My guest for Episode 6 is Edik Mitelman, who is the General Manager for Privacy Cloud at AppsFlyer. In this episode, we not only delve into the intricacies of data clean rooms but also explore the mobile ads ecosystem and demystify the concept of a mobile measurement partner (something I could relate a lot with given similar measurement challenges in the retail media). This episode is the last part in the series on data clean rooms. The contents are available both as a podcast, and in a condensed and lightly edited written format below.
In this episode, we discuss:
(01:43) Edik's key career learnings
(05:17) AppsFlyer's genesis as mobile measurement partner (MMP), and the difference between web and mobile ads ecosystem
(10:21) Evolution of AppsFlyer
(13:19) AppsFlyer’s expansion from MMP into data clean rooms
(20:27) AppsFlyer's USP in data clean rooms, especially for retail media
(25:19) View on clean room for retail media
(26:55) Cross-channel attribution for retail offsite media and clean room interoperability
(31:25) Cloud interoperability
(32:19) Adoption challenges and what's next for AppsFlyer
____________SLIGHTLY EDITED Q&A VERSION BELOW__________
Q: Can you tell us a bit about AppsFlyer’s origins?
Edik:
AppsFlyer started in 2010. Considering startups and scale ups, this is a mature, old company. Essentially, the genesis point for founders Oren and Reshef, who are still our CEO and our CTO, was the Rise in the Apple app store and specifically, I mean the iPhone creation, but the app store that came with it. Mobile advertising was still very nascent and nobody understood what happens. And then Oren and Reshef realised that in this new ecosystem, there is nobody representing the advertisers. There were many startups who came in and started selling or providing value into the app developers, and into the ad networks (e.g., SSPs) but nobody took care of the actual advertiser.
How can advertisers know that they're spending their money properly? How can they know that they are working with the best partners? How easy is it to work with partners? Because back then if you wanted to work with a network, you had to install their SDK into your application. SDKs are heavy, SDKs are hard, it's a black box.
So if you want to work with 10 different networks, you need 10 different SDKs. That becomes unscalable and that is very hard on the tech part. But also measurement was the privilege of the networks. And this was and still is in some cases, to let the students grade their own homework, or to use a more common metaphor, to let the cat guard the milk.
So Oren and Reshev realised that there has to be a neutral, trusted, unbiased player that is in the middle and really represents the advertisers and works with these partners to really show the true, single source of truth picture of how mobile advertisement works. So that every player in this ecosystem can make the best decisions, the smartest decisions to grow, to increase their business and to provide the best user experience to their customers. So that's how the the company started.
Today there is a category for such services and it is called Mobile Measurement Partner or Mobile Measurement Platform (MMP). And you need to actually be certified to do that. So this term was actually coined by the networks themselves. So Facebook or Meta, they have their own MMP program, and then others have, and this is why they invented this term, and it transferred to become a category. But there are only a few companies in the world who are certified measurement partners for these large ad networks. And of course, all the smaller ones are, are working with them.
And not every startup can just come up and become an MMP. It's not how it works. You need to go through a long, hard process. For the last decade plus, we've been building this trust. We've been building this ecosystem. We've been building this machine that performs measurement and attribution and adds a lot of value on top of it.
Q: What are the difficulties with mobile measurement?
Edik:
On the web, it's pretty easy to ‘attribute’ because there are cookies and you can just add stuff to the URL and then if a visitor comes to your website, you can just look at the URL with UTM parameters or look at the cookie and you know where they came from. On mobile, just the existence of the app stores, be it Apple or Google, created a cut. So somebody who clicks on an advertisement in a mobile environment, let's say on Facebook, they are redirected to the App Store [to download the advertiser’s app]. And once they're redirected to the App Store, this data is a property of Apple or of Google, it's not your website. And then somebody clicks download the app, and this might happen two days after. And then it's not enough to download the app, they need to actually click on it, to tap on it, to open it. And then you're asking, okay, I see a new install here. Where did it come from? Is it organic? Is it from channel A? Is it from channel B? Is it from campaign A or from campaign B? So the existence of app stores made mobile attribution difficult.
Q: How big is AppsFlyer today? And how has it evolved in last 13 years?
Edik:
AppsFlyer is a very big company today. We are more than 1500 employees. We have about 70% market share in MMP in some markets. We are market leaders globally. We have offices in more than 20 countries.
Today the MMP solution seems obvious, but it wasn't obvious at all for many years and it's still a very hard problem. We're still constantly tuning, updating, improving our algorithms, and our machine learning to make sure that we really provide the best trusted, unbiased measurement.
And a lot of value on top of it - fraud protection, ROI calculations, analytics, creative analytics etc - so that advertisers can look at the entire picture and make the best decisions. Really, we are their growth partners.
Of course, helping our partners understand the source of app installs is just the foot in the door. This is not enough. Essentially, nobody does user acquisition just to bring installs. You need to see LTV lifetime value. You need to see return on ad spend. You need to calculate how much was your cost of acquisition, and you need to make sure that you're optimising for that growth.
Eventually the very easy formula is lifetime value minus cost of acquisition. This is average revenue per user. And you want that to grow because this is how you make money. So we are getting our customers send us their in app events and they send us a lot of their different signals, whether it's from our SDK or from server to server integrations. And we are integrated with hundreds and thousands of different partners, be it ad networks or tech partners or the entire ecosystem. We have a huge marketplace, probably one of the largest B2B marketplaces in the world.
And the customers are able to track and to monitor and to measure the entire user flow. It's not just the install. It's what the users did later. What did they buy? What did they do? It's their users, it's their first party data, So it's a full fledged attribution analytics platform.
And for the past few years, we're not mobile anymore in the sense that we're doing full cross platform measurement. Connected TV is now very hot. We are doing PC and consoles. We are doing any touchpoint that you have. We provide you the opportunity to add it to your media mix, to measure, to analyse, to get insights off, and to make the next decision about your next campaign.
Q: What prompted AppsFlyer to build a privacy cloud / data clean room? And can you briefly explain your offerings in this space including your recent marketplace launch?
Edik:
AppsFlyer always took privacy very seriously. We always took a very clear side of being neutral. And being trusted and unbiased. And I think these values are really what brought us to the market leadership position.
And we treat them religiously. So, there are some companies in the world, some of our competitors, some companies that are a mobile measurement partner, but also an ad network. Or a mobile measurement partner and also have applications. We don't do that. And believe me, we had a lot of business opportunities to do many such things. It's very lucrative in terms of the potential revenue, but this is an absolute no go for us. So we never commingled data of different advertisers, even before privacy was a thing. And just think about it - we see what happens in every single app in the world almost. So we could easily generate valuable insights and tell gaming app Y a lot about gaming app X and Z so that they can do something about it in their own game.
We will never, we would never, didn't ever, and won't ever do that. So privacy was always at the core of AppsFlyer products and values. But then 2.5 years ago, the entire industry shifted. Of course, Apple led with ATT, with the prompts that the mobile identifiers now need an opt in consent instead of an opt out consent. Not to mention GDPR in Europe and CCPA in the United States and COPPA. So, a lot of regulators started playing this game, but also what is more important is the actual platforms and the operating systems took on themselves the role of the regulator.
And then after Apple introduced iOS 14. 5, which brought these changes, Meta came in and made a very brave and amazing decision. They said no more user level data. The entire industry, be it mobile or web, is addicted to a drug that is called user level data. And this drug exists since the dawn of the internet and everybody built their systems for decades, starting ground up from the user level data. But the paradox is that nobody needs user level data for the value. No company targets Keshav or Edik. All the targeting online or digital is based on segments and cohorts of people like Keshav, people like Edik. The output is always groups of people. But it was so easy and accessible to have user level data, that people were just building their audiences off it.
But then Meta came in and everybody else started to join the party and said no user level data access anymore - just segments, cohorts, and groups. And this was the catalyser for us. When we started the privacy cloud and the data clean room, in December 21, we looked at the different solutions in the market and we realised that we can do much better.
We realise that the world is going to be very private. But people think till today that privacy and utility or privacy and user experience are ‘either or’ and we believe and we prove it every day that privacy and utility and user experience are ‘and’. You can maximise privacy And optimise user value and user experience. It's not a contradiction. Building data collaboration solutions that are privacy by design enables you to really be super private, and keep the same value. It allows you to make the same decisions you were used to when you were built on top of fully accessible user level data. So that’s what led us to this vision. Based on that, my organisation was created, and we really started building the products and the product suite has the data cleanroom at its core.
But data cleanroom is yet another piece of technology. What matters is what do you do with it. So we are building solutions and products that are based on a cleanroom for private measurement. How can you still perform measurement and attribution and do all that when you don't have identifiers? When you cannot just simply take IP addresses or email addresses from the world and do matching? How people can share their first party data, which they have consent for? They collected it religiously from their users, but how they can leverage this data without sending it to a third party? How can you pick a startup that claims to do the best predictive algorithm in the world when you can’t even send them the data to the start-up to show you a demo?
So this is where our marketplace, and this is where our privacy cloud comes in. We believe that the world is already there and with the advances of AI and with the advances of the technology goes to a place where if previously you would bring your data to the logic, these days you bring your logic to the data.
So the data is key. The data is what's important. This is sacred. It should never move. It should be never copied. And utility still can be derived from it. Because if the logic developers or your partners come into the same safe, private environment - let's call it a data clean room because this is what it is - then they can still run the best state of the art AI, machine learning algorithms or what have you, on different data sets (first party, second party, third party) and give you the value you need. But you are not afraid or not concerned or not thinking twice because you know that what is yours is yours and is fully protected both in terms of the government regulators and in terms of the platform regulators.
Q: What are AppsFlyer’s points of difference as a data clean room provider?
Edik:
Technologically speaking, all of us are different. But from a user perspective, I think it's too geeky. And may require a PhD person to understand whether the anonymity algorithm from this vendor is more private than this differential privacy algorithm from that vendor. So when vendors try to compare on feature parity or head to head parity, I try to discourage them from that. Because privacy tech is algorithms, proved by the academia, developed by the smartest people in the world, and they work. All of us are audited constantly by neutral third parties. It works.
But the differentiation comes from what do you do with it? We bring the demand. We already have data and a lot of it because one of our most unique assets is that we have an SDK in the vast majority of the mobile apps in the world. So, in app events, first party data, CRM data, we already have the data, and that's because we have the trust. You want to work with a vendor in the privacy space that you trust and we already bring trust because all of these advertisers already trust us for years and decades with their most precious data and all of their identifiers and all of their in-app events. So if a retailer comes to us, it's a turnkey solution. We don't need to wait while we do business development or sales into their target audience. You can test it. You can demo it. You can do even a trial for free, with no commercials involved. Let's just see what is the match rate; you click a button and you immediately get to see match rates, identity resolution, how it works, and you can decide whether this retailer is worthwhile or not worthwhile for you to engage with. And then commercials, and then pricing, and then all the heavy lifting.
That is a huge differentiator. Onboarding to the AppsFlyer’s Privacy Cloud takes days, instead of months. It is a very low lift on the company side. Many companies in the industry today are concerned that to install a cleanroom solution will take a lot of resources from them. That they need engineering, they need BI people, and it's hard to justify because these people are busy doing something else. In our case, we do all the heavy lifting for them, and its really minimum effort in terms of resource investment from the customer side, and they can start seeing value from day one. If they deem the value enough, then let's talk commercials.
Of course, the sweetest spot is those retailers who are working with AppsFlyer for years now as their MMP on the user acquisition side. So now we come to their monetisation side and say, ‘Hey guys, we already work with you. We have the paperwork. We already did all the loops and hoops of legal. Let's make our partnership deeper and you use the same product, the same ecosystem, the same privacy cloud, both for your user acquisition and for your monetisation. This is definitely something that others don't have because they're not MMPs and hence they don't have a relationship with these people elsewhere. And we do.
We are also cloud agnostic. We won’t say we will work only if both sides are on Amazon or both sides are on Snowflake. I don't care. We are a standalone privacy cloud. We're a standalone cleanroom. We don't copy your data. We don't move it. And we're not relying on you having a certain infrastructure. So a brand can be on Google Cloud and the retailer can be on Snowflake. And for us, it's the same. And from a user experience perspective, for you, it looks like you're querying one table.
And we even added chatgpt on that now, so you write in English, it translates it to SQL, SQL runs, you think that it's one table, but really it's not just one table, it's different clouds.
Q: How do you see the retail media space as a data clean room provider?
Edik:
AppsFlyer’s core business is measurement. So the first product we had and still have is a data cleanroom for measurement purposes, where many of the ad networks restricted access to their data to the advertisers, but we still have it. And for us to have it, we need to ensure that we meet their contracts and meet their privacy standards, right? So the first use case we solved for [through data clean room] was measurement. Whereas the entire industry of cleanrooms went to solve the activation piece, because it doesn't require a prerequisite. To solve the measurement and attribution piece, you need to have a measurement machine first. And then you add the privacy on top of it. But if you don't have measurement data coming from the social networks or from iron sources of the world, there's nothing for you to protect. So we started from that use case, we built it, we gained a lot of traction, a lot of great positive response from the user base, we have public case studies, I welcome listeners who are interested to learn more, go to our website and, and, and read about it or reach out to me directly.
Then we expanded — it's not a pivot — it's an expansion to the activation use case And specifically in the retail media domain, because activation is existing in mobile as well. You activate audiences on Facebook, right? You activate audiences on Uber. You activate audiences everywhere. So activation in retail is just one use case. And yes, we are now focusing on it, uh, just like others, because the retailers are now very popular channels. It's what our customers want. So this is how we evolved opposite some of our competitors who naturally started to help a retailer and brand put their data together for activation.
Q: Given the trust you have from big ad networks like Meta, you get back disaggregated metrics from those ad networks. When brands use your data clean room, can you provide them de-duped attribution across ad networks?
Edik:
This is what AppFlyer is doing since 2010. Let alone privacy for a second.
You run a campaign on a plethora of different channels, on a plethora of different platforms and then you need to comprehend all of that and to have one place where you see your entire marketing performance and see all the channels and compare and make decisions. This is what this company is solving very successfully for more than a decade now. So what you just described, this need did not appear because of privacy. With privacy, it became harder to do. But it's still the same need.
Having multiple cleanrooms is breaking it. This is why I'm advocating for cleanroom interoperability, for example, through my work at the IAB. I believe that if we as cleanroom providers will not find ways to interoperate and collaborate, we will all lose instead of opening this market. Because today, every cleanroom, by nature, creates a walled garden ecosystem. If a brand has a cleanroom A, the retailer has to have the same cleanroom.
And this goes back to the problem I described in 2010 when Oren and Reshev created AppSlayer. If you wanted to advertise, you had to install 15 different SDKs. Somebody had to create one! And suddenly it became easier. We are getting to that same position with the cleanrooms in 2024 or 2025, where if a retailer who has enough data gravity, like the huge ones, the Amazons, the Walmarts of the world, they can dictate technology. They choose a vendor for whatever reason, and then whoever wants to work with them must choose the same vendor. This is more and more walled gardens. And we believe in open ecosystem. Our marketplace and everything we do is opposing walled gardens.
But until it's not there, you at least, need to have your entire funnel [from planning to activation to measurement] closed in the same cleanroom. So in the AppsFlyer Privacy Cloud, you really close the loop. You build an audience, Based on your data or the third party data or together, you activate it, then you measure it, then you enrich it with other data points, you combine it with all of your other channels (Google, Facebook, Web, CTV, Roku, what have you, it doesn't matter). You look at the entire picture, you make decisions, rinse, repeat. Having different clean rooms for different use cases is contradicting the entire point of a cleanroom which gives you your entire marketing picture in one place and in a single source of truth.
Q: From your early conversations, is the industry headed towards clean room interoperability?
Edik:
I really hope. As part of the IAB tech lab, we released in February or March, the first cleanroom standards paper. It was written collaboratively by pretty much every clean room company out there.
And this is a very big step. Again, not saying that everything that IAB releases is being adopted, but this is an attempt. We believe as part of the industry leadership, that this is required. We also released the OPGA, which is an attempt to have an open protocol for joining audiences. So this is for activation. We are now working, led by me representing AppsFlyer, on something similar on the measurement side. So this is one effort.
Another effort is between the clean room companies themselves to see whether interoperability of two different cleanrooms can generate more revenue. I really believe that there is a business model where both of the collaborating clean room companies generate actually more. Yes, it needs to be proven, but it's definitely something I'm pushing for.
And thanks to the cloud platforms, they are also a very important player in promoting an open ecosystem. They provide the infrastructure that enables different clean rooms built on top of it to inter-operate.
Q: What are the main adoption challenges you encounter when trying to sign up clients to your data clean room?
Edik:
The entire industry of data collaboration is still in the education / early adopters phase. So if we are looking at the famous crossing the chasm paradigm, we still did not cross the chasm. And this is as an industry, not just AppsFlyer.
Some people might still think that privacy is not something that is here to stay. Some people are trying to push their luck and to play till the very last day with user level data and with exposing PII until it will be like the really last moment. So there's many such cases where we are jointly, as an industry, interested. And thank you for your great podcast. I think it's also a very heavy help in that. The more content we have, the more education we have, the more it helps. People need to understand that the world changed. Not is changing, not will change. It already changed. Some verticals like mobile already felt it hard. some channels like the web are already bracing themselves. And this is the challenge for me and for my friends in, in the different other vendors is the market education. It's crossing the chasm.
Share this post