these seven rows report upwards

that's another thing that that I didn't really understand when I was part of a much bigger marketing team yeah it's like executives don't really understand what costs per these seven pairs is yeah I mean they do but it's just like it's not a very like it needs you to understand and to report on metric and as a result like Visa Visa Ross is significantly easier everybody understands it you know it's how much money returns by day seven right you know the heart is the better or not really so that's what the article is about and I'm glad that you know people really it seems that people really enjoyed it and I'll try to and in terms of like so what some of the key points from your article is that you know basically the monetization profile based upon channel by geo by platform all look different so if if you are gonna use the g7 or as target man you need to separate that stuff right yeah exactly yeah and I think the other thing that I'm not sure is what I understood is why like just

even going back even further why and how but these seven rounds metric is computed right and so I think some people don't understand about audience basically what the way these seven rows is is actually calculated is you you try to take your LTV and you're assuming some type of RP curve right and you're assuming that there's a monetization profile that grows over time yeah and you're trying to think about if you're on target in terms of the growth what where should the row ass be at be seven right yeah but the other thing that that isn't taken into count is that for a lot of games that are cucurbits going to look different yeah and so like your tape it's basically just a multiple small yeah yeah and but you know those those are peepers could all look very different and yeah you know users in Russia behave completely differently than users in the u.s. right because of like multiple

factors and you cannot really use the same the same solution to estimate what what the LTV is going to be or right money is gonna pay back I mean we internally we we have predictive LTV models and we use those I understand that's that's a challenge for a lot of companies but at least you know figure out what are your top ten sources right you know whether it's like Facebook is in the US and Facebook Android and in Germany and then try to have like custom multipliers for those right and so I think where you could run into trouble is like let's say you your typical RP curve your d7 or ass target is let's say it's 30% but then if you're running sales or if you know you're running across a specific channel that month sizes very quickly and then flatten it out yeah then you can have a significant miss right yeah especially if you did in the first seven days so I think is whatever you optimize for one metric that means that you can also like you know a spoof it or just like hit that

metric keep it metric and not necessarily hit the goal right that's what happened with like CPI initially then like we have installed fraud with these seven rights we have our space bidding yeah with you know whatever the target is well again with CPI you can have misleading ads so whatever the time is you can hit it but that doesn't mean then you're gonna win and then I think d7 rise is great when you have existing data right so like let's say you've done 50 match 3 games and you're doing a new match 3 game you're like oh yeah these seven rows target should be this because we have a lot of data you know exactly what that yeah our poker is gonna look like but for a new game like how do you guys how would you guys come up with a d7 row as target for for a game that's you know that's never been yeah watch before yeah it's really tough I mean it was not deterministic you can yeah you don't have the data to to you know and also for new games you can't expect the money to pay back significant pretty fast yeah and then you can kind of extrapolate

based on the early data what the yeah I think of other things I like to do is just basically have a soft launch in soft launch look at the shape of the curve exactly yeah and then map not the not the absolute values but the shape of the curve to your you know like you mean like if you vote for a western market try to find a market that has similar curve but then just kind of extrapolate it up yeah so I guess my counter question to that is like if it's a brand new game you don't have the data you can at least see that early to shape of the curve yeah how do you how do you know what you own the longer-term engagement of that that's why in soft I'd like to see as long of a curve as possible and and then just just you know I mean you have to do something so I would model that curve to the hard launch curve so with that said what would you recommend as like a decent soft launch period so

that you can get that data so that would depend on the genre of the game if it's match three three years that's the problem if it's hyper casual seven days right you know so I think like you know going back to play Studios that is our Israeli team pops lots to do I did an incredible job and kudos to them for building out a predictive LTV model to kind of measure performance because like they did come to realize like just looking at these seven rows like they're not all the same right so they said she's making variable as I said Michael what geo did this user come from what is the user engagement within the first couple of days like how length of session depth is I mean cost yeah and often like yeah and then and they ran that test for over here right because they realize like you can't it's not something you can just do overnight and over a course of the year they fine-tune it and it got to a point where they're close like 90% plus accurate with their

prediction and that's insane really and this is I think one of the issues that a lot of people a lot of like novice game studios don't understand is that the companies that have done 50 match threes or 50 slots games they've got so much data that they have you know PL TV predictive LTV or ARPU model so that when they launched a new slots game where we match 3 games they're way ahead of the curve then you are in terms of being able to understand how much they can buy you know what to set their row as target what to set their payback window to all that kind of stuff if you go into any like startup especially gaming startup meetup or whatever that's probably like the number one question that's being asked how do you define your early indicators book so that's because lobbying don't know they don't have that data and I kind of have seen that come up with a number of times but going back to your blog post so then so you're not necessarily saying don't

use g7 whereas right but like what what approach would you specifically recommend in terms of yeah it shouldn't be used as an isolating metric basically if that's the only because the reality is like most most themes don't really have a 10 percent data science team yeah or most companies and as a result of that you have to be a little more scrappy and use metrics that are easier to understand help calculate calculate so if you end up using this address at least define the benchmarks for different traffic types yeah and different sources and that should help you hit your goals like business goals faster so and I I guess the main point of the article is just to be more gonna come as enough of what what are the what are the changes in the market that happen because of our space bidding yeah and you know if you take those into account then you'll be you know it's more likely to succeed in in achieving your marketing goals or like pumping goals right so that's it I think just to add on to that like a lot of times even

certain established companies when they launch a new game they there's two questions that often comes up and one is like how much budget you know when but I think it's a different conversation but we didn't bring a little bit in here it's like how much budget should we allocate to the launch of this game or with the channel breakdown right and and I think sometimes the disconnect is that they don't look at how the channel breakdown will baking to the DLT vo of that game right they just try to consolidate into a single thing and so it's all one curve not right if you decide to launch with 60% on same network no says 30% your kill t-butyl good that should be naughty but your repayment window might be very very very different so that's a case like that should also be taken into consideration and then in terms of like when you're separating out different monetization profiles or RP curves so

you know some of the some of the key key factors to look at imagine were you know channel platform geo is there anything else or are those three the creative side could be one more you know if a user is coming from like a playable or says the video can be quite different assuming that the playable works as a kind of creep tutorial yeah so there's a lot of different it's really up to like the the scale to spend because if you know you're driving a hunting solo dates gonna be really hard to go that granular right but if you're driving tens of thousands of installs a beta and you can go pretty granular early and you know do different like analyze different slices and make different gate guys for those okay cool any other thoughts in terms of this topic but it sounds like you got a lot of great feedback in terms of your blog post right yeah I mean I expect me to do well I did that clean baby article headline in I had money on fire as my

like feature photo I mean I'm super happy with the with the feedback yeah oh I'll try to write more it's it takes a lot of time and I'm not necessarily very feeling really comfortable writing in English but I think I should did more for sure and cover some of the of the topics that are these are not really general talk they're like targeting just people doing who who've been doing you a for a while I really like the articles vaguely because it calls all the question I think everyone at some point but night but I do also think that it might be kind of specific to you a folks about do you think that if more executives really understood these concepts more fundamentally they'd stop making kind of down the state's Facebook quality a party ska their feedback was really good because they they kind of like sense that you know this is happening but they were not really sure okay because everybody is just like using video and

it's people are using it because it's working but at the same time you know what are the problems of having the product that's that's creating a hyper competitive environment so yeah exactly I mean it's good that executives that are aware what what are gonna hopefully will be will be aware even on that on the on the sales side is important for the product teams of the companies of Google or Facebook to understand this yeah right because oftentimes like in the past when I've interacted with Facebook product feels like they're they're also building in the vacuum right they think that I'm building something hey I can optimize tours payers like you should pay as much as you need to get this individual but then they didn't realize that goal they're you're creating such a hyper competitive environment that you're isolating it to only certain companies that can afford to pay that if I step it becomes a you know how deep is your pocket tempo game yeah versus you know LTV and you a skill yeah which is always

interesting because like when you're talking to startup legged people are like oh I just wanted to start spending should I spend my money on Facebook on like how much money yeah you know yeah so I think that's a very interesting topic but no well-written article thanks guys alright great and that's a wrap for our neck latest are you a coffee doc yeah thank you [Music] [Applause] [Music] [Music]