ABOUT MISHA: LinkedIn  | Twitter | Huuuge

ABOUT ROCKETSHIP HQ: Website | How Things Grow | LinkedIn  | Twitter | YouTube

KEY HIGHLIGHTS

💵 What programmatic is – and how it’s different from buying on Facebook & Google.

🕹️ What sort of companies or apps is programmatic a good fit for.

❌ The common mistakes that advertisers make with a DSP.

🧪 How to vet a DSP before working with them.

🖼️ The role of creative in programmatic, especially considering that it is a black box.

🏠 What it takes to do in-house programmatic buying rather than through a DSP.

KEY QUOTES

Your spectrum of DSP options

There’s a relatively new self serve DSP or self serve programmatic options available on the market. This would be very similar to Facebook in the sense that you have full control, you can make a decision where you want your ad to be served, on which audience lists, on which look-alikes, etc., which bidding options. However, the challenge here is that because you have all the control, you want to do changes very quickly and you may end up not maybe waiting enough time, or not plugging in the right algorithms.

Why look beyond Facebook and Google for ads

It’s also good for the companies who probably have some limitations with Google and Facebook; for example, gambling is a highly limited area where media buyers cannot purchase media everywhere. In our case, we are social casino advertisers. We also have limitations on Google, in which countries we can purchase media; for example, in Russia, UAC is not available for social casino, whereas Facebook is available but it’s rather limited in terms of inventory. So if we, as a social casino advertiser, want to purchase media in Russia, it’s good to go beyond Facebook where Google is not available.

The programmatic learning curve

If you go on Facebook, for example, and you promote, as in our case, social casino games, Facebook already knows who is the right user. So if you upload another social casino game, Facebook knows what kind of user they had to show ads to. 

If you go to a programmatic media space, unfortunately, they cannot use other advertisers’ learnings onto your company, so they have to start from scratch, and the first few weeks there would be exploration where the main purpose of programmatic would be to lower the customer acquisition cost, so the CPI. Then once this is done and they are able to collect payers’ information, they will start adjusting to modify the algorithms to acquire the right payers, not just the lower cost per install, for instance.

What really makes campaigns work

The creative is not the number one reason that a campaign will be working or not, but the targeting is something that would define whether campaign and cooperation is working. 

The importance of transparency

One DSP partner who has purchased a few hundred installs from publisher, let’s call them ABC. The quality was great. However, the concern that I had was that this publisher was—where a few friends of mine are working—doesn’t serve ads, and they have a different model and so I reached out to them directly and I asked, hey guys, did you change the monetization model and start serving ads. The answer was no. 

So I brought it up to our DSP partner which understood the case and took it with exchange where the installs were purchased from, and then we got charged back for the amount of money that we spent on this publisher, which then was fraudulent. 

FULL TRANSCRIPT BELOW:

Shamanth: I’m very excited to welcome Misha Syrotiuk. Misha, very excited to have you on the Mobile User Acquisition show.

Misha: Same here, thank you for having me.

Shamanth: Yeah, and we’re going to dive into an area that I’m very, very curious about myself, just because you’ve been in this area at such a huge scale for so long and developed so much expertise. This is absolutely something I would like to dive in with you, and this is programmatic. To start with the basics, tell us what programmatic means and how it’s different from what a smaller app developer might do with doing UA your way on Google, Facebook, Snapchat.

Misha: Sure, yeah. So programmatic, the definition of programmatic is find media in real-time. What’s important here is on multiple ad exchanges in multiple ad formats and using CPM as pricing model, so this would be a very basic definition. When we speak about Facebook here, we mostly buy Facebook inventory, so you see Instagram etc., and a media buyer has a massive control over their campaigns and budgets and how they perform. When we talk about Google and in Google there is more, there are more limitations, but still user pretty much is buying Google inventory, that includes as well search inventory which unfortunately is not available on Facebook or programmatic. 

So programmatic has a lot more in common with Google in my opinion than with Facebook. So similarities here would be that, you know, as similar to as in Google you would need to wait a few days, a few weeks for campaigns to start spending. It’s also not advised to make drastic changes on a daily basis, and you kind of would have to trust machine learning and the black box. But there are also a few differences, so in programmatic world, there is mostly no self serve dashboard, you have to talk to your account manager, there is very little control of where the ad is served. 

So unless you specifically highlight, I don’t need, don’t want this publisher, then you most likely will end up serving your ads all over in all the list of publishers. There is also very little targeting options in terms of interests or demographics. So this would be relatively similar to Google examples. And now what’s important here is also that there are two kind of programmatic media buying options. One is self serve and one is managed. When we speak about programmatic DSP, we usually talk about managed programmatic DSPs, and those would be big companies, few examples of them, Liftoff, Aarki, CrossInstall, Chartboost DSP, etc. Now, the way they would be identified, I would put it that way, is that, first of all, you would talk to your AM, there would be very little things you can do by yourself other than set up the daily caps, send the creatives, or even talk what campaigns you want to have. They would be also needed more time to split up, and they would be again very similar to Google UAC in terms of understanding what is this.

There’s a relatively new self serve DSP or self serve programmatic options available on the market. This would be very similar to Facebook in the sense that you have full control, you can make a decision where you want your ad to be served, on which audience lists, on which look-alikes, etc., which bidding options. However, the challenge here is that because you have all the control, you want to do changes very quickly and you may end up not maybe waiting enough time, or not plugging in the right algorithms.

So most likely your user acquisition campaigns will not be performing well, at least they were not performing well for us. And so, two examples of those self serve DSPs would be Kayzen and Appreciate, and there probably would be more coming soon. And I do recommend using them, especially for retargeting. For user acquisition, unfortunately, we did not make them, we did not happen to make them work.

Shamanth: So it sounds like this is media buying outside of Google, Facebook, Snap, so their walled gardens, and which is perhaps why some smaller apps that we work with don’t necessarily need to look at programmatic. But I’m curious as to what kind of companies programmatic is a good fit for, and also tell me what inspired your team to start testing on programmatic the first time you guys did it, what was your trigger at the time?

Misha: Yeah, so maybe first part of the question, what kind of companies programmatic is good for, I think, first of all, it’s good for companies that have already explored media buying on Facebook and Google, maybe some other channels like Snapchat, Twitter, etc., and I’ll look in to explore further options. It’s also a good fit for the companies that have big investment opportunities in user acquisition, have patience and actually can wait a few weeks or months for the results to mature. It’s good for the companies who have product with not necessarily low customer acquisition cost. And one more thing,

it’s also good for the companies who probably have some limitations with Google and Facebook and, for example, gambling is highly limited area where not everywhere media buyer can purchase media. In our case, we are social casino advertiser. We have also limitations on Google, in which countries we can purchase media and, for example, in Russia, UAC is not available for social casino, and whereas Facebook is available but it’s rather limited in terms of inventory. So if we, as a social casino advertiser, want to purchase media in Russia, it’s good to go beyond Facebook where Google is not available.

Another example could be Belgium. Recently Facebook has not stopped allowing advertising social casino in Belgium. So except Facebook, then there is Google available, and obviously everybody else, so there are, there are limitations in geographics that you should consider whether your product is suitable for. And the reason why we jumped into programmatic because we have been doing UA for some time, and we have those two limitations that I already mentioned. But also we are always looking to acquire new media sources that will be able to bring us extra revenue in the very niche market that we operate. So it happened a few years ago and we do work with those programmatic media partners up until today, and they became, probably in between 20 to 40% share in terms of UA spend depending on OS or country in Huuuge Games’ portfolio.

Shamanth: Wow. Yeah, and those are substantial numbers. You did say that programmatic is usually right for companies that are, A, willing to make a big investment, and, B, are able to wait for the machine learning, by which I’m also hearing that early on the numbers are going to be terrible. Is that always the case, can you explain why companies need to wait for the machine learning to learn and why those big investments are necessary, and what type of big investments are we talking about here, what just numbers wise?

Misha: Sure, sure. So majority of the managed DSPs out there, well, want to sign insertion orders for at least a few thousand dollars, so, for example, $20,000 or $50,000. And the reason why there is even a need for exploration is because they are starting promoting your game without knowing what this game is all about, what kind of users are the right users.

If you go on Facebook, for example, and you promote, in our case, social casino games, Facebook already know what is the right user. So if you upload another social casino game, Facebook knows what kind of user they had to show ads to. 

If you go to programmatic media space, unfortunately, they cannot use other advertisers’ learnings onto your company, so they have to start from scratch, and the first few weeks there would be exploration where the main purpose of programmatic would be to lower the customer acquisition cost, so the CPI. Then once this is done and they were able to collect payers information, they will start adjusting to modify the algorithms on to acquire the right payers, not just the lower cost per install, for instance.

So this is the reason why the exploration needs to be on few weeks’ period of time, sometimes it happens faster, there are a few DSPs out there where you can actually pre-send them a selection of your user base from history. For example, top engaged users, medium engaged users, and not engaged at all, and they would analyze this and feed the machine learning with this information. So your initial first weeks of UA campaigns might be more successful than would be with somebody else out there.

Shamanth: Right, right, so it’s the learning phase, except, of course, these DSPs don’t have quite the amount of data that Facebook has which is what takes the time to build that learning phase.

Misha: Exactly.

Shamanth: Got it. And when working with the DSP, what are some of the common mistakes that you’ve seen people make? 

Misha: Yeah, so by common mistakes I see people make, I would rather say I see, we made or I made while working with the, with the piece. So those could be, mistakes or some of the things that maybe could have avoided. So A, scaling too fast. So as a user acquisition manager, you want to scale, if the results are good. It could be that they are good, because of the random few purchases, a few payers. So A, scale too fast. B, pausing too fast, similar but the opposite one. C, 100% trusted the account manager on the DSP side. 

I’m not saying we shouldn’t trust our account manager, but we should challenge them very often, and not necessarily 100% listen to the best practices that are recommended. We could be, we should be more creative and, I would say, suggest different ideas than they would be able to accept and push it. 

And one more is not being transparent, so not necessarily that we are not depending on DSP to be transparent, but us as an advertiser not being transparent with our DSP partner. So what I mean here sending, for example, postbacks in the real time, so our DSP partner would be able to verify the quality and sharing as well a list of devices they already have, our games installed, so they wouldn’t be targeted again etc. And maybe last but not least is not asking hard questions. 

So what I mean here is just to go very deep into details with your account manager, into all different machine learning details that you can, and the example here that I can bring is we a few, some time ago, a few years ago, we started working with one DSP and that was also great. So after a few weeks of scaling we actually ended up asking very hard questions, like, what kind of exchanges they are buying from, how the machine learning works, how they optimize towards payer rate or any further down the funnel events, and it turned out that the answers were not existing. So we realized the reason why this DSP was successful was rather luck and not actually any big machine learning behind it. And few, some time after that there is also dropping performance and we ended up pausing this, this partnership.

Shamanth: Yeah, I’m curious to dig into some parts of that, because when you are like, look, I want to understand what’s happening with machine learning, what are some of the aspects of machine learning that you seek to understand, what are some of the questions you might ask to say, oh let me understand how machine learning works. I am trying to understand what are some of the things one might look for as a UA manager that’s trying to evaluate, you know, a DSP from a machine learning capabilities point of view, what would you advise them?

Misha: Sure, sure. So maybe let’s look at the example of Facebook, what Facebook has few, or even Google UAC have a few ways of optimizing. So in Facebook you can help mobile app install campaigns, you can have app event optimization, you can have value optimization, right. So these three different algorithms has its reason behind it and has its way to optimize. So similar with also the other DSPs have. Some DSP would have machine learning approach to optimize on the day seven payer rate, some would be optimizing on lifetime value, some would be optimizing on cost per payer or cost per purchaser, etc. And majority of them would be changing models, they wouldn’t just have one model and they would run forever. So it’s good to challenge them and actually ask, okay, why your model is optimizing on lifetime value if maybe I want to have the cheapest installs at this moment because I’m looking to buy as many as possible to be ranked number one in the App Store. So this would be examples of hard questions that you can ask, not necessarily to an account manager, but feel free to bring it further and talk to their product team and see what they have to say about it.

Shamanth: Yeah, and I imagine that’s also important and necessary just because there are so many DSPs in the market. I know you probably get tons of pitches, you’ve evaluated so many of them, right, and certainly something we talked about earlier, the last time you spoke. So what are some of the things, other questions you might ask other than machine learning, and how would you assess that test period once you start to work with the DSP?

 Misha: Sure, yeah. So I guess, before we would onboard a new DSP partner or maybe any other partner, we would be able to divide this new potential leads into verification stages. So I would call it that away. So the first verification stage would be when we actually talk to them directly and ask about their past performance, so let’s say, a new programmatic partner wants to work with us and sell amazing inventory that they have. I would ask, okay, tell me whom do you else work with in the gaming industry, send me a successful case study. I also have some connections, so we have some connections within the gaming industry, so I can reach directly to that company that they would mention and verify whether they actually worked with this, with this new programmatic partner, etc. And moving forward, we would, with all the new partners, we would actually ask for test budget which could be a few thousands of dollars, and the reason – and this test budget would be actually for tests, used for tests with no obligations from our side. And during this period of a few weeks we would require at least 50 installs per day for a period of at least 10 to 14 days where we would be able to assess longer cohorts quality of the longer than seven days’ cohorts. And after that if the results would be satisfying us, they not necessarily have to meet expectation 100%, but they have to show the positive trends. We would start moving with the regular IO and this is our current verification process that we would apply to all the new partnerships.

Shamanth: Understood. And when you speak about the test budgets, and you said 50 installs per day, that’s what you look at to evaluate, my understanding, and I worked in social casino a couple of years ago, my understanding is social casino is very whale driven, and do you find that just 50 install today gives you enough of a good idea to where a DSP can identify payers and whales for you?

Misha: It’s a good question. So of course, it’s not enough, but it will show a trend. So let’s imagine we’ll have one whale per every 20,000 installs. So 50 installs per day times 10 days will not be enough, since we will only got 500. But we already see what are the engagement KPIs, whether these users monetize progress, the retention, there will be payers already, there should be actually payers already, not necessarily whales. So those payers, of course, they will see whether they buy the cheapest package of 1.99 or they buy more expensive package, so there are already methodologies of verification of this traffic, and they will be rather lucky, they will catch the whale and they would be very great, I would say, situation. However, this is not something we are looking for.

Shamanth: Right, so you’re basically saying, oh we have a payer profile, are they able to identify the payer profiles, and other users who have same profile, are they’re going to apply those for us, that’s what you’re looking for?

 Misha: Exactly.

Shamanth: Got it, got it, that makes complete sense. The other aspect that I’m very curious about is creatives, because again, programmatic is generally considered a black box with something like Facebook and Google with fairly extensive creative experimentation that can happen – how does creative on programmatic differ just because you don’t even have a dashboard in most cases to manage some of these things, how do you work with creatives, what’s the typical creative strategy like?

Misha: Right, so when we work with programmatic, we work with managed partners, so they manage our campaigns on our behalf, on their own dashboards. That means they also create their own creatives from our own assets. So the reason why they do it is because they buy in CPM and they want to make sure that cost per install would be minimized. So they would rather experiment with creatives on their side where they can iterate quickly and not wait for us to send them the new creatives. 

But to move forward,

the creatives is not number one reason that a campaign will be working or not, but the targeting is something that would define whether campaign and cooperation is working.

So once the right targeting is achieved, which is because of machine learning and algorithm targeting the right users, then the creatives should actually be showing the casino elements. So I mean, in coins, diamonds, slot machine, etc. Unfortunately, we haven’t seen that misleading creatives would work for us, which makes sense because the casino audience is very niche, let’s say, it’s a rather 40-plus users; and if they would see very misleading creative and even if they would download, they would be very confused as what this game is; they don’t really like changes that much, they would rather see the creative that would actually reflect slots games that they want to play. So if we did experiment with misleading creatives where we ended up attracting users that are not the right users and they would churn right away. And if we show the right creatives to the right users, that’s a way to work in this area.

Shamanth: Yeah, no, I think that makes sense in the context of creatives, just because I think it sounds like the most important thing in programmatic is for the DSP to identify a potential player, identify a potential installer, engager, if the creatives are secondary to that identification process it sounds like, yeah. And, you know, what, how, what would you say transparency means in the context of programmatic and why is it important?

Misha: Yeah, actually transparency is really important while working with DSPs. We haven’t realized this at the beginning, like few years ago, but now this is, this is must, and transparency should be on all different areas. So we need to see in the real time what is the publisher name, what is the exchange name, what are the creatives’ name, etc. There are a few reasons, let’s say, one of them is fraud. So there will be cases when fraud will be as well on programmatic available, and it’s not the DSP’s fault, or it’s not necessarily DSP’s fault. 

So we had one time situation that I would like to talk about, so and

one DSP partner who has purchased a few hundred installs from publisher, let’s call it ABC. The quality was great. However, the concern that I had was that this publisher was where a few friends of mine are working, doesn’t serve ads, and hey have a different model and so I reached out to them directly and I asked, hey guys, did you change the monetization model and start serving ads. The answer was no. 

So I brought it up to our DSP partner which understood the case and took it with exchange where the installs were purchased from, and then we got charged back for the amount of money that we spent on this publisher, which then was fraudulent.

It’s really important to have transparency when speaking about anti-fraud. Another one is negotiation power – we also work with some of the ad networks directly and two examples here could be named Unity and Applovin which you can purchase media from directly, but you also purchase through DSPs by the ad exchange. So knowing how much actually you spent in these specific media sources, direct versus indirect, is good information for your further negotiation power for different deals. And maybe, last but not least, it’s also important to spot different trends and verifying machine learning. So example could be if our user acquisition campaigns would start purchasing installs from games for kids, something is definitely wrong here, because majority of our users are 40-plus users, and I would check for sure. So it’s good to verify and double check whether this is good approach.

 Shamanth: Yeah, no, that makes complete sense. Out of curiosity as a marketer and a user acquisition manager, is there a way you educate yourself about: oh what’s happening with machine learning, how do I learn and figure out what’s happening, and what’s perhaps a technical subject?

Misha: Yeah, of course, of course, it’s not easy because usually the people we talk to from our programmatic partners are not the product people, so of course they know somewhat, but product people would be the one who would answer all of your questions. So I do ask them all the questions I have, and there is often a situation where they would suggest or I would suggest that we jump on the call or meet with their product people and I would continue to answer this question. So every once in a while, at least once a year, I try to meet with the product people from all partners at conferences, events or while just visiting the offices, and try to understand what new they are working on, how we can help feed in algorithm to help it buy in users for us in most effective way.

Shamanth: Right, right, it sounds like you go above and beyond, and it’s indeed, that indeed explains why you see the success you do, what typically has to happen before the team says, right, I’ve been buying on a DSP, we are doing a lot, let’s try to do all the programmatic buying in-house, what has to happen to have that happen?

Misha: That’s a good question actually. So every once in a while when you think about how much money you spent on programmatic and you know that part of that money, actually the fees that the company take for themselves, you realize, oh maybe it would be a great idea to build own bidder and bring programmatic in-house. It does sound like a good idea, and I’m sure a lot of companies have that in their mind. So a few things have to happen, first of all, the company has to be super BI savvy. 

So they need to have already, they need to know everything about the users and how to calculate the LTV, and get the user in the first one hour after acquiring a user as to whether they’ll become a purchaser. And second is the company should be able to allocate at least few full-time people, so few people in BI, and at least one person in UA for at least months or even years in this project as it will take time to build, to verify, to verify the model, etc. It also takes just so much time to connect all the ad exchanges, it could take months, the ad exchanges have to verify you and accept it’s not that easy. 

And company also needs to be aware that the investments will be at least a few million dollars, and for the period of at least probably years before you see any light at the end of the tunnel. So I know a few companies are thinking about it and trying, you know, maybe researching, however, I know one or I’ve heard about one that relatively succeeded and few that failed. And I do want to highlight here that, of course, it would be great to bring programmatic in-house, but it also would be great to maybe bring attribution in-house, and not pay the attribution companies fees, and maybe it would be great to, you know, take off the App Store and iTunes Store fees and not pay them as well. So I think we, as a mobile gaming companies, are good in what we do best, mobile games, and I’m happy Huuuge Games is focusing on doing more mobile games instead of maybe trying to bring something else in-house where we may fail, and that’s the reason why a lot of other companies succeed that actually just doing what they are doing best full time.

Shamanth: Indeed, indeed. Yeah, no that makes sense and that explains why, you know, building programmatic bidders isn’t a mobile gaming company’s core competency, explains why it’s so hard to build programmatic infrastructure, and indeed I think it also points to how much effort it takes to succeed with programmatic which is absolutely something you have done for the many years, for many years out. Misha, this was an honor to have you on this show. Thank you so much for being a guest on the Mobile User Acquisition show. Can you tell our listeners how they can find out more about you?

Misha: Sure. Well, thank you very much for inviting me. It was a pleasure. I am actually also following your podcast, so looking forward for this to be online. I’m very active on LinkedIn, so my LinkedIn is Misha Syrotiuk. You can find me, I’m sure with majority of people out here I already have few common connections. My email is misha@huuugegames.com, so that’s another way to connect with me. And I am trying to be a few times a year in industry back, so that’s an opportunity to meet face to face.

Shamanth: Absolutely, absolutely. And now that we’re in the same continent, definitely looking forward to meeting you at one of the events. I’m excited for that. Thank you so much for being on the show.

Misha: Thank you Shamanth.

A REQUEST BEFORE YOU GO

I have a very important favor to ask, which as those of you who know me know I don’t do often. If you get any pleasure or inspiration from this episode, could you PLEASE leave a review on your favorite podcasting platform – be it iTunes, Overcast, Spotify or wherever you get your podcast fix. This podcast is very much a labor of love – and each episode takes many many hours to put together. When you write a review, it will not only be a great deal of encouragement to us, but it will also support getting the word out about the Mobile User Acquisition Show.

Constructive criticism and suggestions for improvement are welcome, whether on podcasting platforms – or by email to shamanth at rocketshiphq.com. We read all reviews & I want to make this podcast better.

Thank you – and I look forward to seeing you with the next episode!