John Koetsier is a journalist and analyst. He's a senior contributor at Forbes and hosts our Growth Masterminds podcast as well as the TechFirst podcast. At Singular, he serves as VP, Insights.
Meta will charge the equivalent of $127.40 for an annual subscription to its services in the EU starting in November 2023. That’s essentially double what Meta makes in ad revenue per user in Europe, which is a total of not quite $64 over the past 4 quarters.
Meta announced today that due to changes in EU regulations, it will be offering an ad-free subscription to Facebook and Instagram. The core reason: EU legislators have rejected Meta’s use of “contractual necessity” as a legal basis in Europe for processing user data for personalized advertising and have pushed Meta towards a subscription option which would then offer EU citizens a data-processing-free means of accessing Meta’s global-scale social platforms.
The subscription will be optional, Meta said today in an announcement post. The option is simple:
Either pay for the service
Or, consent to targeted, relevant ads
How many subscribers would it take to replace Meta revenue in the EU?
Thanks to Meta’s detailed quarterly and annual reports, it’s easy to understand both how much Meta makes from advertising right now, and how many Europeans would have to subscribe to Meta’s services to replace that ad revenue.
Meta has averaged 408 million monthly average users over the past 4 quarters
Each user returned an average of $15.99 per quarter
Meta made $63.97 per MAU from advertising
Note that total revenue per user is slightly higher than ad revenue per user, simply due to the fact that Meta offers some products for purchase.
But Meta’s subscription plan would theoretically bring in far more revenue, per average user, than ads.
Subscribers on the web will be charged $10.60/month for an annual total of almost $130 (so far there is no mention of an annual discount, though that could come). In-app subscribers will pay more, but I’m using the web numbers as Meta is following Twitter’s lead in charging more for in-app purchases to cover Apple’s and Google’s cuts.
That $130 is almost twice what Meta makes from targeted advertising per average user.
At these rates, Meta would need 208,572,327 European users to buy a subscription to replace all ad revenue.
Of course, as we’ve seen from Twitter (OK, X), very few people will subscribe. On X, about 640,000 people pay for premium, formerly Twitter Blue. If we take Twitter Ads Manager’s estimate of 372.9 million addressable users, that’s far less than 1% of users. To be precise, it’s under .2% of users. And there’s very little to indicate that Meta’s users would be substantially different enough to impact the economics on Facebook and Instagram.
This is not about a shift in Meta’s business model
208 million Europeans are not going to start paying the Euro equivalent of $130/year to access Facebook, when you can get it simply by consenting to ads.
Rather, this is simply about dotting I’s and crossing T’s so that Meta can point to the subscription model and tell European regulators that citizens can totally and completely opt out of data processing for personalized advertising if they choose to pay for the service.
That’s good news for advertisers, who don’t want to lose a valuable way of connecting with consumers, players, customers, and users.
It’s also good news for Europeans, who will continue to have a valuable service that connects them with friends, loved ones, communities, and celebrities, and who can continue to do it for free now that Meta has (almost certainly) cleared a legally plausible rationale for continuing to process user data for personalized advertising.
Google is starting to quietly signal an upcoming Chrome feature called IP Protection that will act much like Apple’s Private Relay feature, which hides IP addresses to make tracking — and marketing measurement — more challenging. Add IP Protection to Google’s soon-to-come Privacy Sandbox technology, and you’ve got interesting parallels between Apple and Google privacy technology, plus some parallels — and gaps — between the two tech giants’ technologies for marketing and attribution.
Comparing Google’s and Apple’s privacy, marketing, attribution tech
From Google, this set of software and standards includes:
Privacy Sandbox on Web and Privacy Sandbox on Android
Clearly, we’re seeing the emergence of separate but often related suites of software, standards, frameworks, and requirements from the tech giants. These tech giant initiatives are in 2 distinct but very related areas:
Privacy enhancement
Marketing measurement
The key reason for the connection: marketing measurement has typically required tracking, and that tracking has significantly impacted privacy. These tech giant initiatives are intended to rip out granular tracking as a vector for measurement and replace it with something very different: cloaked deterministic evaluation of advertising impact, with some noise sifted into the data, to provide analytics and preserve privacy.
Here’s what I’m seeing so far. (Let me know if I’m missing anything!)
* See the Privacy Sandbox website: “Privacy Sandbox also helps to limit other forms of tracking, like fingerprinting, by restricting the amount of information sites can access so that your information stays private, safe, and secure.”
These are complex beasts on both sides, with some parts baked in as OS-level components in iOS and Android, some grafted into the app submission and review process, and some that act more as platform-level directives than actual hard-coded realities. They are not monolithic projects or programs that are neatly subdivided, which makes them harder to fully grasp, and to fully understand the overall impacts on privacy as well as marketing measurement.
And, of course, they both deal with the world of mobile apps and the world of the open web, further complicating the overall landscape.
Intelligent Tracking Prevention vs the new Google IP Protection
Apple Intelligent Tracking Prevention, first introduced in iOS 11 and MacOS High Sierra in 2017, fights cross-site tracking by blocking third-party cookies, quickly deleting many first-party cookies, and blurring device characteristics to make fingerprinting harder. In conjunction with Private Relay hiding your IP address and App Tracking Transparency for requiring permission for the IDFA on mobile, it’s a powerful tool for privacy, plus a challenge for marketing measurement.
Now there’s a similar technology coming from Google for the Chrome browser, increasing an interesting degree of similarity — and divergence — between the Apple and Google stacks for privacy, marketing, and measurement.
The new technology from Google has been signaled in a Google Groups post by a member of the Chromium team. Chromium is an open source browser engine that forms the foundation of Chrome itself, as well as any other Chrome-based browsers, like Microsoft Edge, the Brave browser, and Opera.
“IP Protection is a feature that sends third-party traffic for a set of domains through proxies for the purpose of protecting the user by masking their IP address from those domains,” writes Brianna Goldstein, a senior software engineer at Google.
It’ll be an opt-in feature that will roll out in phases, she says, and will be “just focused on the scripts and domains that are considered to be tracking users.”
Functionally, this will work very similar to Apple’s Intelligent Tracking Protection, Goldstein says. The experiment does not currently impact Android WebView, the technology that allows an Android app to display web content, and will be limited in the beginning to Google’s own domains. It could cause some security concerns, Bleeping Computer notes, because proxied traffic “may make it difficult for security and fraud protection services to block DDoS attacks or detect invalid traffic.”
Private Click Measurement and SKAN vs Privacy Sandbox everywhere
Despite the fact that Apple absolutely needs privacy to be its crucial calling card as it expands its mobile universe to an ever-more personal PC that you wear on your face with no fewer than 12 cameras on it and in it looking both at your world and your face — plus 6 microphones — the company understands that that advertising drives free apps and the free web.
And that requires measurement, because advertisers need validation that they are getting ROI.
Google, of course, as an ad network primarily — at least in terms of revenue — never needed to learn that lesson.
Private Click Measurement measures both web-to-web and mobile app-to-web ad clicks, providing an 8-bit identifier on the source for up to 256 simultaneous ad campaigns per website or app, and a 4-bit identifier on the conversion, enabling measurement of 16 different conversion events. There’s a built-in time delay of between 24 to 48 hours, similar to SKAdNetwork, and measurement postbacks for both advertiser and ad network are handled in-browser and on-device.
Along with SKAdNetwork for mobile apps — which I won’t talk about here since we’ve covered it prettyexhaustively on the Singular blog — Apple is iterating through an increasingly richer advertising measurement framework. Yes, PCM pales in comparison to cookies (first or third-party) and SKAN pales in comparison to unfettered IDFA access, but that’s the point: they’re privacy-safe, and Apple will continue to add features over time.
On the other side of the fence, Privacy Sandbox on Web and Privacy Sandbox on Android are full-fledged initiatives to redefine the basics of how advertising works. Apple’s initiatives are more about mitigating adtech’s problematic capabilities; Google’s are about reinventing the world within which adtech exists.
Again, I won’t go into huge detail on Privacy Sandbox in this post: we’ve done it extensively already (focused, of course, on Android and not so much web, because mobile is where Singular primarily lives):
The one big obvious difference between the two suites in the marketing measurement area is that Google has provided capability for needed functionality in advertising and marketing: targeting and retargeting. It’s privacy-safe, which means it’s limited and restricted, but it’s there. Apple, on the other hand, while it will offer the ability for a retargeting signal in SKAN 5 so you know you’ve marketed to an existing user, player, or customer, does not offer any capability for targeting at scale in a privacy-safe way, or retargeting existing or former users.
That, perhaps, will wait until SKAN 6 or SKAN 7?
Google & Apple’s privacy/marketing/measurement suites: parallels and divergences
Ultimately when you boil down the privacy requirements of our evolving digital marketing ecosystem, you need a combination of items to limit tracking via cookies, identifiers, or device characteristics.
As far as all of that goes, there are clear parallels between the Google and Apple technology platforms. Despite the fact that Apple banned third-party cookies much earlier (2020) and that ITP has been in-market for years, as has Private Relay, Google’s coming Privacy Sandbox along with IP Protection will achieve roughly the same results. (Note: likely those technologies will be applied with varying degrees of vigor: Google after all makes almost all of its revenue from advertising, whereas Apple makes almost all of its money from devices, but the broad strokes are similar.)
But there are clear divergences as well, like for the same reason just mentioned.
Google’s Privacy Sandbox is in essence a reinvention of the entire advertising model, as we’ve already said. It’s something that has been called a 360-degree advertising suite by InMobi’s Sergio Serra:
“Privacy Sandbox for Android is a complete advertising suite … it goes 360 degrees from targeting, retargeting, fingerprinting crackdown, and attribution.”
That’s clearly beyond the scope of Apple’s ATT and SKAdNetwork, which focus entirely on privacy and privacy-compliant marketing measurement, disregarding targeting or retargeting.
The emerging privacy-safe marketing infrastructure means we need hybrid measurement
Put it all together, and you have the emerging privacy-safe advertising infrastructure.
It’s defined by:
Increasing respect for the individual and therefore, respect for the individual’s privacy
Decreasing data-gathering capabilities for the adtech ecosystem
Decreasing ability to track people from site to site and app to app
Increasing marketing measurement complexity
Growing reliance on semi-independent attribution frameworks and technologies (Privacy Sandbox, SKAN)
All of this is happening while we’re seeing increasing complexity in marketing mix, moving from just web or just mobile to web AND mobile AND CTV and outdoor AND custom SMS AND retail media AND influencer AND desktop AND console AND more and more channels and platforms … all of which is pumping the tires of the growing need for media mix modeling (MMM).
It’s also increasingly requiring what Singular calls hybrid measurement: marketing and advertising attribution based on a multiplicity of platform, cost, campaign, delivery, attribution, and first-party signals. Some of those are derived from deterministic sources such as SKAN or Privacy Sandbox, even if they are aggregated and noise has been added. Some are based on probabilistic technologies, like MMM itself. And others are based on deterministic signals that are the most accurate and detailed and precise of any that a marketer could hope for: your own first-party data.
All of this is a tremendous shift that is literally pulling the rug out from under the feet of marketers. But it’s both an industry and global legislative shift that won’t stop.
The one thing Singular can guarantee in all the change is that we will be providing everything you need for marketing measurement, optimization, and growth.
Singular CTO Eran Friedman spent some time with Redbox CTO Samual Chorlton on the AdBites podcast. The topic: everything SKAdNetwork, especially SKAN 4 help for those working on a transition from SKAN 3.
Hit play to watch it now, and keep scrolling for some of the highlights …
Data return to advertisers: from SKAN 3 to SKAN 4 help
In SKAN 3, as we know, Apple provided privacy thresholds to anonymize users. Low volumes of conversions from a campaign results in few or no conversion values. That works, but it punishes smaller advertisers, reducing the feedback they receive from their ad campaigns and lowering their trust in ROI and ROAS numbers.
“The idea is for anyone to be able to use SKAN. If you’re just beginning, you have barely a budget, you’re just testing things, you’re going to get some limited information, but not too much,” Friedman says. “But as you scale and you need to become more advanced, you’ll get more and more granular information for optimizations.”
Under SKAN 4, just 15 installs per campaign will start to give advertisers at least some data: at least a coarse conversion value: low, medium, or high. It’s not much, but at least it’s some signal to start calibrating and optimizing.
If that provides confidence to boost your ad spend, you’ll get more conversion values and will not just get coarse but fine values: 64 potential values. Increase scale even more, and you’ll get source identifiers, providing more detailed data you can use to tag campaigns, geos, or ad placements. And that tagging informs campaign optimization and improvement: getting more of what you want.
SKAN 4 help: defining terms in SKAN
One of the more challenging parts of SKAN in general is learning the language. That’s especially true for people who are new to mobile marketing, but it’s also the case for veterans of the industry, because many of the terms are new, or used in different ways.
So Friedman defined the terms for the AdBites audience:
Conversion values
A number that you choose that represents the value of a user. When SKAdNetwork encodes that number into a postback, and your MMP decodes it for you, you get clues about the effectiveness of an ad campaign.
Coarse conversion values
Low-volume campaigns like those we just talked about can only have coarse conversion values: 3 potential values like low, medium, or high to represent user value, and therefore campaign effectiveness.
Fine conversion values
When campaign volume is high, SKAN 4 permits more data to be encoded into conversion values: not just the 3 possible values of coarse conversion values, but the same 64 possible values that were available in SKAN 3.
(Note: in SKAN 4, you can only get a fine conversion value for the first postback. The second 2 postbacks are always going to be coarse conversion values.)
Source identifier
In SKAN 4, the source identifier is additional data you can get from your campaigns. Like conversion values, it is connected to crowd anonymity: high volume supplies more potential data than low volume.
If you achieve high crowd anonymity, your source identifiers will be 4-digit numbers that you can encode with data about your campaigns, geos targeted, ad sets used, ad placements, and more.
What an MMP does for you under SKAN
When SKAN first came out, some thought it meant there would be no need for MMPs anymore. After all, SKAdNetwork can send postbacks right back to advertisers themselves, potentially short-circuiting the need for independent results measurement.
Complexity turned out to be one of the core challenges. Plus the ability to be able to interpret advertiser models for ad networks so they could optimize based on known good results.
That’s one of the core reasons SKAN 4 help is so desperately needed.
“This is where we believe it’s the perfect kind of world for MMPs to provide the technology and management of all the SKAdNetwork framework: basically using the APIs, managing those conversion values, getting back those postbacks, and essentially trying to abstract all those technical terms and details so the advertiser doesn’t even need to think in terms of those encoded numbers and all the details, and they just get kind of the bottom line,” Friedman says.
That means campaigns, installs, dollars, registrations: human terms.
Plus, given the privacy-centric obfuscation of SKAdNetwork, including randomness Apple adds to the numbers, being able to use Singular’s AI-driven modeling in SKAN Advanced Analytics restores missing data in marketing measurement while not impacting user privacy.
SKAN 4 adoption: yet to scale
One other topic the two hit on the podcast: SKAN 4 adoption, which is lagging right now for many ad networks and especially the big platforms.
“I think all of them for sure are working to upgrade to SKAN 4,” says Friedman. “Some of them have, for example, started beta testing SKAN 4 and have selected advertisers that are already working with and running SKAN 4 campaigns. Others have done full launches and we already see most of their traffic has arrived to SKAN 4 … it’s on a network by network level.”
The timelines I’ve heard most industry experts mention are in the Q1 2024 range. More on that, likely, in a future Singular blog post, but the key point is that if you’re needing SKAN 4 help, you still have some time.
Looking for guidance on your SKAN 4 transition?
Watch the video above, but also go check out our SKAN 4 transition guide here. It will give you all the details you need to get started.
Once you’ve kicked that off, book a session with a Singular expert to go through your planned implementation, and how Singular can help make it all much, much easier.
App monetization changes from genre to genre of app and across different countries. Globally, casual games almost exclusively monetize via rewarded ads and interstitials while hypercasual games lean more on banner ads, and social apps monetize across a wide range of ad units but at a much lower scale.
But the data varies significantly from country to country.
I love data. I’m a sucker for reports that tease insights based on massive gobs of evidence about what’s actually happening across our digital ecosystems.
So when I saw a recent report by Kayzen citing insights based on …
24 trillion mobile ad bid requests
For 630,000 apps
Touching 1.4 billion people
Built via 10 billion machine learning decisions
… I had to have a chat with one of the authors. Click play on the video above (and subscribe to the Growth Masterminds podcast), then keep scrolling to see what I learned from Kayzen’s Tomas Yacachury.
App monetization varies from country to country
One of the super-interesting parts of the report is the app monetization profiles of different app types. Globally, games offer vastly more inventory than social apps or tool and utility apps and monetize better from rewarded and interstitial ads than banners or native ads.
But the profile changes from country to country.
“You see those radar charts in different countries and they are completely different, right … India and the U.S. are completely different,” Yacachury told me.”
USA vs India vs Germany
In the U.S., casual apps monetize largely via rewarded ads and interstitials, with a bit of banner thrown in. In India, banner is huge, followed by native, with a very little bit of rewarded and interstitial as well.
From the advertiser perspective if you’re looking for the biggest possible audience, casual games let you access about a third of all daily active mobile users in the U.S. In India, however, focusing more on tools and utility apps will capture almost two-thirds of daily active users.
In terms of ad formats banners and interstitials will cover 70% of the India market, while in the U.S. advertisers need to select a more balanced portfolio of ad unit types, and app publishers need to enable that broader portfolio to be able to ad space.
In Germany, it’s all about weather.
“So German users are very much concerned about the weather because you see, you see a lot of weather apps amongst like those apps that actually provide the highest reach,” Yacachury says.
Germany is also an exception to the global rule that iOS inventory is more expensive. In Germany, there are plenty of high-end Android phones, and Android inventory for banner ads and native ads are often more expensive than iOS inventory, so app monetization strategies need to adjust.
Interestingly, programmatic reach is highly atomized in the German market, Yacachury says. No single app reaches even 7% of the total available daily active users.
SKAN support, IDFA, and IDFV
SKAN support currently sits at 85% of all iOS bid requests, which is significantly high and, of course, likely to only continue trending higher. But almost all of that is SKAN 3, as we’ve repeatedly shown.
Interestingly, about 25% of bid requests have IDFA availability.
“What we see there is that IDFV … on non-IDFA inventory, it’s available on 75% of those bid requests, whereas it’s only available on 39% of IDFA traffic,” Yacachury says. “So it’s actually useful that … the coverage of IDFV is quite high on non IDFA inventory.”
Of course, beware of using the IDFV in France, where use of the IDFV now requires end-user consent.
App monetization: Singular can help
If ad monetization is a significant part of your revenue, Singular can help. Not only is ad monetization now available in Singular’s free tier, Singular’s admon solution provides quick and highly accurate insight on ad monetization, offering a much better picture of your LTV and ROAS.
Earlier this year I wrote about bad ads: high-CTR playable ad units that refuse to disappear, that interpret every touch as a click, or that crash the game you’re playing. The concern I had at the time was the tragedy of the commons: bad ad experiences ruining the mobile ad ecosystem. The question I didn’t answer at the time was: do these high-CTR playable ads that pop up insanely high click-through rates actually work?
In other words, do they achieve what advertisers want to achieve: installs and revenue?
High-CTR playable ads: data
Recently for a webinar with Kaizen, I had an opportunity to pull some data and analyze it for insights to exactly this question: do bad ads work well? Emotionally, I wanted the answer to be no. But I also wanted to let the data speak. Here’s an overview of what I found.
The data was from a campaign for a mobile gaming app:
88 million ad impressions
Mix of ad partners including search, big social networks, and SDK networks
$110,000 in spend
Here’s the first thing I found:
High CTRs are correlated with low CVRs
CTRs in the 60-70% range got CVRs of .4%, .6%, 1%, or even .07%
CTRs in this range generally originated from SDK networks
Low CVRs are correlated with high CVRs
CTRs of .5% or 1.5% are associated with CVRs such as 18%, 28%
CTRs in this range generally originated from the traditional blue-chip big platforms
The obvious question, of course, is whether a much higher CTR — even with a much lower CVR — result in similar performance?
A simplistic example: 1,000 clicks and 100 clicks look very much the same in the end if the conversion rate is 10X on the 100 clicks compared to the 1,000 clicks.
It turns out that the answer is yes.
I looked at the ratio of impressions to installs and installs to impressions to see how many impressions were needed from a particular network in order to get an attributed install. The result: 2 of the 3 SDK networks with high CTR playable ads actually do have higher install rates per impression than traditional large platforms.
And not just by a little: between 2X to 6X.
Wait … what about revenue?
I wanted to take it a little farther: all the way down-funnel to revenue. Does this still hold true?
So I compared the number of top-funnel impressions hitting people’s eyes to bottom-funnel dollars: money in your pocket. The question: how many impressions does it take with different ad networks to end up with $1 of publisher revenue after a player clicks an ad, installs an app, and converts to some form of revenue?
The result: for this campaign, most of the high CTR networks required fewer impressions to drive positive ROAS.
Far fewer.
That is, of course, why they’re doing what they’re doing in high-CTR playable ads: multiple clicks with multiple invocations of SKOverlay.
But … there’s a big caveat here
Comparing ad units is not always apples to apples.
Most of the ad units from the SDK networks were rewarded ad units, generally with playable ads, and pretty much always with aggressive end cards, which monetization expert Felix Braberg says are almost more important than the ads themselves. Many of the ad units from the traditional big publishers were banners or videos with much less aggressive end cards.
They demand more attention (you have to click out of them)
They receive more attention (you might play them, and you pay at least enough attention to get your reward)
The other big caveat: I’d really need to look at much more data to draw very detailed conclusions with a high degree of certainty.
What I can say is that there’s definitely a reason app and game publishers are using high-CTR ad units. They may not be the healthiest for the overall ad ecosystem, or for consumers’ impressions of advertising and app monetization based on ads, but they do work.
Want to see the full webinar?
If I may say so myself, it was a pretty good webinar with high-level participants, including:
Singular’s free tier now has free admon insights included along with global-best mobile attribution, ROI analytics, the best available SKAN analytics, fraud prevention, and deep linking solution. Click play below and keep reading for more details …
The new capability is thanks to product innovation from the Singular R&D team that will aggregate ad monetization postbacks for your ad network partners, enabling them to get insight and optimize campaigns without deluging them with postbacks for every single ad a user or player watches.
The even better news?
Singular’s free admon solution offers exactly the level of insight and transparency for marketers, product managers, and monetization experts that you’ve come to expect from Singular in user acquisition and cost data.
From all available datasets, just like UA data.
“When you’re monitoring your UA data, you’re going to be checking: is it the same that it looks like with the reports?” says Singular director of product Lisi Gardiner. “You have to look at the MMP data, the network reported data, and then the modeled data. And so the same thing with admon: you want to see: what’s the mediator reporting versus what is the monetization network reporting?”
“So the Singular solution basically gives you access to all these different datasets so that you have transparency and control over what’s being reported across all the different platforms that you’re working with to be able to market and monetize your application.”
Singular’s free tier (yes, it exists)
It might be a fairly well-kept secret in the mobile attribution and marketing measurement space, but Singular does have a free tier of service. You can test it without handing over your credit card details, and it includes the best of what Singular has to offer as the leading privacy-safe attribution platform.
That list includes:
15,000 paid conversions
Mobile attribution
ROI analytics
Data management
Fraud prevention
Cross-platform measurement for email, SMS, web, social, and mobile campaigns
Internal cross-promotion
Deep linking, deferred deep linking, and QR code links
SKAN attribution
SKAN cohorts
Customizable dashboards
Creative reporting
And more … now including free admon revenue monitoring and measurement. (Find out more about what that means in our ad monetization product overview.)
Ad monetization: adding a critical dataset
If you offer a free app, especially in the casual or hypercasual game sectors, ad monetization is likely a significant part of your revenue. For many apps, it might literally be ALL of your revenue.
How can you possibly calculate accurate ROI, ROAS, and LTV without having a way of measuring and managing the amount of ad monetization revenue you take in? Plus, to really know how your marketing campaigns are working, you need to match that revenue to cohorts of users, partners, and specific campaigns.
Then you might try to stitch that admon data together with user acquisition data, pulling revenue data from your mediation partner and applying it with a pinch of data science and a dab of magic to tie it into your marketing, costs, and acquisition data.
Good luck.
“For you to be properly able to optimize and make improvements, you need to be able to go more granular. And that’s where having the full combo of being able to pull in all your marketing costs and activity and combine it with the admon data is critical,” Gardiner says.
When you’re starting out, you don’t have as sophisticated a data management platform or growth stack as the bigger players. So achieving that is hard if not impossible.
The solution is connecting advertising cost, user acquisition numbers, and revenue (ad monetization and IAP/subscription, if you have it) together in a very granular way.
“We receive user level data for every time an ad is watched and we understand the value of that ad being watched as reported by the mediator,” Gardiner says.
Being able to connect that data with cohorts, campaigns, and ad partners then allows marketers to understand the ROI of their campaigns and the LTV of their users.
Free admon: how Singular was able to do it without DDoSing ad partners
Singular wasn’t previously able to offer ad monetization in the free product tier for a simple reason: too many postbacks.
Traditionally, MMPs like Singular send a postback to an ad network to inform them that an install happened or a revenue event occurred. This closes the loop on user acquisition, providing insight that helps ad networks optimize their campaigns on marketers’ behalf.
But while you can do that for an IAP that is $5 or $10 or $25, and while it makes sense for a significant event such as an app install, ad views are different. Ad networks aren’t built to accept postbacks for $0.001 of value accrued due to an ad impression. Not only is the amount too small, the number of postbacks would be overwhelming. Free admon wouldn’t be possible.
“We’ve heard from partners where it’s problematic for them to ingest this level of volume, because especially with hypercasual, the volume of ads being served can be quite high,” Gardiner says.
Imagine a hypercasual game with millions of players watching tens of millions of ads each day inundating an ad network with 30 million, 40 million, or even 100 million postbacks on a daily basis. The effect for some could be similar to a DDoS (distributed denial of service) attack … which would not be great for their core business. It would also be expensive on Singular’s side.
Singular’s innovation: batching postbacks.
“So the user, for example, watches 10 ads in one day, or in one session,” Gardiner says. “Maybe that accumulates to $1 per user. We can send that instead of saying: ‘Okay, there were 20 events.’”
Fewer postbacks equals less processing with the same amount of information being delivered. And since it’s at a much lower cost, it all works for both Singular and network partners, making free admon feasible.
Ad networks supported
Singular has integrations with the top mediators available, Gardiner says, so you’re pretty much fully covered on the major players.
That includes Unity/ironSource, AdMob, AppLovin, and more.
As Singular gets additional requests to add more mediation platforms and data connectors, the team is adding more, and the free admon solution will get even more comprehensive.
Plus, of course, ad monetization in your SKAN conversion models
iOS app developers make casual and hypercasual games too. And they also make ad-supported apps. But they don’t have the same level of visibility that Android developers currently have with GAID. (Until Privacy Sandbox, of course. More on that here.)
So ad monetization revenue has to be visible in an SKAdNetwork scenario as well for iOS developers to build out their acquisition, growth, marketing, and revenue plans.
So, what you can do is set up ad monetization events as part of your SKAN conversion model in Singular.
“We’re making sure that there are no gaps in terms of your marketing strategy and you have the full solution in terms of UA, seeing that with traditional attribution, seeing that within the SKAN framework, and being able to audit all the data across all the different activity that you’re doing,” Gardiner says.
Plus, there are some additional product features the team is working on that will be added to the ad monetization solution in the near future.
Next steps: how to get free admon and MMP services
Sounds interesting? Wondering how to get it … without having to fork over your credit card information or sit through a long sales pitch?
Start here: you’ll be able to sign up and get going almost immediately.
If perhaps you’re a bit bigger … you’re in growth phase or even enterprise, but being able to see ad monetization data side-by-side with cost, campaigns, and other revenue sounds good, there’s a solution for you too.
What are the 15 worst user acquisition mistakes you can make? The most awful, horrible, shot-myself-in-the-foot mistakes?
Pablo Gonzales is a performance marketing director for Admiral Media who has worked in performance and digital marketing for over a decade for brands like MercadoLibre and Banco Galicia. A few weeks ago he posted on LinkedIn, saying that he and his team had managed 26 million Euros in ad spend to scale apps over the last 12 months and learned 15 of the worst performance-killing mistakes.
Hit play here, and keep reading:
So, clearly, I had no choice, there was no other option: I had to invite him onto Growth Masterminds and dive into them all. Here are all 15 of the worst mistakes marketers can make in user acquisition campaigns.
(Note: if we’re missing some, let us know and maybe we’ll have a chat on the podcast too!)
The 15 worst user acquisition mistakes
Not diversifying across multiple ad platforms Just like investing money, Gonzales says, you diversify to protect your investment and to be able to maximize learning. “If you’re investing everything into Google Ads, TikTok … whatever channel it is, then first of all, you have nothing to compare it with. And secondly, if that bucket starts getting holes, then you will lose everything.”
Reducing iOS spend because of SKAN tracking limitations “Since ATT was introduced almost two years and a half ago I have seen so many times advertisers and marketers actually stopping or reducing iOS spend … but … it’s usually the best performing operating system … it means that you’re missing big, big opportunities.”
Failing to account for campaigns not meeting SKAN minimum thresholds Not getting data under SKAN is a huge challenge, and user acquisition mistakes while using new methodologies are easy. But this is very fixable, especially with SKAN Advanced Analytics from Singular, and with some smart strategy. “You need to actually take care of your campaigns, check the numbers, analyze what are the number of installs you’re getting on a daily basis, and then of course, rework or rethink the conversion volume mapping strategy that you have applied.”
Not consistently testing creatives, copy, and CTAs (experiment, always!) It’s tempting to just jump in with both feet and advertise your app precisely how you think it ought to sell. But you might be wrong. People might not think that way. And, you might also miss opportunities. “By testing, you can actually identify which opportunities you can find out there.”
Not having a creative testing approach It’s not enough to test. Testing is good, but it’s better to have a consistent creative testing approach that clearly lays out what you’re testing, how much data you need for a high-fidelity test with good predictive value, and what you’ll do based on the different potential outcomes of the test. Having a defined testing methodology is also important because it ensures that tests can be compared against each other in a reasonably apples-to-apples way.
Not aligning ad messaging with the app’s core value proposition User acquisition mistakes aren’t only about technology or spending. They’re also about strategy. For example, fake ads or even ads that just don’t hit your core value prop might have great CTR, but won’t necessarily translate to good CVR, monetization, or retention. In fact, if someone feels betrayed by the creative you used to trick them into an install, they’re likely to delete your app quickly — yep, this is me — which will have a negative impact on your retention and CAC.
Over-boosting underperforming campaigns Maybe you don’t have enough data. Maybe there’s not enough volume in each of your SKAN campaigns to generate sufficient insight. Fix the problem before boosting the campaign. “There are some self-attributed networks that might be over-reporting revenue or installs or different in-app actions. And if you don’t compare this with any other data sources such as an MMP or your own internal tracking system and reporting system, then you might also think that a campaign is actually working very good and in the end it might not.”
Keeping poor-performing ad creatives running, “hoping” they will turn around soon I get it. You love that one creative, or that one call to action. But it’s just not working. The best option is to switch to something that is, even if it breaks your heart. “I personally like to base my decisions on data. So if you don’t actually change anything, then data is not likely to be improved just because of magic, right?”
Ignoring ad placement data and other breakdowns User acquisition mistakes are easy to make because there is so much complexity in adtech. Not all ads are the same. Not all placements are the same either, even on the same platform, partner, or channel. Vertical 9×16 ad units might not fit well in feeds that optimize for a 5x:4 or 1×1 ratio. And other dimensions matter too, not least for double-checking that you’re getting what you’re paying for: “We got attributed install signup conversions for a placement we were not targeting … it wasn’t getting any impressions, but because of Meta’s own attribution modeling based on SKAN, which doesn’t report ad level data … they were assuming that, or attributing install/signup events to a placement that wasn’t getting any impressions.”
Only looking at in-platform reporting (hint: your MMP can be a treasure!) As #9 shows, you have to look at different sources of data: especially your MMP. Most reporting sources are at least partially based on modeling, which can be great but can also introduce error. So by checking as many data sources as possible, you achieve a greater opportunity to zero in on the most correct interpretation of reality. “Definitely you need to understand if the trend that you are seeing in one data source is actually following the trend that you see in a different, in another one … if not, then you need to deep dive and understand why there is such discrepancy.”
Ignoring retention & churn rates We’re closing in on a decade past the time when a user acquisition expert could just get users, throw them over the cubicle wall, and tell product or live ops that the users are now their problem. “There needs to be a correlation between the product team and the marketing team, because otherwise you will be paying for users who are actually being retained for one day, or they are not converting.”
Ignoring app uninstalls as a feedback metric App uninstalls are incredibly important to monitor for marketers, even if you’ve never historically looked at this metric. One of the critical variables: when are people uninstalling. After an update? Right after installing? At the paygate? When they can’t pass a level? Analyzing this can help inform creative strategy, onboarding efforts, major app design transitions, and much more. “You also need to take a look at how many users are uninstalling the app, and when they’re doing: when in the user funnel, the user acquisition funnel.”
Not being aware of the synergies between UA and ASO User acquisition mistakes are about more than UA. ASO, or app store optimization, is also critical. Good UA boosts good ASO. And good ASO boosts good UA. But understanding their synergies also helps you save wasted ad dollars spent on unnecessary campaigns. “You might be ranking in the top five for specific search terms. So you need to understand if it’s actually worth it to actually spend money on user acquisition … because if it doesn’t bring any incremental installs or revenue, then it doesn’t really make sense to spend money on those search terms that you are covering through ASO or organically.”
Underestimating the power of ASO ASO is powerful, and it’s often the last link in the paid user acquisition chain. Having a strong app store listing page that aligns with marketing strategy, language and creative is incredibly important.
Not analyzing user ratings and feedback regularly Ratings and reviews can dramatically impact both ranking and conversion rates in both the App Store and Google Play. Paying close attention to them will help you understand when you might need to add more focus on acquiring more positive ones, which is a form of insurance against poor conversion rates on your app listing pages as well as drops in search ranking.
User acquisition is hard: mistakes are easy
User acquisition mistakes are easy to make because there is literally so much to do, so much to learn, so much data to analyze.
You need to build creative
You need to design calls to action
You need to manage campaigns
You need to operate in multiple platforms’ systems
You have to learn new jargon, new math, new strategies, and entirely new systems of marketing measurement (SKAN, Privacy Sandbox) pretty much all the time
You have to collect data from dozens if not hundreds of systems
You have to analyze in-platform data, off-platform data, and your own first-party data
You have to Vulcan mind-meld that data into something standardized and normalized that makes sense
And you have to do it all quickly, so that you can optimize everything all the time
Apps are now just like websites and movies. Going forward, app publishers will need Chinese government approval to list their apps on the iOS App Store for China, as Apple is now requiring an ICP (Internet Content Provider) license filing to submit an app for publishing.
For movies, China restricts the number of foreign films to under a hundred, and they must pass review before release. For websites, all hosts and operators must apply for an ICP from the Chinese Ministry of Industry and Information Technology. If China detects a website without one, it will block the website. Mobile games as well have been heavily regulated.
Now, essentially, the same is true for all mobile apps.
iOS apps in China and ICP
After weeks of rumblings from China, Apple began requiring app developers to apply for a submit an ICP filing when they publish new apps, Reuters says. Apple has been resisting this for years, but given that China sets the rules for what happens in China, submission was inevitable.
Exactly what hoops you need to jump through will depend on exactly what type of app, game, or content you provide. From Apple’s developer documentation:
Apps: must possess a valid Internet Content Provider (ICP) Filing Number
Games: must secure an additional approval number
Additional app certifications required:
Apps with book/magazine content: must secure an internet publishing permit
Apps with religious content: must secure an Internet Religious Information Permit from China’s National Religious Affairs Administration
Apps with news content: must obtain an Internet News Information Permit from the Cyberspace Administration of China
There is no guarantee, of course, that applying for a certificate or permit ensures that you will get one. In addition, if China’s app policies follow their web policies, foreign companies that have no presence in China will not be able to apply for the ICP. Those who can apply include:
Partially or completely Chinese-owned companies
Chinese nationals with Chinese passports
Foreigners who are in China can apply for individual licenses
In other words, non-Chinese apps will have to work with a local app publisher, and may have to work on an arrangement for partial local ownership or licensing.
The big purge? What does this mean for apps already published in China?
The number of apps available on all app stores in China has been decreasing over the past 6 years, according to Statista.
That’s generally counter-trend to Google Play and the iOS App Store in general, where the overall number of apps has been either steady or growing (minus a Google Play crackdown on poor quality apps in 2018 and an Apple crackdown on old un-updated apps last year).
Chinese regulations have resulted in massive purges from the App Store in the past. In 2020, Apple removed 39,000 games from the iOS App Store for China in one fell swoop. Games have always required more onerous regulatory approvals. At that time, only 74 of the top 1,500 highest-grossing games survived the cut.
One likely scenario now is that Apple will be forced to do the same with app in the near future, presumably after giving apps some kind of grace period in which to apply for publishing permission. The other, perhaps more palatable option is that any app update will require proof of permission, so that in the near future, no apps without permission will be available on the iOS App Store in China.
But if history repeats, hundreds of thousands of apps could be affected almost overnight.
What other countries will do this in the future?
China has always been very selective about what gets through the Great Firewall. But I don’t see China as being alone in this trend towards balkanization of the digital ecosystem.
India has recently blocked many Chinese apps. Russia is cracking down on foreign sources of information. The U.S. could very well grow tired of the situation under which Chinese app publishers make billions in America, but American publishers can’t access the Chinese market. Just this year we saw that 17 of the top 52 app publishers on the planet are based in China, surpassing the United States for the first time ever.
That list includes:
Tencent
ByteDance (TikTok)
NetEase
miHoYo
37 Entertainment
Alibaba
Lilith
Baidu
Long Tech Network
Ultrapower
Zhejiang Century
G-bits
Most likely, we’ll see more of this division of cyberspace along real-world country lines, and it will be an additional headache for app developers who used to be able to write once and publish everywhere.
It’s also a risk factor for app publishers in any country to consider when taking their apps to foreign markets.
Once again we’re interviewing Singular software architect Ron Shub and CTO Yuval Carmel and getting all the insight about how Privacy Sandbox is evolving, how it works, and how marketers will be able to use data from it in Singular reporting.
Live Privacy Sandbox tests: A lot has changed since part 1
A lot has changed since part 1. We’re now running actual live-in-the-wild tests of Privacy Sandbox with Google and gaming customers like Gameloft. Plus, we know a lot more about Privacy Sandbox, how it works, what data marketers will receive from it, and how Singular will be able to build strong attribution, measurement, and optimization capabilities from the data it will provide.
Here’s how it will work on the device side. There’s more than a few steps to register sources (views or clicks) and triggers (events) and report them …
Privacy Sandbox on Android: on-device flow
And here’s how it will work on the server side: privacy-safe aggregated data will be summarized on an aggregation service, then decoded and displayed by Singular in reports that look like typical campaign measurement reports you might see today …
Privacy Sandbox on Android: server-side flow
What we’re learning about data and reducing added noise in Privacy Sandbox
As you probably already know if you’ve been following Privacy Sandbox development over the past few months, there are 2 limitations on the data you’ll get from Privacy Sandbox.
User level The sum total of all the events from a user will not exceed 65K (more will just not be recorded)
Aggregated level For each line in the result table, Google will add noise: fake data to obscure actual user activity and protect privacy
If you’re thinking about Apple’s SKAdNetwork, this is essentially a different solution to the same problem. Apple only provides data on installs and events when you surpass Privacy Thresholds in SKAN 3 or achieve Crowd Anonymity in SKAN 4. In Privacy Sandbox, Google provides way more data on events while obscuring user/player/customer identities and adding noise to the data so that marketers can’t reverse-engineer precisely which users came from which campaigns. That’s useful and good for privacy, but marketers also want to get the most data they legitimately can.
So how do you reduce noise?
First, you understand that there’s a trade-off between granularity and events versus more noise. If you want more granular data, you’ll get more noise. But if you’re OK with grouping users, you can reduce that noise level.
The default might come in around about 15% noise. But if you aggregate user activity to report on groups in each line of your measurement report, currently it appears that you can get it down to less than 1%. In other words, for big campaigns in big geos, it should be easy to reduce noise … but in small campaigns, noise could hit 50%.
Data you can expect from Singular
As in SKAN, marketers can choose metrics and events. Of course, in Privacy Sandbox, you’ll be able to choose many more, at least in the upper funnel. Like SKAN 4, you’ll get 3 kicks at the can for post-install conversion data.
Current thinking at Singular product indicates that mixed models will likely work very well, just like in SKAdNetwork. That means reporting some combination of post-install events and revenue.
Singular will decode all the data provided by Privacy Sandbox and supply modeled totals on:
Cohorted revenue
Marketer chosen metrics
Marketer chosen events
All of that will be presented with confidence levels so you know exactly how certain to be about the data you’re basing decisions on.
Protected Audiences in Privacy Sandbox (the new FLEDGE)
If you like names that say what they mean, you’ll like the new name for FLEDGE: Protected Audiences.
Singular has already integrated Protected Audiences, and the goal is to make it seamless for customers. Whereas in audiences as we know them today you have the data, in the privacy era the data is saved on-device.
How you use this audience will change:
Singular’s SDK will see that a user added to cart
Singular’s SDK will call the Protected Audiences API
Singular’s SDK will add the user to a segment
Google will distribute that audience to ad networks
Ad networks can bid on providing ads to that audience
Marketers won’t know who the user is specifically but will get aggregated reports that show a certain number of users have been added to the “Added to cart but did not buy” segment, for example. You won’t be able to export that audience and import it elsewhere, but it can be anonymously exposed to ad networks for retargeting campaigns.
Want to run your own tests?
It’s not exactly early days anymore: Privacy Sandbox has been around for a couple of years now, and sometime in 2024, Google will sunset the GAID and transition to the new technology for ad targeting, delivery, and measurement.
So it might be time to start thinking about your Android measurement strategy going forward. And maybe even testing what kind of data and measurement you can achieve.
Brands are using generative AI for marketing to get better faster. I recently chatted with Winclap general manager Avi Ben-Zvi about how he’s saving money, increasingly velocity, driving better ad performance, and boosting bottom line sales and revenue with generative AI.
Generative AI and performance marketing: 9 things that are possible now
What’s possible with generative AI in marketing?
Probably way too much.
1. Ideation: making new ideas
Humans are creative. But humans are also busy, tired, and (let’s be honest) fresh out of new, innovative, creative ideas. Having an infinite idea box handy is a wonderful thing, which is why Ben-Zvi is using tools like ChatGPT to get fresh input in his gray matter.
“We’re using Gen AI to start the process of actually ideating,” Avi Ben-Zvi says.
ChatGPT might not give you a concept that you can just run with verbatim. In fact, it probably won’t. But human brains are amazing creative association machines: bounce ideas off of them, and new ideas appear almost spontaneously. Kick off the process with generative AI, finish it with good old-fashioned wetware.
2. Generative creative assets
Sure, this one is a no-brainer. But it’s not just images for an ad: creative assets include text, images, voice, video, and more. And that’s a game-changer because it makes every team member their own team. Now writers can design, English speakers can publish in Spanish, and graphic designers can make videos.
(Sure: we need to add a bunch of caveats to each of those, and mistakes can be costly, but the fact is that marketers are already using generative tools to 10X their output, and the tools are getting better by the day. Put the right guardrails in place, and you’re off to the races.)
3. Accelerating creation of almost everything
When I go to Creative Diffusion to make some art for a podcast episode, blog post, or social post, I get a result in about 20 seconds. Usually there’s something interesting, but often I’ll finetune my prompt or just regenerate until I get a result that works.
“We’re using Gen AI to start the process of actually ideating,” Ben-Zvi says. “But also most importantly, generating assets at a faster velocity, and generating them for a fraction of the cost.”
That’s art in minutes.
And that’s powerful.
The same is true for translations, for voiceovers, for legit deepfakes (keep reading for kosher use cases!) and much more. Again: employ guardrails to ensure that outputs are quality and won’t damage your brand.
4. Reducing the cost of creative
Whether you’re using Creative Diffusion or Midjourney or a custom model trained to your brand’s look, feel, and standards, it’s pretty clear that when you can create in seconds or minutes, you’re saving bigtime.
5. Localizing ads
This sounds dangerous, right? I mean, screwing up on a translation is a really good way to instantly annoy people that you want to please.
Absolutely.
But … localization isn’t just about text and language. Sometimes it’s about ceviche. Or spaghetti, butter chicken, or Chairman Mao’s Red-Braised Pork.
“We’ve done something with DiDi when it comes to food delivery down in the Latin market, right? We wanted to do something in relation to celebrating national food dishes. So we gave it some direction and we gave the AI some direction in terms of … hey, we want to generate a dish that’s typically known in Mexico, for instance, or in Peru, and we want it to be ceviche, and we want it to show this and give me something with steam and have it fading out.”
Make sense: now you can localize ads with food, or activities, or sports, or clothing, colors, and much more.
6. Voiceovers? Yes, voiceovers.
ChatGPT might not do it but Fliki will not just do text-to-video but also voice cloning and voiceovers. Which means you can do some really cool stuff.
One example I heard recently:
A brand wanted to do something hyperlocal in China. with a big international star. So they used deep faking technology to basically clone the star into a virtual self, and then had him or her say things that were hyper geolocated for their various stores or locations
The result: potentially millions of pieces of creative from this one mega star.
(And yes, if you read above about legal deepfaking … this is what it referred to.)
7. Running better A/B or multivariate tests
We used to run tests on the colors of buttons. Avi Ben-Zvi isn’t impressed by that kind of minutiae, and doesn’t think that significantly moves the needle over time.
Because generative AI can help you make so much more, so much quicker, and so much cheaper, the scope of what you can test vastly expands.
“It’s the type of person you’re featuring, the way you’re featuring the product, obviously things like the length of your video or … voiceovers,” he says. “Is it a male voiceover? Is it a female voiceover? What type of value proposition? Are you testing out different brand messaging in each of these? There’s so many different elements to test within a video. And time again, every study shows, right: with the amount of brands that are out there today, especially direct to consumer brands, users are more inclined to purchase from a brand that they feel connected to.”
Better tests means better outcomes. And better outcomes means more revenue.
8. Delivering 10X more assets
Faster and cheaper?
2 + 2 = 4, and faster + cheaper = more.
Where creative proposals were once accompanied by 8 to 10 different creative elements, now they’re bundling 100 or 200. In other words: 10X’ing creative.
9. Allocating budget
This is always hard, and has a lot to do with measurement, historical results, and future predictions. AI can help here, whether we call it generative AI or not.
“We use something called our budget allocator, which is a predictive model,” Ben-Zvi says. “That’s looking towards the future and saying, okay, if I’m spending $5,000 here, $2,000 here, $1,000 there, what is going to happen to my cost per customer acquisition? What is going to happen to my return on ad spend … and playing with those numbers a little bit to see how the predictive model is going to respond.”
Useful, especially if it’s accurate.
The good thing is that models like these generate a lot of data, and machine learning can process that data and continually refine the model — as long as it’s fed with accurate results data — into future improvements in precision.
10. Creative avatars
Sometimes you need a spokesperson. A real human takes time, costs money, and maybe they’ll pull a Jerod-Subway on you and do something truly horrific that damages your brand.
AI avatars are almost free, and brands can control exactly what they say and do. It may not work for everyone, but there’s definitely a role here for marketing.
Much more on generative AI and marketing in the full podcast (including its impact on measurement)
John Koetsier: How can you use generative AI to turbocharge your ad creative?
Hello and welcome to Growth Masterminds. My name is John Koetsier. Everyone needs more creative all the time, much more, but building it at scale is hard. It’s challenging. It’s expensive. You need people for it. Can generative AI do the job for you or at least help you somewhat?
To figure it out, we’re chatting with Avi Ben-Zvi.
He’s Winclap’s general manager for the United States and North America. And works with clients like Shell, Paramount Plus, Rappi, Etermax, Lemon, and MovieStar. Welcome, Avi. How are you?
Avi Ben-Zvi: Great. Thanks for having me, John. Super excited to chat today.
Building with generative AI
John Koetsier: Super pumped to have you. What are you building with Generative AI?
Avi Ben-Zvi: There’s a lot we’re doing here. And ultimately the creative space has long been hampered by the fact that generating those assets can be expensive. It can be timely. It could be difficult to start the process from start to finish.
What we’re trying to do is change that trend right now. So we’re using Gen AI to start the process of actually ideating, helping us ideate, but also most importantly, generating assets at a faster velocity, and generating them for a fraction of the cost, whether it’s assets that don’t have humans in it, or even you actually using humans as well. And avatars that are replicating what creators are typically doing in something like a TikTok video that you’ll see, and it’s definitely a new frontier that we’re excited to explore.
Making your own style
John Koetsier: Super interesting that you’re talking about avatars and that brings up interesting points, right?
Because a lot of games have characters or they have a look, they have a feel, right? And so if you’re going to use a generative AI solution, you can’t just go to Midjourney or Creative Diffusion or Dall-E and say, blah, blah, blah, prompt and get what you want necessarily. You’ve got to give it a style.
You’ve got to give it some training. You might say, here’s a, here’s some images. How did you do that?
Avi Ben-Zvi: Yeah, absolutely. So that’s when our team has the human element, right. And the understanding to say we understand what typically works with this advertiser. We’ve done some creative testing in the past … here are some scripts to potentially work off of. Here are some ideas of the ways we want to approach this.
So, for instance, we’ve done something with DiDi when it comes to food delivery down in the Latin market, right? We wanted to do something in relation to celebrating national food dishes. So we gave it some direction and we gave the AI some direction in terms of … hey, we want to generate a dish that’s typically known in Mexico, for instance, or in Peru, and we want it to be ceviche, and we want it to show this and give me something with steam and have it fading out and doing all these different things. Right?
So we’re giving it that direction. We’re iterating off of it, and then we’re pulling in all the other things that we want, like a voiceover.
Or text in the background that we want to potentially test highlighting the national day or highlighting the dish itself, whatever it is that we can then AB test, whereas in, if you’re doing that all with the human element, it’s a lot to go in and create those things to professionally shoot that to even source the dishes in that example.
In this case, we’re able to do it all within the clicks on the computer. And then also add different variables so we can actually AB test, which is what every performance marketer is actually looking for.
Voiceovers in multiple languages
John Koetsier: Absolutely. So it sounds like you’re using it as part of the process with the people you already have.
Is there a piece where there’s just, it’s just surprised you it’s created something that you didn’t know that you needed, didn’t know that you wanted, but you saw it and thought, whoa. That’s amazing.
Avi Ben-Zvi: I think there’ve been a couple areas. I think one of the most interesting use cases I’ve seen set for it is in international businesses, right?
To have an avatar speak … one avatar speaking many different languages, I think is truly amazing. And so you’ll have it in English, then French, then Portuguese. And especially in a world where everybody’s focused on being globally enabled, that’s really important. And it’s really hard to do that with just the human element.
And here we’re able to do it fast and at scale, purely through these avatars.
And I see it, I’m like, problem solved. Unbelievable.
John Koetsier: You’ve discovered the digital influencer, the digital star, the digital hero. There was a story I heard, I guess it’s about three weeks ago now about a star in China, and they wanted to do something hyper local with this big international star.
And they used deep faking technology to basically clone this individual into a virtual self and then say things that were hyper geo located for their various stores, locations, whatever it was all over the country. So there’s literally a million pieces of creative from this one mega star.
Multiplying creative
Avi Ben-Zvi: That’s amazing.
It’s truly amazing that you’re able to do that and a lot of the proposals that we’ll put in front of our advertisers — now with human based creator influencer content, we can maybe start with pure human, and then as we look to scale that and get more assets, that’s where we’re going to, when we’re going to lean into that generative AI aspect.
And it’s going to go from, Hey, our proposal has eight, 10 different assets to a hundred, 200 different assets, which was a world that was nearly impossible to think about before when you had the creator/influencer marketing things starting to become a major focus for paid marketers. But now this is becoming a reality where you can do that.
And especially as creative as the star of the show, when it comes to marketing right now, it’s really hard to achieve great performance and a lot of the other variables like audience, algorithmic bidding, whatever it is, you really need to focus on creative. So especially as creator content is starting to scale and become popular, this is a great way to lean into a performance variable that is increasingly important in the market.
John Koetsier: The possibilities are amazing.
Hey, you can get The Rock and say, Hey, come on down to this place in this neighborhood in Caracas, Venezuela, or something like that. And there you go. Boom. You’ve got him.
Prompt engineer or just creative worker?
There’s a bit of a debate in the generative AI world about, do you call it prompt engineering or is that maybe too high a level? Where do you fall on that? Is it a prompt engineer or is it just, you’re asking the AI for stuff.
Avi Ben-Zvi: Well, I think it’s a balance prompt engineer is probably underselling it a little bit because I think it’s got to be a lot more than that. The human element is still really important.
I think even if we go back to before AI, automation in general has become a thing, but even as automation has started to escalate. You still need that human element for the strategy for driving things forward in a more innovative way. And I think that’s still the case here. And like I said, having that historical knowledge, haven’t done testing before.
So yes, you’re leading the prompts and that’s where prompt engineer comes from, but you’re going even further than that. And you’re understanding the nuances at a much deeper level than just simply saying, I need something like this. Give me something like X, Y, Z. You’re going much deeper with it.
And especially when you eventually end up pairing that with smart measurement tactics, that’s when you can go even further into those nuances to develop the right types of generative AI creative.
John Koetsier: Let’s hold those measurement tactics for a moment because I want to get into those, understand those, and I’m interested in those that are sometimes in — I want to say in the creative in terms of how long somebody is looking or engaging or watching or playing or something like that — as well as what happens after the creative, let’s hold those for just a second.
Generative AI technologies
I want to ask about the technology you’re using. Did you grab something open source? Are you using Creative Diffusion, Midjourney or using something different … using something you created yourself?
Avi Ben-Zvi: Yeah, we were definitely tapping into some of the third parties out there and there’s a lot of good stuff.
We’re certainly using Midjourney for images, for videos. We’re using something called Fliki for voiceover, which I think is super great also. People often forget about the voiceover element, which can be great because you can have assets you’ve already generated and then you’re adding just that influencer creator element on top simply using voiceover.
And even getting those scripts out there and working just within the ChatGPT to ensure that it’s relevant and it’s spot on with what we need, but it’s amazing. We just needed a couple of our key points to hit on and it can really get us a great starting point for our teams to build off of and get even more creative.
John Koetsier: It’s … if you just sit back for half a second and you put yourself in the mindset of two years ago and you just replay what you just said, this is insanity, right? It’s amazing. It’s literally almost incredible because of the creative toolbox. has evolved so much and so quickly.
You just labeled off like four or five different AI solutions that are part of your workflow, popping in at different parts for video. AI generated video that’s pretty cutting edge for your stills, for your voiceovers, all this stuff for your text, for even for the ideas itself … That’s a massive shift. And you know what? We don’t even really have the like out of a box tools or out of the box suite for gen AI yet. We have some stuff, right? Adobe’s got a few things, but a suite that pulls it all together. We haven’t seen it yet. It’s still very early days.
Early days for generative AI
Avi Ben-Zvi: Yeah it’s super early.
And you mentioned this seismic shift within the industry. I’ve been working in digital for 13 years now. And as creative has become more important, it’s become harder and harder — I think — for marketers to develop creative, that’s really good. And the key element of really good, I think, is the creative in creativity has often been lost.
And I think here, because you can do variations so quickly, and you can do much more testing, we’re getting that back. But that’s why the human element is certainly important too, because it drives a lot of that creativity. You could just do it faster now and you can test out different ideas.
Whereas before … you’re getting killed on budget. You’re getting killed … like I need this asset tomorrow. I don’t have time to generate something that’s super out of the box or going to really transform the way people think about our brand. And now you can actually do some of those things and be a little bit more creative when it comes to your creative, which is great.
But yeah, I think we’re just at the tip of the iceberg. Like you said, go back 2 years … we’re not really talking about any of these things. So I’m excited to see how this develops and how advertisers start to approach this in a really scalable way because we’ve just started this process.
John Koetsier: 13 years of working in digital? Clearly you started at 10!
So I had that all my career as well. Hey, it’s a good problem as we age, so that’s all good.
Marketing measurement and generative AI
Let’s get into some of the measurement stuff. Tell me a little bit about the measurement technology you’re using and tell me a little bit about some of the early results you’re seeing in terms of, are you seeing better CTR?
Are you seeing improved conversion rates? Are you seeing … better … What are you seeing, how are you measuring and what are you getting out of it?
Avi Ben-Zvi: Yeah, absolutely. So there’s a variety of different ways. You mentioned it earlier. There’s certainly the in-platform stuff, CTRs, conversions based on a pixel that’s in the particular platform that you’re advertising on.
We’d like to take this to the next level though, because for a lot of these consumer journeys, this isn’t just a sort of last click type of play. This is an engagement play. This is them getting adept at interacting with the brand. So we’re doing things like incrementality testing, doing something like geo holdouts, right, to understand when we’re going live with this particular creative in a given market. Are we actually starting to see more sales come through? And the key word being there, sales, not media conversions. Are we actually seeing sales come through? Is this really driving a true business impact?
So making sure that you’re tying back this creative testing towards something that’s real intangible from a business perspective, the sort of second piece to that is tech and what we’re building there to make sure that you are optimizing your media throughout.
So speaking of AI, we use something called our budget allocator, which is a predictive model. That’s looking towards the future and saying, okay, if I’m spending 5,000 here, 2,000 here, 1,000 there, what is going to happen to my cost per customer acquisition? What is going to happen to my return on ad spend and playing with those numbers a little bit to see how the predictive model is going to respond.
But another form of AI … in the measurement section to help us go out and scale and achieve efficiency from the outset.
The evolution of creative testing
John Koetsier: Nice. Very nice. Okay. So generative AI is changing a lot. Are you seeing any other changes or evolution in the creatives for mobile ad tech right now?
Avi Ben-Zvi: I think there’s a lot of changes as it comes to here.
Well, if I look back three, four, five years, The big sort of creative change was always like a change of background color, is it pink? Is it blue? Is it green? Is it black? Right? And everything else stayed the same. That was like the way performance marketers interacted with the world back then.
And I’m not sure that kind of stuff works. Maybe it does to a certain extent, but I think it actually caps at a wall with the way with how complex the digital media landscape has become and how complex the consumer journey has become within that digital media landscape. There’s so many touch points to being hit with ads and different organic journeys, and they’re on countless apps compared to 5 years ago, right?
So I think just how savvy creative has become or needs to be is a massive change in the industry from where we were four or five years ago.
John Koetsier: What does that mean? So let’s say that five years ago: let’s make the button red, let’s make the button green, which one performs more? AB test! Multivariate!
Okay. So that’s where we were. We were taking baby steps. So what’s the next evolution that you’re seeing? It’s not just the background or the color of the button. Are they totally different things? Is it totally out of left field stuff? And we’ll just throw stuff out there and see what happens. What is it … is it an evolution of the brand story?
Avi Ben-Zvi: I think it’s like all the above, right? It’s the type of person you’re featuring, the types of the way you’re featuring the product, obviously things like the length of your video or let’s look at like voiceovers. Is it a male voiceover? Is it a female voiceover? What type of value proposition?
Are you testing out different brand messaging in each of these? There’s so many different elements to test within a video. And time again, every study shows, right? With the amount of brands that are out there today, especially direct to consumer brands, users are more inclined to purchase from a brand that they feel connected to.
And they’re going to connect more to a brand that’s honing in more on what really works. as opposed to a brand that’s just trying to click convert right away. And that’s why at Winclap, we’re very focused on this sort of growth transformation story. And that’s really about sustainable, profitable growth, not like that quick hitting, let me just acquire a customer for one time.
It’s developing that connection with a particular consumer.
Customer journeys, MMM, and device identifiers
John Koetsier: I found it interesting that you’re talking earlier about the customer journey, and you mentioned it now, just as you’re talking about the complexity of the customer journey. And three years ago, five years ago, if we’re talking about the customer journey, we’d be talking about lots of measurement, customer data points, touches, cookies, IDFAs, GAIDs, all that stuff, right?
And some of that still remains relevant, but what you talked about earlier was basically triple M, lift, incrementality, right? And that is the new way of measuring. It’s not as precise, but it’s actually in some senses more accurate because you’re measuring actual full on complete results. Did I, did my sales improve?
Did my bookings improve? Do I have more subscriptions? All that stuff and then somehow all these different factors played a role. That’s an interesting transformation as well.
Avi Ben-Zvi: Yeah, I think so. You’re just seeing, right, that there’s, it’s hard, it’s getting harder to think about how you’re going to measure your media and the effect and the impact of your media four or five years ago, I think, you look down and people could get away with being very last click focus, even though we’ve known for a long time, that’s an imperfect understanding of the way a user interacts with the world, but they could get away with it and they could be successful.
Look at the amount of direct to consumer companies who built their business off of a pure Meta-then-Facebook acquisition strategy that was driven by something like that type of measurement, and I think today … complexity, the privacy changes, all that has obviously shifted how users interact with the world and also how brands can measure the efficacy of their media.
So you have to be a little bit more strategic when it comes to that on the flip side of that. I’ve seen brands who are also like, hey: I know measurement is imperfect period in digital, and like, I’m just gonna let go of measurement. I’m going to do things that are purely engagement focused, and I’m just going to measure, do I see lift in my sales?
And I know a very popular new-agey soda brand that takes that approach to TikTok and they’ve seen incredible success and they haven’t been hampered by finding the perfect measurement model and figuring that out before they go out in scale,
John Koetsier: There is no perfect measurement model. Everything that you can measure has some value, some more than others.
But yeah, ultimately does it result in growth of the brand and growth of sales? Very interesting stuff. Thank you so much for this time, Avi.
Avi Ben-Zvi: Yeah, thank you for having me, John. It was great.
