ABOUT JAYNE: LinkedIn  | Twitter | Electronic Arts




ABOUT ROCKETSHIP HQ: Website | LinkedIn  | Twitter | YouTube

KEY HIGHLIGHTS

📈 How to future-proof growth strategy

⚖️ The case for bringing the decision engine in-house

✅ We still have to be SKAdNetwork compliant

🏗️ How to build bidder as a service in-house

🤝 The shift of UA towards engagement and retention

👩‍🔬 The expertise necessary to build a decision engine

KEY QUOTES

The future lies in self-sufficiency

What if we brought the decision engine in-house? What if we didn’t have to share data with the likes of Facebook or others to create lookalike audiences and to be reliant on their algorithms? What if we could do that internally? Which isn’t impossible, but it does change the landscape of a marketing team internally from relying on others to do a lot of the algorithmic work and bringing that in-house.

SKAdNetwork is a part, not the whole.

There’s no doubt that we’re still driving towards being compliant with SKAN and we are buying SKAN and we are focused on partnering with our partners to get them compliant with SKAN.

But it’s a fraction of overall inventory. I can’t help but think, as any gaming company will look at, let’s say emerging markets, Asia, where iOS isn’t the majority, it’s not even a sliver of the total.

There are no shortcuts

The crawl-walk-run approach is you start with what you would consider linear optimization every day. You look at your inventory coming in and you cut the bottom 10% or basically the most inefficient sources that your campaigns are running on. And you continue on. Then you get more sophisticated where you actually start doing bid modifications.

The perceptible shift towards retention

We’re going to see is the deprecation of IDFA as a forcing function for a lot of gaming companies to become more sophisticated in their life cycle. And one of those aspects is bringing a decision engine in-house to help with promos, offers, where to send users and, and manage it in a way more sophisticated manner.

There are going to be challenges

Programmatic, for a lot of people, “failed”. Not because it wasn’t working overall, but it wasn’t as efficient in comparison to channels like Facebook. If you have channels that aren’t accessing as much data, you can’t leverage, you can’t send your own first party data as much as you did. The comparison and the performance is now becoming more, I wouldn’t know, I wouldn’t say at parity yet, but the gap between the performance of programmatic versus a performance of something like a Facebook, that gap is closing.

Start asking questions of your DSPs

If you’re already working with the DSP, start to engage with them. Not just as an inventory source in terms of throwing budget at them, but asking, “Hey, can you walk us through a little bit more on how your algorithms work?”

How to build bidder as a service on top of a DSP

Bid modifiers. Hey, do you allow bid modifiers? Can we upload our own bid modification sheets? Hey, do you allow us to host our own algorithms and then just leverage your tech essentially to access inventory? I mean, bring your algorithms is a generic kind of an umbrella term. Proprietary decisioning that the advertiser can bring to a DSP.

How to discover efficiencies

The fee that a DSP, that an SSP, has with the DSP and the exchange. So if you add up all of the SPO supply path optimization, so if you basically take inventory of all of the cuts along the path of how that one media dollar that you’re intending to spend on inventory gets shared. And then at the end result, how much of that actual original dollar is spent on media. And you try to rake that back. That can and should be factored into the efficiencies that you’re gaining with a decision.

In-house decision engines are a longterm investment

Decision engines are hard. It is more than a quarter endeavor or a two quarter endeavor. This takes over typically a fiscal year, if not more to, to get a V1 going. I would also suggest that you don’t just think about SKAN in the context of iOS, but think about all of the protocols that you might be engaging with in the next two to three years and where that point of diminishing returns is in terms of how focused you want to be on overly optimizing conversion values on one protocol.

FULL TRANSCRIPT BELOW

Shamanth: I am very excited to welcome Jayne Peressini to the Mobile User Acquisition Show. Jayne, welcome to the show.

Jayne: Thanks for having me.

Shamanth: Excellent. Jayne, let’s jump in. I’ve been looking forward to having you on the show because you’ve had some very strong opinions. And you’ve had a lot of clarity of thinking around how to approach what’s going to happen, in terms of the effective deprecation of the IDFA. So, you have a very unique perspective. We’re going to talk about how you’re thinking about potentially just sidestepping SKAdNetwork. And I’m excited to dive into all of that just because that’s something a lot of people take for granted and assume that that’s the only path.

Your team manages multiple titles, and at least a part of your approach involves completely bypassing SKAdNetwork. Tell us about this approach where you’re contemplating bypassing SKAdNetwork, and also tell us what inspired this particular approach.

Jayne: I wouldn’t say bypassing SKAdNetwork, I would say future-proofing protocols—future-proofing our strategy for multiple protocols. My concern is this could become a little bit of a nightmare for advertisers to manage if other app stores or inventory sources go that way.

Then, as an advertiser, it becomes a question of: is the juice worth the squeeze to even leverage the protocols if there are multiple? Even as we’re looking today on the demand side or advertiser side, we’re only seeing a fraction of SKAN compliant inventory.

So if we’re only seeing 13-15%, what’s next year going to look like? What’s the following year going to look like? When will the majority of the inventory that we’re buying SKAN compliant? Will it ever be the majority? And, I think if we think about it from that approach, then practically should we be overly concerned about conversion values and campaign ID numbers?

Or should we be thinking about:

what if we brought the decision engine in-house? What if we didn’t have to share data with the likes of Facebook or others to create lookalike audiences and to be reliant on their algorithms? What if we could do that internally? Which isn’t impossible, but it does change the landscape of a marketing team internally from relying on others to do a lot of the algorithmic work and bringing that in-house.

That does take technical capabilities, that takes data science, that takes additional servers, and the like. But, it makes sense for companies that spend a lot. Look at Zynga that just acquired ChartBoost. When I was at Machine Zone we had a DSP in-house. It used to be when advertisers flirted with this idea, because it’s not a novel idea to say: “Oh, we should have bidder as a service.”

There’s plenty of vendors out there that do bidders as a service, but it used to be that in comparison of your performance with the likes of Facebook it never really panned out. It was always second or third tier compared to Facebook or Google performance. But with the impact of the deprecation of IDFA, everyone’s CPIs are going up. I don’t know any person I’ve talked to to date that hasn’t been impacted with their campaigns. But that’s not really my concern right now.

The trends I’m seeing that are more concerning are the fact that Facebook, and the partners that we used to rely heavily on acquiring high value users, don’t have that data anymore. And they’re not driving as high a value of users. So, that would be more of my focus as an advertiser: bringing the decision engine in-house to acquire high value users where those cohorts, after the deprecation of IDFA, have gotten smaller and smaller, have contributed less revenue over time. And that’s more of a concern, right? And in the free to play space, your high value users, cohorts ended up contributing the majority of the lion’s share of your revenue.

Shamanth: Yeah. I like how you characterize this. The fact that there are very different protocols on different platforms. That’s a reality. That’s easy to lose sight of when the loudest voices just speak of the IDFA. Whereas, even if IDFA deprecation is a hundred percent complete, all of Android would not get coverage under that. Not to mention, if there’s another app store.

Jayne: Yeah. You have Unity and you have all these others.

Shamanth: And I think that’s a very important characterization. I think that makes so much sense. And, would you be looking to rely on SKAdNetwork for iOS traffic? Or would you supplement the workings of your decision engine that you spoke of? Which I’m sure we’ll dive into in more detail. So would you combine iOS optimization with SKAdNetwork plus the decision engine? How would that look?

Jayne:

There’s no doubt that we’re still driving towards being compliant with SKAN and we are buying SKAN and we are focused on partnering with our partners to get them compliant with SKAN. But it’s a fraction of overall inventory. I can’t help but think, as any gaming company will look at, let’s say emerging markets, Asia, where iOS isn’t the majority, it’s not even a sliver of the total.

Where is the point of diminishing returns in terms of being overly compliant and also building tooling or automation? And all of the tools that a lot of gaming companies or larger advertisers have built over time that were relying on IDFA, they’ve had to reverse engineer all of those toolings and all of those automated tools.

And it probably is very painful for them, but to do that over and over again, if multiple app stores do the same thing, I see this as a disruption to business. I absolutely want to be privacy compliant, but I think it doesn’t take much to Google or to search the news right now, in terms of Apple’s decision to hire a pretty notorious ad tech person to lead their ad platform for a short period of time.

So, it makes me wonder what the motives are, I think, of moving towards SKAN. And, user privacy should be taken seriously, it should not be used as a pawn to one up another business partner, in my perspective. That’s not how we should be doing business. We should, if we’re focused on user privacy, we focus on user privacy, but it’s not a pawn or a bargaining chip in the market.

Shamanth: Yeah, certainly. And if we look at everything that Apple is doing, it’s clear that privacy is a facade for a strategic decision to move into advertising. And as you pointed out from their recent hiring decisions, that makes sense. And, what I’m also hearing you say is, as a marketer, not relying completely on the protocol that is proposed by Apple is a way of future-proofing yourself.

Jayne: Yeah. I don’t think a hundred percent relying on that protocol or that being the backbone of your business from a UA perspective is strategic in the long term.

Shamanth: Okay. Certainly. And you spoke of building a decision engine. It’s my understanding that it’s going to be your strategic focus to make sure there isn’t as much reliance on SKAdNetwork. Tell us more about what this looks like. What is the decision engine and what does that look like?

Jayne: Sure. Again, this is not novel. I’m not the smartest one in the room. But others have gone this route, too. Some have been successful. Some have not. Most decision engines typically start with something simple, in terms of either white label a bidder as a service, and getting access to programmatic channels. Or they develop on top of APIs to buy programmatically. The engine itself can go a few ways similar to how life-cycle marketing has gone into these phases.

The crawl-walk-run approach is you start with what you would consider linear optimization every day. You look at your inventory coming in and you cut the bottom 10% or basically the most inefficient sources that your campaigns are running on. And you continue on. Then you get more sophisticated where you actually start doing bid modifications.

So, depending on the characteristics of big calls in the inventory that you’re looking at, you can go up, go down on your bids. If it’s a high end device, if there’s something about that call that you consider an indicator of a high value or just even a high propensity to convert a player, then use bid modifiers.

And then after that, then you get into actually sophisticated algorithmic decisioning. So that kind of crawl-walk-run approach is typically how it goes. Where you start with linear optimization, it doesn’t take a lot of tech or anything like that. Usually it’s in Excel. And then you go to bid modification, which can be in Google sheets and Excel to some degree and you house it somewhere like an S3 bucket or something.

And then you go to full-blown building algorithms on top of these bidders as a service; or if you partner with the DSP that can allow you to bring-your-own algorithms. That can also be something you can do.

Shamanth: Yeah. So from what you’re saying, this could become a more sophisticated version of something you could basically execute with spreadsheets. And what you’re also saying is, look, there are contextual variables that could be very powerful in determining what could be valuable users. And, would not a lot of this be applicable primarily in programmatic? Or would you say that some of this could also be applicable, let’s just say, on owned and operated inventory?

Jayne: Oh yeah, absolutely. You look at the acquisition of Zynga and Chartboost. Zynga has a massive player network. I would assume that they are going to be way more sophisticated with up-selling and cross-selling across their own network with the acquisition of ChartBoost.

I mean, it’s now a race for larger gaming companies that have a pretty diverse portfolio and a sizable audience: to think of how to better manage your own players, rather than just being a hundred percent reliant on only acquiring new players. But it is a shift. It’s the evolution of where we’ve come from in UA and growth. It started with pretty basic stuff: hit a certain install goal and, that’s about it; to ROAS and back-end, and deeper metrics; to revenue, velocity, and all these things. And now it’s actually quite expensive too, to retain users. So what are we doing to retain and also grow those retained users, as a contribution of our overall revenue? And I think that’s what

we’re going to see is the deprecation of IDFA as a forcing function for a lot of gaming companies to become more sophisticated in their life cycle. And one of those aspects is bringing a decision engine in-house to help with promos, offers, where to send users and, and manage it in a way more sophisticated manner.

Shamanth: Yeah. And would it be accurate to say that for platforms like, let’s just say Facebook or Snap that rely on SKAdNetwork for optimization—if you’re saying cost per add to cart or cost per purchase, that’s what still lean on SKAdNetwork? Or is that not quite the case?

Jayne: I can’t tell you exactly what’s going on internally at Facebook in terms of what their plans are. But a majority, let’s say 40-50%, of most UA budgets goes to Facebook. That’s pretty consistent, no matter who you talk to. It’s not because we all love Instagram and things like that. It’s because it performs. And similarly on the monetization side, people have FAN implemented as a publisher on the monetization side, because it performs, it drives high CPMs.

Well, on the advertisers’ side, we get higher value users. When we leverage Facebook and if the lookalike audience is not performing on Facebook we have a problem, because you’re talking about half of your budget going to one partner and the lion’s share of that budget that is going to that partner going to modeling for lookalikes.

As an advertiser, that might be something you’re going to have to do yourself if Facebook doesn’t have that data anymore.

Shamanth: So a couple of years ago a number of folks tried to in-house programmatic and it just wasn’t very successful at the time.

What are some of the key factors that would contribute to this having a better shot at success going forward?

Jayne: I think it’s really just a comparison of performance. I mean,

programmatic, for a lot of people, “failed”. Not because it wasn’t working overall, but it wasn’t as efficient in comparison to channels like Facebook. If you have channels that aren’t accessing as much data, you can’t leverage, you can’t send your own first party data as much as you did. The comparison and the performance is now becoming more, I wouldn’t know, I wouldn’t say at parity yet, but the gap between the performance of programmatic versus a performance of something like a Facebook, that gap is closing.

And then it becomes more interesting to say, “Hmm. Well, it’s not like now we have to make a big leap to impact performance as much as we just have to do incremental steps to get to a place where we can use it to scale.”

Shamanth: And for a company that’s exploring building a decision engine of this sort, what are typically good starting points? Are there off-the-shelf solutions, or would it be a sort of spreadsheet-based modeling that you described? What sort of resourcing should they plan for? What do you think would be good starting points for somebody that’s looking to go down this path?

Jayne: I think the first thing is get super close with whatever DSP you’re currently working with. Ask them first: “Do you all allow bring-your-own algorithms?” A lot of mobile specific DSPs don’t. And I can’t tell you why, because programmatic inherently is not supposed to be black box, but here we are. And, if in the off chance that your mobile DSP or a DSP that you’re working with does allow for bring-your-own algorithms or is open to it, I would heavily rely on them to help partner and consult for your internal team on how to develop your own algorithms.

If you aren’t currently working with a DSP, you can look at bidder as a service. It does behoove you though, to leverage a consultant or someone that has ad tech and programmatic experience because it’s sophisticated tech, and it’s not just a typical UA buyer that can manage this himself or herself or themselves.

This requires coordination on the data science side, on your infrastructure, on the media side. So it’s a beast. This is development work. This is mathematics. This is buying. So

if you’re already working with the DSP, start to engage with them. Not just as an inventory source in terms of throwing budget at them, but asking, “Hey, can you walk us through a little bit more on how your algorithms work?”

And if they’re not open to that, I would start asking why they aren’t open to showing their algorithms. Most should be able to give a white paper, at least, of how their algorithms work. And if they aren’t willing, then you should be very concerned about why you’re working with them.

And then also ask, “Do you allow bidder as a service? Or do you allow bring-your-own algorithm?” Yes/no. And then kind of go down that route. Okay. Well, we’re going to start looking for a DSP that will, and we’re going to partner with them.

Shamanth: Sure. And when you say, bring-your-own algorithm, you’re basically saying…?

Jayne: Being able to use anything.

Bid modifiers. Hey, do you allow bid modifiers? Can we upload our own bid modification sheets? Hey, do you allow us to host our own algorithms and then just leverage your tech essentially to access inventory? I mean, bring your algorithms is a generic kind of an umbrella term. Proprietary decisioning that the advertiser can bring to a DSP.

Shamanth: And that decisioning would involve contextual variables like you described earlier. Got it.

Jayne: A lot of it too is taking an inventory. What apps drive the majority of your traffic, and going down the route of asking “can we access them programmatically? What kind of apps are like those that we can create?” Yes. Some bid modifier modifications for there’s a lot of uses for programmatic in general. It’s also not just RTB programmatic. It just negates the need really to engage with it with people over email as well.

Shamanth: At what level of scale would it make sense for a developer to start exploring some of these solutions? Because certainly as you said, it’s very development intensive, very engineering intensive to be able to, or data science intensive. What level of scale would it make sense for somebody to look at this?

Jayne: I would say over about $50 to $70 million annual spend is something that would make sense. It would at least make sense to flirt with. Anything over a hundred million, you should absolutely be doing that.

Shamanth: Certainly, certainly. And at that level of scale, I imagine you’re paying significant amounts to the DSPs and the intermediaries at that point of time, so the savings can fairly easily be justified at that level of spend against, I would imagine.

Jayne: Yeah. And I was just only explaining the efficiencies on the fee by: there’s middlemen fees, right?

The fee that a DSP, that an SSP, has with the DSP and the exchange. So if you add up all of the SPO supply path optimization, so if you basically take inventory of all of the cuts along the path of how that one media dollar that you’re intending to spend on inventory gets shared. And then at the end result, how much of that actual original dollar is spent on media. And you try to rake that back. That can and should be factored into the efficiencies that you’re gaining with a decision.

Shamanth: Definitely, certainly. And not to mention, as you said, the optimization is so much stronger, so much more powerful when you own it.

Jayne: Right. Yeah. 

Shamanth: Wonderful, Jayne. This certainly is a lot of food for thought. A very unique and interesting perspective. As you pointed out, this may not have worked very well a couple of years ago, just because the proprietary algorithms platforms were so strong. But perhaps this is a time when this sort of strategy of owning your own decisioning can work. Certainly this has been very, very instructive and this is perhaps a good place for us to start to wrap Jayne. Any closing thoughts, any other things folks should keep in mind as they think through the way forward?

Jayne: I would say don’t give up on it.

Decision engines are hard. It is more than a quarter endeavor or a two quarter endeavor. This takes over typically a fiscal year, if not more to, to get a V1 going. I would also suggest that you don’t just think about SKAN in the context of iOS, but think about all of the protocols that you might be engaging with in the next two to three years and where that point of diminishing returns is in terms of how focused you want to be on overly optimizing conversion values on one protocol.

When, if there’s three or four coming down the pike as well.

Shamanth: Indeed. Yeah. And as you said this lets you plan, not just for SKAd, not just for IDFAs effective deprecation, but for your longer term future. Future proofing, as you said. Jayne, thank you so much for being on the Mobile User Acquisition Show.

Before we wrap, can you tell folks how they can find out more about you?

Jayne: Of course I’m very present on LinkedIn, so you can always set up time with me on my Calendly links. I’m an open book, so reach out to me. I’m Jayne for Twitter and if you want to get a hold of me, I’m sure you’ll be able to. I’m very SEO friendly as well. So I’d love to hear from anyone.

Shamanth: Wonderful. And we will certainly link to your socials from the show notes for folks to find you just as well. Wonderful. Jayne, thank you so much for being on the show. Thank you.

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!