Adam Smart (Director of Product at AppsFlyer) and Piyush Mishra (Lead Growth Marketing at Product Madness) receive Alexandre Noirot-Cosson (Head or Marketing - Live Games at SocialPoint) and Norberto Degara (Head of Marketing Analytics at SocialPoint) to discuss the latest on media buying, automation, and the state of the demand side adtech ecosystem in 2021.
A DSP is a combination of several pieces of tech: the bidder, the ad-serving mechanism and the different algorithms that enable you to decide which ad to serve and which impression to actually buy.
Transparency and control is why SocialPoint wanted to build something in-house:
- Transparency: you can see which bids you were offered, which ones you successfully purchased, which ones you made an offer on but didn’t win. You also know exactly what you buy and where you buy it: there is no black box like there can be with ad-side DSPs.
- Control: you decide what you buy and at which price you buy it. Controlling the algorithm and the decision making means you control the type of users you want to buy and for which purpose in the game.
SocialPoint hasn’t built an in-house bidder, they use a bidder as a service technology. Then they built their algorithm on top of that.
Building a bidder in-house is extremely difficult: the technical infrastructure that you need is very high. On one side, a bidder works in real-time which requires a huge tech stack, and on the other side you need to have and maintain all the connections/API with the ad exchanges. SocialPoint decided to instead focus on the algorithm part.
When you build in-house you get bid requests, win logs and lost logs. So if you’re in control of the algorithm, you can enhance your bidding strategy to acquire users that you want at the lowest possible cost and find additional pockets of traffic.
There are 2 types of buys in programmatic: user-based educated buys and contextual buys. If you know something about the user and that he is worth spending for, the CPI doesn’t really matter.
In-house DSPs can’t compete with large managed DSPs on Android, or can’t compete with Google and Facebook’s algorithm and their ability to leverage their data to create lookalikes (including competitors’ data). However, when moving to contextual buys (e.g. due to ATT), it levels up the playing field which allows in-house bidders like SocialPoint’s to compete.
When you start working with an in-house bidder, you have to think about your long-term strategy.
- If you want immediate return on investment, use whitelists and what you already know. You cut DSP fees which gives you room for higher CPM buys.
- If you’re going for the long term, you need to buy every single demographic and start looking for patterns. This can be extremely costly because you need a lot of data if you look at things with a lot of granularity (combinations of OS, zip code, creative type, etc.).
Before iOS 14.5, managed DSPs were collecting IDFAs and able to use them across the different companies advertising. With contextual bidding, this advantage has been reduced a lot (even though they might still be able to find correlation at the publisher level).
Not everything is driven by algorithms. The UA team still has a major impact on how the campaigns are progressing in terms of performance, for example on the creative side.
There are no more blockers in terms of potential performance between using a SaaS bidder and a managed platform. Before ATT, SocialPoint ran blind tests against DSPs for LAT US campaigns (without device graphs, only contextual, same constraints, same creatives, same support) and monitored and shared performance across all partners.
Before, data was king and could make or break the success of a DSP. Now, data science and algorithms' cleverness and complexity are what will make the difference in the new ATT landscape.
By analyzing the bids you lost, you can run an experiment where you bid at a rate you would typically not consider in order to find out by how much the quality of the user could increase. SocialPoint has been surprised by bidding at crazy CPMs, getting very high CPIs yet seeing increased conversion rates in-game.
Before, with device graphs, DSPs knew where the high value users were and you would end up not getting access to that inventory through contextual buys. With ATT, those whales are pushed back to the pool that’s available if you’re working with contextual advertising.
When it comes to bidding, using the conversion value is a challenge. Not only because of its value, but also for budget control: DSPs usually have a very accurate budget control/pacing and the 48h delay that comes with SKAN prevents you from reacting quickly.
If everything is contextual, the benefit of having several DSPs bidding on the same SSP is much lower than when each DSP had device graph data they could leverage. So the question of incrementality from your DSP partners becomes even more relevant (i.e. Is it worth adding a new partner?).
Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.
In-house data is immensely valuable because you can get both pre-install and post-install data in order to have a feedback loop and identify which creatives drive retention and activity.
A DSP is a combination of several pieces of tech: the bidder, the ad-serving mechanism and the different algorithms that enable you to decide which ad to serve and which impression to actually buy.
Transparency and control is why SocialPoint wanted to build something in-house:
- Transparency: you can see which bids you were offered, which ones you successfully purchased, which ones you made an offer on but didn’t win. You also know exactly what you buy and where you buy it: there is no black box like there can be with ad-side DSPs.
- Control: you decide what you buy and at which price you buy it. Controlling the algorithm and the decision making means you control the type of users you want to buy and for which purpose in the game.
SocialPoint hasn’t built an in-house bidder, they use a bidder as a service technology. Then they built their algorithm on top of that.
Building a bidder in-house is extremely difficult: the technical infrastructure that you need is very high. On one side, a bidder works in real-time which requires a huge tech stack, and on the other side you need to have and maintain all the connections/API with the ad exchanges. SocialPoint decided to instead focus on the algorithm part.
When you build in-house you get bid requests, win logs and lost logs. So if you’re in control of the algorithm, you can enhance your bidding strategy to acquire users that you want at the lowest possible cost and find additional pockets of traffic.
There are 2 types of buys in programmatic: user-based educated buys and contextual buys. If you know something about the user and that he is worth spending for, the CPI doesn’t really matter.
In-house DSPs can’t compete with large managed DSPs on Android, or can’t compete with Google and Facebook’s algorithm and their ability to leverage their data to create lookalikes (including competitors’ data). However, when moving to contextual buys (e.g. due to ATT), it levels up the playing field which allows in-house bidders like SocialPoint’s to compete.
When you start working with an in-house bidder, you have to think about your long-term strategy.
- If you want immediate return on investment, use whitelists and what you already know. You cut DSP fees which gives you room for higher CPM buys.
- If you’re going for the long term, you need to buy every single demographic and start looking for patterns. This can be extremely costly because you need a lot of data if you look at things with a lot of granularity (combinations of OS, zip code, creative type, etc.).
Before iOS 14.5, managed DSPs were collecting IDFAs and able to use them across the different companies advertising. With contextual bidding, this advantage has been reduced a lot (even though they might still be able to find correlation at the publisher level).
Not everything is driven by algorithms. The UA team still has a major impact on how the campaigns are progressing in terms of performance, for example on the creative side.
There are no more blockers in terms of potential performance between using a SaaS bidder and a managed platform. Before ATT, SocialPoint ran blind tests against DSPs for LAT US campaigns (without device graphs, only contextual, same constraints, same creatives, same support) and monitored and shared performance across all partners.
Before, data was king and could make or break the success of a DSP. Now, data science and algorithms' cleverness and complexity are what will make the difference in the new ATT landscape.
By analyzing the bids you lost, you can run an experiment where you bid at a rate you would typically not consider in order to find out by how much the quality of the user could increase. SocialPoint has been surprised by bidding at crazy CPMs, getting very high CPIs yet seeing increased conversion rates in-game.
Before, with device graphs, DSPs knew where the high value users were and you would end up not getting access to that inventory through contextual buys. With ATT, those whales are pushed back to the pool that’s available if you’re working with contextual advertising.
When it comes to bidding, using the conversion value is a challenge. Not only because of its value, but also for budget control: DSPs usually have a very accurate budget control/pacing and the 48h delay that comes with SKAN prevents you from reacting quickly.
If everything is contextual, the benefit of having several DSPs bidding on the same SSP is much lower than when each DSP had device graph data they could leverage. So the question of incrementality from your DSP partners becomes even more relevant (i.e. Is it worth adding a new partner?).
Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.
In-house data is immensely valuable because you can get both pre-install and post-install data in order to have a feedback loop and identify which creatives drive retention and activity.
A DSP is a combination of several pieces of tech: the bidder, the ad-serving mechanism and the different algorithms that enable you to decide which ad to serve and which impression to actually buy.
Transparency and control is why SocialPoint wanted to build something in-house:
- Transparency: you can see which bids you were offered, which ones you successfully purchased, which ones you made an offer on but didn’t win. You also know exactly what you buy and where you buy it: there is no black box like there can be with ad-side DSPs.
- Control: you decide what you buy and at which price you buy it. Controlling the algorithm and the decision making means you control the type of users you want to buy and for which purpose in the game.
SocialPoint hasn’t built an in-house bidder, they use a bidder as a service technology. Then they built their algorithm on top of that.
Building a bidder in-house is extremely difficult: the technical infrastructure that you need is very high. On one side, a bidder works in real-time which requires a huge tech stack, and on the other side you need to have and maintain all the connections/API with the ad exchanges. SocialPoint decided to instead focus on the algorithm part.
When you build in-house you get bid requests, win logs and lost logs. So if you’re in control of the algorithm, you can enhance your bidding strategy to acquire users that you want at the lowest possible cost and find additional pockets of traffic.
There are 2 types of buys in programmatic: user-based educated buys and contextual buys. If you know something about the user and that he is worth spending for, the CPI doesn’t really matter.
In-house DSPs can’t compete with large managed DSPs on Android, or can’t compete with Google and Facebook’s algorithm and their ability to leverage their data to create lookalikes (including competitors’ data). However, when moving to contextual buys (e.g. due to ATT), it levels up the playing field which allows in-house bidders like SocialPoint’s to compete.
When you start working with an in-house bidder, you have to think about your long-term strategy.
- If you want immediate return on investment, use whitelists and what you already know. You cut DSP fees which gives you room for higher CPM buys.
- If you’re going for the long term, you need to buy every single demographic and start looking for patterns. This can be extremely costly because you need a lot of data if you look at things with a lot of granularity (combinations of OS, zip code, creative type, etc.).
Before iOS 14.5, managed DSPs were collecting IDFAs and able to use them across the different companies advertising. With contextual bidding, this advantage has been reduced a lot (even though they might still be able to find correlation at the publisher level).
Not everything is driven by algorithms. The UA team still has a major impact on how the campaigns are progressing in terms of performance, for example on the creative side.
There are no more blockers in terms of potential performance between using a SaaS bidder and a managed platform. Before ATT, SocialPoint ran blind tests against DSPs for LAT US campaigns (without device graphs, only contextual, same constraints, same creatives, same support) and monitored and shared performance across all partners.
Before, data was king and could make or break the success of a DSP. Now, data science and algorithms' cleverness and complexity are what will make the difference in the new ATT landscape.
By analyzing the bids you lost, you can run an experiment where you bid at a rate you would typically not consider in order to find out by how much the quality of the user could increase. SocialPoint has been surprised by bidding at crazy CPMs, getting very high CPIs yet seeing increased conversion rates in-game.
Before, with device graphs, DSPs knew where the high value users were and you would end up not getting access to that inventory through contextual buys. With ATT, those whales are pushed back to the pool that’s available if you’re working with contextual advertising.
When it comes to bidding, using the conversion value is a challenge. Not only because of its value, but also for budget control: DSPs usually have a very accurate budget control/pacing and the 48h delay that comes with SKAN prevents you from reacting quickly.
If everything is contextual, the benefit of having several DSPs bidding on the same SSP is much lower than when each DSP had device graph data they could leverage. So the question of incrementality from your DSP partners becomes even more relevant (i.e. Is it worth adding a new partner?).
Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.
In-house data is immensely valuable because you can get both pre-install and post-install data in order to have a feedback loop and identify which creatives drive retention and activity.
Notes for this resource are currently being transferred and will be available soon.
Alex
[💎@02:23] A DSP is a combination of several pieces of tech: the bidder, the ad-serving mechanism and the different algorithms that enable you to decide which ad to serve and which impression to actually buy.
The bidder is the technical stack that connects you with the supply and enables you to listen to the requests that come to you and have either an algorithm or a set up to say that you want to buy the impression/bid request and at which price, with which creative and with which link.
Alex
[💎@03:40] Transparency and control is why SocialPoint wanted to build something in-house:
[💎@05:34] SocialPoint hasn’t built an in-house bidder, they use a bidder as a service technology. Then they built their algorithm on top of that.
They initially used Beeswax: it gave the benefits of in-house (listen to the bid requests + have control on the decision making). Beeswax was listening to millions of bid requests from SSPs and as a customer you’d be charged based on how many you decided to listen to: you could limit to one country, one OS, etc.
Norberto
[💎@07:27] Building a bidder in-house is extremely difficult: the technical infrastructure that you need is very high. On one side, a bidder works in real-time which requires a huge tech stack, and on the other side you need to have and maintain all the connections/API with the ad exchanges. SocialPoint decided to instead focus on the algorithm part.
Alex
[💎@09:10] When you build in-house you get bid requests, win logs and lost logs. So if you’re in control of the algorithm, you can enhance your bidding strategy to acquire users that you want at the lowest possible cost and find additional pockets of traffic.
When working with partners, you have little visibility on why a DSP is bidding on a user: do they know something about that user? Or about other external, contextual factors (e.g. combination of US x Monday morning x iOS user, etc.)?
[💎@10:29] In-house is not for everyone. Having all this data doesn’t mean you can activate it.
Managed services is the majority of what SocialPoint still uses: for in-house, on top of building the algorithm you still have to activate the campaign and optimize it. Building the algorithm is a long term game and very difficult.
[💎@12:15] There are 2 types of buys in programmatic: user-based educated buys and contextual buys. If you know something about the user and that he is worth spending for, the CPI doesn’t really matter.
Initially they tried to decrease the CPI and succeeded, but did not pay enough attention to the quality of the traffic.
The KPI they’re looking at is the cost of acquisition of someone that’s going to generate revenue inside their games:
Alex
[💎@15:19] In-house DSPs can’t compete with large managed DSPs on Android, or can’t compete with Google and Facebook’s algorithm and their ability to leverage their data to create lookalikes (including competitors’ data). However, when moving to contextual buys (e.g. due to ATT), it levels up the playing field which allows in-house bidders like SocialPoint’s to compete.
The question becomes: how much prospection do you need to do to get to the point where you know the patterns and which chunks of users are valuable VS. how much can you jumpstart your campaigns by already knowing specific targeting.
[💎@16:23] When you start working with an in-house bidder, you have to think about your long-term strategy.
Norberto
You can use previous data but only for some time because things change, even week to week.
Managed DSPs also use competitors’ data to identify pockets of users.
[💎@20:06] Before iOS 14.5, managed DSPs were collecting IDFAs and able to use them across the different companies advertising. With contextual bidding, this advantage has been reduced a lot (even though they might still be able to find correlation at the publisher level).
Managed DSPs can still do some fingerprinting but it is more limited.
Alex
In-house bidders/DSPs are going to thrive under ATT because it’s more comfortable to control your data.
Now, SaaS platforms come with account managers and algorithms so you don’t need your own UI or an analytics team building your own algorithm. But to go to the next level, requirements and barriers of entry are pretty high: it requires the right data infrastructure and team.
[💎@24:05] Not everything is driven by algorithms. The UA team still has a major impact on how the campaigns are progressing in terms of performance, for example on the creative side.
[💎@24:54] There are no more blockers in terms of potential performance between using a SaaS bidder and a managed platform. Before ATT, SocialPoint ran blind tests against DSPs for LAT US campaigns (without device graphs, only contextual, same constraints, same creatives, same support) and monitored and shared performance across all partners.
[💎@25:40] Before, data was king and could make or break the success of a DSP. Now, data science and algorithms' cleverness and complexity are what will make the difference in the new ATT landscape.
Alex
Analytics can now discuss with DSPs about what the best conversion values are and this is very valuable. All games are different and in-house offers the granularity/flexibility.
With their current partners, they looked at the patterns in the data gathered from previous campaigns, then at the weight of every data point in the success of their campaigns and used this to define the conversion values for SKAN.
There is also value in the pre-install data and understanding which bid you win, why, where + working on IPM.
[💎@27:50] By analyzing the bids you lost, you can run an experiment where you bid at a rate you would typically not consider in order to find out by how much the quality of the user could increase. SocialPoint has been surprised by bidding at crazy CPMs, getting very high CPIs yet seeing increased conversion rates in-game.
[💎@28:45] Before, with device graphs, DSPs knew where the high value users were and you would end up not getting access to that inventory through contextual buys. With ATT, those whales are pushed back to the pool that’s available if you’re working with contextual advertising.
SocialPoint thinks that CPMs will go down, that the quality of the users will go up, and that if you have good tech plus save on DSPs fees then an unprofitable campaign before ATT could turn into a profitable campaign after ATT.
Norberto
Before iOS 14.5: by default, SaaS bidders provide their own algorithms. Once they had everything else set up, SocialPoint started working on their algorithm.
They went for a machine learning approach because rule-based approaches might lead you to mistakes. With machine learning, you train algorithms through cross-validation techniques: you train on a set of data that is not where you’ll evaluate the performance of the algorithm. It also allows for a complexity that humans can not come up with.
Post iOS 14.5: when it comes to measuring performance, the challenge with the conversion value is that the value is not that high (63), and only gotten through the first 24 hours. But their team makes predictions on the value of the purchaser up to 720 days. So they start working with more complex machine learning models linked with media mix modeling and incrementality.
[💎@34:16] When it comes to bidding, using the conversion value is a challenge. Not only because of its value, but also for budget control: DSPs usually have a very accurate budget control/pacing and the 48h delay that comes with SKAN prevents you from reacting quickly.
Alex
You need 2 sets of algorithms: the bidding algorithm and another algorithm that adds another layer for extra refinements.
The algorithm can not take over fully because there are strategic decisions that also need to be made, so you need to allow for additional input/exceptions.
So far they haven’t experienced the bottleneck of still relying on UA managers.
Alex
They had a lot of time to prepare for ATT and train contextual algorithms that would work even if fingerprinting is killed. On Android, device ID remains king.
[💎@39:57] If everything is contextual, the benefit of having several DSPs bidding on the same SSP is much lower than when each DSP had device graph data they could leverage. So the question of incrementality from your DSP partners becomes even more relevant (i.e. Is it worth adding a new partner?).
[💎@41:08] Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.
[💎@41:48] In-house data is immensely valuable because you can get both pre-install and post-install data in order to have a feedback loop and identify which creatives drive retention and activity.
Having this will be a competitive advantage.
Norberto
SocialPoint is still testing different solutions that new partners come up with, because right now there is a lot of opportunity since DSPs won’t be able to use their device graph.
Alex
[💎@02:23] A DSP is a combination of several pieces of tech: the bidder, the ad-serving mechanism and the different algorithms that enable you to decide which ad to serve and which impression to actually buy.
The bidder is the technical stack that connects you with the supply and enables you to listen to the requests that come to you and have either an algorithm or a set up to say that you want to buy the impression/bid request and at which price, with which creative and with which link.
Alex
[💎@03:40] Transparency and control is why SocialPoint wanted to build something in-house:
[💎@05:34] SocialPoint hasn’t built an in-house bidder, they use a bidder as a service technology. Then they built their algorithm on top of that.
They initially used Beeswax: it gave the benefits of in-house (listen to the bid requests + have control on the decision making). Beeswax was listening to millions of bid requests from SSPs and as a customer you’d be charged based on how many you decided to listen to: you could limit to one country, one OS, etc.
Norberto
[💎@07:27] Building a bidder in-house is extremely difficult: the technical infrastructure that you need is very high. On one side, a bidder works in real-time which requires a huge tech stack, and on the other side you need to have and maintain all the connections/API with the ad exchanges. SocialPoint decided to instead focus on the algorithm part.
Alex
[💎@09:10] When you build in-house you get bid requests, win logs and lost logs. So if you’re in control of the algorithm, you can enhance your bidding strategy to acquire users that you want at the lowest possible cost and find additional pockets of traffic.
When working with partners, you have little visibility on why a DSP is bidding on a user: do they know something about that user? Or about other external, contextual factors (e.g. combination of US x Monday morning x iOS user, etc.)?
[💎@10:29] In-house is not for everyone. Having all this data doesn’t mean you can activate it.
Managed services is the majority of what SocialPoint still uses: for in-house, on top of building the algorithm you still have to activate the campaign and optimize it. Building the algorithm is a long term game and very difficult.
[💎@12:15] There are 2 types of buys in programmatic: user-based educated buys and contextual buys. If you know something about the user and that he is worth spending for, the CPI doesn’t really matter.
Initially they tried to decrease the CPI and succeeded, but did not pay enough attention to the quality of the traffic.
The KPI they’re looking at is the cost of acquisition of someone that’s going to generate revenue inside their games:
Alex
[💎@15:19] In-house DSPs can’t compete with large managed DSPs on Android, or can’t compete with Google and Facebook’s algorithm and their ability to leverage their data to create lookalikes (including competitors’ data). However, when moving to contextual buys (e.g. due to ATT), it levels up the playing field which allows in-house bidders like SocialPoint’s to compete.
The question becomes: how much prospection do you need to do to get to the point where you know the patterns and which chunks of users are valuable VS. how much can you jumpstart your campaigns by already knowing specific targeting.
[💎@16:23] When you start working with an in-house bidder, you have to think about your long-term strategy.
Norberto
You can use previous data but only for some time because things change, even week to week.
Managed DSPs also use competitors’ data to identify pockets of users.
[💎@20:06] Before iOS 14.5, managed DSPs were collecting IDFAs and able to use them across the different companies advertising. With contextual bidding, this advantage has been reduced a lot (even though they might still be able to find correlation at the publisher level).
Managed DSPs can still do some fingerprinting but it is more limited.
Alex
In-house bidders/DSPs are going to thrive under ATT because it’s more comfortable to control your data.
Now, SaaS platforms come with account managers and algorithms so you don’t need your own UI or an analytics team building your own algorithm. But to go to the next level, requirements and barriers of entry are pretty high: it requires the right data infrastructure and team.
[💎@24:05] Not everything is driven by algorithms. The UA team still has a major impact on how the campaigns are progressing in terms of performance, for example on the creative side.
[💎@24:54] There are no more blockers in terms of potential performance between using a SaaS bidder and a managed platform. Before ATT, SocialPoint ran blind tests against DSPs for LAT US campaigns (without device graphs, only contextual, same constraints, same creatives, same support) and monitored and shared performance across all partners.
[💎@25:40] Before, data was king and could make or break the success of a DSP. Now, data science and algorithms' cleverness and complexity are what will make the difference in the new ATT landscape.
Alex
Analytics can now discuss with DSPs about what the best conversion values are and this is very valuable. All games are different and in-house offers the granularity/flexibility.
With their current partners, they looked at the patterns in the data gathered from previous campaigns, then at the weight of every data point in the success of their campaigns and used this to define the conversion values for SKAN.
There is also value in the pre-install data and understanding which bid you win, why, where + working on IPM.
[💎@27:50] By analyzing the bids you lost, you can run an experiment where you bid at a rate you would typically not consider in order to find out by how much the quality of the user could increase. SocialPoint has been surprised by bidding at crazy CPMs, getting very high CPIs yet seeing increased conversion rates in-game.
[💎@28:45] Before, with device graphs, DSPs knew where the high value users were and you would end up not getting access to that inventory through contextual buys. With ATT, those whales are pushed back to the pool that’s available if you’re working with contextual advertising.
SocialPoint thinks that CPMs will go down, that the quality of the users will go up, and that if you have good tech plus save on DSPs fees then an unprofitable campaign before ATT could turn into a profitable campaign after ATT.
Norberto
Before iOS 14.5: by default, SaaS bidders provide their own algorithms. Once they had everything else set up, SocialPoint started working on their algorithm.
They went for a machine learning approach because rule-based approaches might lead you to mistakes. With machine learning, you train algorithms through cross-validation techniques: you train on a set of data that is not where you’ll evaluate the performance of the algorithm. It also allows for a complexity that humans can not come up with.
Post iOS 14.5: when it comes to measuring performance, the challenge with the conversion value is that the value is not that high (63), and only gotten through the first 24 hours. But their team makes predictions on the value of the purchaser up to 720 days. So they start working with more complex machine learning models linked with media mix modeling and incrementality.
[💎@34:16] When it comes to bidding, using the conversion value is a challenge. Not only because of its value, but also for budget control: DSPs usually have a very accurate budget control/pacing and the 48h delay that comes with SKAN prevents you from reacting quickly.
Alex
You need 2 sets of algorithms: the bidding algorithm and another algorithm that adds another layer for extra refinements.
The algorithm can not take over fully because there are strategic decisions that also need to be made, so you need to allow for additional input/exceptions.
So far they haven’t experienced the bottleneck of still relying on UA managers.
Alex
They had a lot of time to prepare for ATT and train contextual algorithms that would work even if fingerprinting is killed. On Android, device ID remains king.
[💎@39:57] If everything is contextual, the benefit of having several DSPs bidding on the same SSP is much lower than when each DSP had device graph data they could leverage. So the question of incrementality from your DSP partners becomes even more relevant (i.e. Is it worth adding a new partner?).
[💎@41:08] Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.
[💎@41:48] In-house data is immensely valuable because you can get both pre-install and post-install data in order to have a feedback loop and identify which creatives drive retention and activity.
Having this will be a competitive advantage.
Norberto
SocialPoint is still testing different solutions that new partners come up with, because right now there is a lot of opportunity since DSPs won’t be able to use their device graph.
Alex
[💎@02:23] A DSP is a combination of several pieces of tech: the bidder, the ad-serving mechanism and the different algorithms that enable you to decide which ad to serve and which impression to actually buy.
The bidder is the technical stack that connects you with the supply and enables you to listen to the requests that come to you and have either an algorithm or a set up to say that you want to buy the impression/bid request and at which price, with which creative and with which link.
Alex
[💎@03:40] Transparency and control is why SocialPoint wanted to build something in-house:
[💎@05:34] SocialPoint hasn’t built an in-house bidder, they use a bidder as a service technology. Then they built their algorithm on top of that.
They initially used Beeswax: it gave the benefits of in-house (listen to the bid requests + have control on the decision making). Beeswax was listening to millions of bid requests from SSPs and as a customer you’d be charged based on how many you decided to listen to: you could limit to one country, one OS, etc.
Norberto
[💎@07:27] Building a bidder in-house is extremely difficult: the technical infrastructure that you need is very high. On one side, a bidder works in real-time which requires a huge tech stack, and on the other side you need to have and maintain all the connections/API with the ad exchanges. SocialPoint decided to instead focus on the algorithm part.
Alex
[💎@09:10] When you build in-house you get bid requests, win logs and lost logs. So if you’re in control of the algorithm, you can enhance your bidding strategy to acquire users that you want at the lowest possible cost and find additional pockets of traffic.
When working with partners, you have little visibility on why a DSP is bidding on a user: do they know something about that user? Or about other external, contextual factors (e.g. combination of US x Monday morning x iOS user, etc.)?
[💎@10:29] In-house is not for everyone. Having all this data doesn’t mean you can activate it.
Managed services is the majority of what SocialPoint still uses: for in-house, on top of building the algorithm you still have to activate the campaign and optimize it. Building the algorithm is a long term game and very difficult.
[💎@12:15] There are 2 types of buys in programmatic: user-based educated buys and contextual buys. If you know something about the user and that he is worth spending for, the CPI doesn’t really matter.
Initially they tried to decrease the CPI and succeeded, but did not pay enough attention to the quality of the traffic.
The KPI they’re looking at is the cost of acquisition of someone that’s going to generate revenue inside their games:
Alex
[💎@15:19] In-house DSPs can’t compete with large managed DSPs on Android, or can’t compete with Google and Facebook’s algorithm and their ability to leverage their data to create lookalikes (including competitors’ data). However, when moving to contextual buys (e.g. due to ATT), it levels up the playing field which allows in-house bidders like SocialPoint’s to compete.
The question becomes: how much prospection do you need to do to get to the point where you know the patterns and which chunks of users are valuable VS. how much can you jumpstart your campaigns by already knowing specific targeting.
[💎@16:23] When you start working with an in-house bidder, you have to think about your long-term strategy.
Norberto
You can use previous data but only for some time because things change, even week to week.
Managed DSPs also use competitors’ data to identify pockets of users.
[💎@20:06] Before iOS 14.5, managed DSPs were collecting IDFAs and able to use them across the different companies advertising. With contextual bidding, this advantage has been reduced a lot (even though they might still be able to find correlation at the publisher level).
Managed DSPs can still do some fingerprinting but it is more limited.
Alex
In-house bidders/DSPs are going to thrive under ATT because it’s more comfortable to control your data.
Now, SaaS platforms come with account managers and algorithms so you don’t need your own UI or an analytics team building your own algorithm. But to go to the next level, requirements and barriers of entry are pretty high: it requires the right data infrastructure and team.
[💎@24:05] Not everything is driven by algorithms. The UA team still has a major impact on how the campaigns are progressing in terms of performance, for example on the creative side.
[💎@24:54] There are no more blockers in terms of potential performance between using a SaaS bidder and a managed platform. Before ATT, SocialPoint ran blind tests against DSPs for LAT US campaigns (without device graphs, only contextual, same constraints, same creatives, same support) and monitored and shared performance across all partners.
[💎@25:40] Before, data was king and could make or break the success of a DSP. Now, data science and algorithms' cleverness and complexity are what will make the difference in the new ATT landscape.
Alex
Analytics can now discuss with DSPs about what the best conversion values are and this is very valuable. All games are different and in-house offers the granularity/flexibility.
With their current partners, they looked at the patterns in the data gathered from previous campaigns, then at the weight of every data point in the success of their campaigns and used this to define the conversion values for SKAN.
There is also value in the pre-install data and understanding which bid you win, why, where + working on IPM.
[💎@27:50] By analyzing the bids you lost, you can run an experiment where you bid at a rate you would typically not consider in order to find out by how much the quality of the user could increase. SocialPoint has been surprised by bidding at crazy CPMs, getting very high CPIs yet seeing increased conversion rates in-game.
[💎@28:45] Before, with device graphs, DSPs knew where the high value users were and you would end up not getting access to that inventory through contextual buys. With ATT, those whales are pushed back to the pool that’s available if you’re working with contextual advertising.
SocialPoint thinks that CPMs will go down, that the quality of the users will go up, and that if you have good tech plus save on DSPs fees then an unprofitable campaign before ATT could turn into a profitable campaign after ATT.
Norberto
Before iOS 14.5: by default, SaaS bidders provide their own algorithms. Once they had everything else set up, SocialPoint started working on their algorithm.
They went for a machine learning approach because rule-based approaches might lead you to mistakes. With machine learning, you train algorithms through cross-validation techniques: you train on a set of data that is not where you’ll evaluate the performance of the algorithm. It also allows for a complexity that humans can not come up with.
Post iOS 14.5: when it comes to measuring performance, the challenge with the conversion value is that the value is not that high (63), and only gotten through the first 24 hours. But their team makes predictions on the value of the purchaser up to 720 days. So they start working with more complex machine learning models linked with media mix modeling and incrementality.
[💎@34:16] When it comes to bidding, using the conversion value is a challenge. Not only because of its value, but also for budget control: DSPs usually have a very accurate budget control/pacing and the 48h delay that comes with SKAN prevents you from reacting quickly.
Alex
You need 2 sets of algorithms: the bidding algorithm and another algorithm that adds another layer for extra refinements.
The algorithm can not take over fully because there are strategic decisions that also need to be made, so you need to allow for additional input/exceptions.
So far they haven’t experienced the bottleneck of still relying on UA managers.
Alex
They had a lot of time to prepare for ATT and train contextual algorithms that would work even if fingerprinting is killed. On Android, device ID remains king.
[💎@39:57] If everything is contextual, the benefit of having several DSPs bidding on the same SSP is much lower than when each DSP had device graph data they could leverage. So the question of incrementality from your DSP partners becomes even more relevant (i.e. Is it worth adding a new partner?).
[💎@41:08] Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.Before, the creative was a way to get a whale to click and come inside your game and clickbait was ok. Now, you need to qualify people with your creative, have a first time user experience that is consistent and hook them up for long enough so that they can discover the value of your game.
[💎@41:48] In-house data is immensely valuable because you can get both pre-install and post-install data in order to have a feedback loop and identify which creatives drive retention and activity.
Having this will be a competitive advantage.
Norberto
SocialPoint is still testing different solutions that new partners come up with, because right now there is a lot of opportunity since DSPs won’t be able to use their device graph.
