Head of Marketing
Apple will soon fully release iOS14 out into the wild. Its impact on digital advertising will be huge. So much so, we wrote this mammoth piece about everything we know so far. Its so big, we thought it would be helpful to have a hyperlinked table of contents to help you get to the bits you want.
Disclaimer: The following information presented is correct to the best of our knowledge, and at the time of writing. All our recommendations in this document are subject to change. We will be updating it as more information becomes available. Where we are unsure, we have tried to make it clear. For more information, please contact us or your media/measurement partners.
Apple has announced new privacy measures will be put in place in iOS14.
Mobile advertising currently relies on the Identifier for Advertisers (IDFA) to track users across apps and services to tie ad impressions to installs and revenue.
The IDFA is opt-in by default. Apple has historically offered people the chance to opt-out before, with their Limited Ad Tracking feature. In iOS14, IDFA will become opt-out by default. Apps will need to gain explicit consent from users to allow them to be tracked and attributed as they are currently (via the IDFA). Adverts will still be shown to all users, whether opted in or out.
Apple has not yet given a firm date for when these changes will happen. Indeed, it has been pushed back a couple of times already. The latest is that it will happen in March of this year.
Apple’s uncertainty about its release date reflects its participation in the solving of the upcoming challenges. As such, much of the targeting and attribution questions the industry has are still not fully answered. No one, not even Facebook or the larger Mobile Measurement Platforms quite knows the full impact expected, and how Apple will police certain practices in the future.
IDFA stands for Identifier for Advertisers. It is a combination of letters and numbers designed to identify a specific iOS device (including iPhones, iPads, and Apple TVs). Marketers use the IDFA to target advertising to specific users at scale, and measure the performance of those ads.
An IDFA is currently sent by the app by default (unless ‘Limited Ad Tracking’ is turned on by a user), meaning performance is easily measured and ads successfully targeted. When consent is not given by a user in iOS14 the IDFA will be ‘zero-ed out’ to anonymise the user.
An IDFA is not used by Android devices (which instead uses a GAID: ‘Google Advertising Identifier’). Apple’s changes do not affect the use of GAID.
Each user will be presented with the App Tracking Transparency (ATT) pop-up asking whether they are happy to be tracked by the app. If they do not give consent, the app will not be able to access the IDFA. Apps that do not present the ATT pop-up and still attempt to track people will be removed from Apple’s App Store.
Apps will have some control over when the ATT pop-up is shown, and a pre-emptive message encouraging people to give consent will be possible. Apple is warning against apps trying to incentivise consent. Non-monetary incentives (such as free bonuses) for consent may be possible. But we would be very cautious in recommending an app tries this. Furthermore, Apple is likely to bar any apps that make giving consent a prerequisite for using the app.
For an advertiser to attribute credit for the new install to a referring app using an IDFA, the referring app will also need to have been given consent to track.
Early estimates on how many people will opt-in to an app vary – but do not look encouraging at around 5-20%. Expected opt-in rates vary by industry and product type. Apps looking to use more personal data, such as financial or health data are bracing for a lower-than-average opt-in rate.
In a recent Singular webinar, it was suggested that users that have switched on Limited Ad Tracking in iOS13 won’t even be shown the ATT pop-up on any app. When they migrate to iOS14, their Apple device will remember and automatically reject consent to track their IDFA in each app they have installed.
Attribution and measurement will be explored in more detail – see section 3: How will measurement and attribution be affected?
For where an IDFA is not present, Apple has designed a solution that allows for permissionless (i.e. does not require user consent), privacy-centric, aggregated attribution: SKAdNetwork.
SKAdNetwork is not new. It was released in 2018 to help marketers track users who turned on the Limited Ad Tracking feature. However, the recent updated version has more functionality.
SKAdNetwork will allow marketers to measure attribution of mobile advertising on iOS. It will provide a variety of functions for mobile advertisers:
But it isn’t perfect:
How SKAdNetwork works
A user clicks on an ad for your app inside another app, such as Twitter.
They install and open your app. Postback data is collected (including Ad Network ID, Campaign ID, App ID, Transaction ID, and Conversion Value) and stored on the user’s device. A 24-hour conversion timer is then set live. In that timer, multiple conversion events (such as a registration or in-app purchase) can be recorded, and the timer is reset every time an event of a higher priority than the previous event is recorded. The highest value conversion is then posted back. For more details, see section 3.5 Events and ConversionValue.
When the timer expires 24-hours after the last post-install event, the postback data is sent to the ad platform or Mobile Measurement Partner to be reported on.
The advertising performance implications of SKAdNetwork are wide-ranging across targeting, attribution, and reportable performance. These will be discussed in the following chapters of this document.
In theory, nothing changes for mobile advertising outside of iOS14. In practicality though, things will change. The way major supply partners (e.g. Facebook and Google) adapt and build products to overcome the approaching challenges will have consequences for all mobile advertising, not just that on iOS14.
The IDFA is already only available in mobile apps, not the mobile web. So a Safari or Chrome user on an Apple device wouldn’t be tracked using an IDFA. Therefore website-to-website tracking is unaffected by this development.
But app-to-website tracking is harmed though in a similar way to how app-to-app is harmed. To help advertisers attribute clicks to conversions in a privacy-centric way, Apple has launched ‘Private Click Measurement’. But it has its drawbacks and Facebook isn’t a fan.
For more information on PCM and its drawbacks, see point 3.7 App to web and PCM.
For more information on Facebook’s solution, read section 5.1.3 Facebook Measurement section.
Broadly speaking, targeting audiences in digital advertising will become more limited with the absence of IDFAs. It is expected that targeting will become more contextual, rather than based on user attributes and behaviours.
There is a feeling that this might mean advertisers increasingly use digital channels for higher up the marketing funnel. If people can not be determined to be in-market and giving buying signals so easily, then performance marketing at the bottom of the funnel might become harder. By focusing on context, app marketing will be forced into reaching broader audiences who will be less ready to buy.
Targeting will be based mainly on data provided by end content providers suggesting details about the intended audience. E.g. app category and genre, audience interests, page content.
There will be severe limitations to other forms of targeting which will detrimentally impact performance marketing and its ability to find high value users and decrease wastage.
Website custom audiences of most (perhaps all) types are expected to decrease in size as people refuse to give consent to be tracked. The small percent of people who do give consent will be invaluable here, but the absolute number of people might be too small to build effective custom audiences. Where in the past website custom audiences would be split by user behaviour (i.e. build a retargeting audience for each stage of the user journey, or a lookalike audiences for ranges of value conversion), marketers may have to group as many valuable events together as possible to make any one new website custom audience big enough.
People who do not give consent will not be available for retargeting and lookalike audiences using pixel-based approaches. This is because it will not be possible to match a person from your audience on your app or website back to the same person on non-owned apps.
It may still be possible to use CRM data such as email addresses, names, phone numbers and so on, for the purpose of building lookalike audiences. But it would go against Apple’s privacy guidelines, and any app discovered doing it may be punished by removal from the app store. Although it is hard to see how Apple will be able to police it. So do this at your own risk, and speak to an Apple representative for confirmation.
It is important to remember that consent will need to be given by both the advertiser’s website/app and the publisher’s website or app. As such, successful crossover is likely to be very small.
Some ad personalisation will be limited. Personalisation using contextual or publisher-provided (2nd party) data will be largely unaffected. However, personalising creative using 1st party data (say for dynamic retargeting ads) will rely on successfully gaining consent on the brand and publisher apps hosting the adverts.
Sequential targeting across different publishers will see very little scale as it would need the user to be opted in to sharing IDFA consent across too many 3rd parties.
Sequential targeting of creative within the same app publisher may be slightly easier because there are fewer times consent is required to be granted. But realistically, sequential targeting will be another tactic with minimal scale.
Hyper-localised targeting will be lost without an IDFA. Instead Apple is happy to share an approximate location. How approximate this is remains to be seen. But this shouldn’t impact campaigns and performance too much as hyper-localised campaigns offer less scale and therefore tend to not move the needle quite so much.
Due to the lack of insights more granular than at a campaign level, the ability to do extensive A/B testing is harmed. Testing the post-click (read: metrics such as conversion rate, and purchases and ROI) impact of small scale factors such as CTA button colours or ads’ colours will not yield a set of comparable results unless separate campaigns are built for each. As such, due to the limit of campaigns allowed per ad network, A/B testing programs won’t be able to manage as many concurrent tests as might be desired.
Lastly, most tests that operate purely within one channel should still be able to yield insights into what drives media metrics (i.e. engagement metrics like click-through rate and cost-per-click).
There are obviously numerous times a marketer will want to purposefully withhold their advertising from certain people. iOS14 will have a detrimental effect on each of these times in those audiences that opt-out of tracking. Not only will this harm performance, efficiency and wastage, but it will also make the user experience of advertising worse.
Limiting the number of times an advert is shown to an individual will be harder on many channels. Frequency capping at the device level will become harder (but within the same session will still be fine). Where log-in data would be available, such as on Facebook and Google, Apple is clear that non-IDFA attempts to identify users is also banned without consent. Therefore frequency capping on these platforms will also be hampered hugely.
There are potential workarounds, but are not 100% effective. Sources of knowledge such as IP address might be usable, but they are not as accurate. Alternatively, app developers have their ID for Vendors (IDFV). This is an ID that a publisher can use across its multiple apps. So this might help frequency cap across larger companies such as Google.
Frequency capping at a creative level across multiple channels will not be possible.
As a result, marketers should be ready for audiences to be overexposed to your adverts. Creatives will need to be part of a larger library of variety, changed regularly and used for less time. Ad fatigue is expected to set in quicker (in terms of time, rather than # of impressions).
With the absence of an IDFA, creating accurate exclusion audiences from 1st party data will be harder. So preventing people from seeing ads who are existing users or customers will be impossible using pixel-based methods. It may still be possible and permissible to exclude people using CRM data (matching email address gained from user to the same user in a signed-in environment like Facebook and Google).
Creating (and maintaining) distinct audiences where there is minimal overlap of people will be harder. Even platforms with an enforced log-in such as Google and Facebook do not expect to be able to do this because Apple has made it clear that non-IDFA identifiers will not be allowed without consent.
Understanding the impact of your mobile advertising in iOS14 will become less clear as fewer new users (along with any revenue generated) can be traced back to an advertising touchpoint at a user-level.
SKAdNetwork is designed to limit tracking of data to the campaign-level. So say goodbye to more granular reporting and insights. While there is much less user-level data, there is however Publisher ID included – giving marketers more transparency which individual publishers within an ad network delivers performance).
Apple is limiting advertisers to 100 campaigns per ad network. 100 campaigns may sound a lot, but these have to include all targeting variations. Apple is limiting it to 100 so advertisers cannot make it easier to track individual users by making each campaign much tighter in its targeting.
Some ad networks and advertising platforms are limiting the number of campaigns further. This will be discussed further in section 5: How are the major channels reacting?
Because SKAdNetwork was designed to link a click to a conversion, it was expected that there would be no view-through attribution where there is no consent.
However, an Apple update (28th Jan) suggested that view-through attribution will be possible with SKAdNetwork for video, audio and interactive advertising. For an impression to be considered ‘viewed’ and be able to claim credit for a conversion, the advert will have to be on screen for a minimum of three seconds.
Additionally, SKAdNetwork will only record up to 15 viewed ad impressions per source app, before it will start forgetting the oldest impression. So if your ad was the 1st impression and the user saw 10 more ads in an app before converting, SKAdNetwork would attribute that conversion to the impression. But, if the person saw, say, 20 impressions after your ad was served, then any later conversion would not be attributed. We are not yet sure how Apple will track and enforce this.
A view-through ad impression must deliver a conversion within 24 hours for it to be attributable.
Attribution will go to the most recent impression with the highest strength. So a clicked ad will take precedence over a viewed ad, even if the viewed ad was served later, but if there are only multiple views, then the last view will take the credit.
Probabilistic methods of attribution in the absence of an IDFA will be out, too, in favour of a deterministic click. Tactics such as ‘fingerprinting’ (where the user’s IP address is combined with their device type, local network data, clipboard contents, and any other accessible data to generate a “fingerprint” that attempts to uniquely identify that user) was being considered as a workaround, but Apple has moved to dismiss this from being allowed.
SKAdNetwork will mean that the data available will be aggregated, delayed and limited. It is aggregated to maintain individual user’ privacy. This means that marketers will get less granular data for analytics and insights.
App advertising data will no longer be real-time, but delayed by potentially a few days. After the install, and the user opens the app for the first time, the device will report the install 0-24 hours after a 24-hour timer expires. This means that installs will be reported 24-48 hours after the install happened.
Conversions after an install will be reported 0-24 hours after a 24-hour timer expires after the final conversion event occurs. See the next section for more details.
ConversionValue is the new way to measure post-install activity to define the value of users
You are given 64 values (0-63) to code events per user within the app. It is important to note that these ConversionValues can only go up, i.e. from 10 to 20. Value can only be updated by a higher value event (e.g. app registration to app free trial) and within the timer. Marketers need to keep their ConversionValue events the same across their full range of channels and ad networks to allow for consistent reporting.
It appears that the ConversionValue of 0 is reserved by Apple to report installs – so don’t try to use it for anything else.
Tying accurate and long-term revenue to advertising campaigns will become more difficult for mobile advertisers. Once the 24-hour timer expires any further events, notably revenue, will not be posted back for reporting.
But it gets even harder. All revenue has to be included in the 64 values you have to report on post install events. A different value would need to be coded for different amounts of revenue. This much less problematic if you are a subscription app with one, two or three prices. But if, for example, you are a shopping app with dozens or hundreds of different price points, then you cannot possibly postback each unique revenue amount. Therefore, you will need to bucket revenue amounts into ranges and postback a value when a user falls into a certain range. For marketers taking this approach, measuring exact ROI will not be possible.
To overcome these challenges, some marketers are attempting to build a model to predict Lifetime Value (LTV). Talk to a Mobile Measurement Partner for more information on how they are helping marketers do this.
Tracking users over the web (including mobile web) doesn’t rely on an IDFA. So app-to-web advertising might not have been too badly affected. But Apple wasn’t happy with allowing attribution of an ad click to a conversion of a single user.
Private Click Measurement (PCM) is an API from Apple that prevents websites, on Safari, from sharing data while still letting advertisers measure the impact of web advertising. Each device will be treated like it is in private browsing mode, meaning attribution will be based on limited and delayed data.
PCM has been criticised by Facebook for not being a usable solution, so they have built a solution that is designed to bring more measurement ability to app to web advertising. See section 5.1.3 Facebook Measurement for more details.
Using URL Parameters such as UTMs to understand where web traffic arrives from should be unaffected.
Cohort analysis involves analysing groups of users with shared characteristics. Typically in marketing, cohorts are subsets of users grouped by a date: either acquisition or ad exposure date.
Because data is aggregated and delayed without IDFAs, cohort analysis will be unusable (although some MMPs are saying they have a probabilistic solution.
A common use case of cohort analysis is determining performance by the day the ad impressions are served. Because data is delayed, that will be inaccurate, and performance will be shown by the day SKAdNetwork posts the conversion data to the publisher/MMP, regardless of when the impression was served, or when the conversion actually happened.
This will harm optimisation and may skew results (e.g. payday looks even stronger than usual, even if much of the impact was delivered by an ad three days earlier).
The data available for reporting in Nara Analytics will be aggregated, delayed and limited as it comes from MMPs and self-attributing networks (like Facebook and Google).
Performance from mobile advertising on iOS14 is expected to look dramatically worse. With shorter lookback windows, no view-through attribution, and much less revenue able to be attributed back to a campaign, visible (i.e. trackable) performance from mobile advertising will appear worse.
This does not mean (especially in the short term) that mobile advertising isn’t still performing. Instead, look at how ‘organic’ performance improves as your BI and attribution tools can’t determine where the conversions and revenue came from.
However, in the longer-term, the algorithms of the major advertising platforms may get less effective at finding relevant and valuable audiences, as there is less data available for lookalikes and retargeting. So performance may well drop in the future.
There will be choices to be made on how to assign budget across the two main mobile operating systems.
Historically, iOS users have been typically viewed as more valuable and therefore cost more to reach in your advertising. But if iOS14 users are less trackable and targetable, we might see them become much cheaper on a CPM basis.
In addition, Android users could become more desirable so more budget might flow there. This would further lower iOS14 prices but also inflate Android CPMs.
Some advertisers may want to stop iOS spending, and feeding spend back in slowly as SKAdNetwork performance is gauged. Some may decide the suddenly lower CPMs of iOS users is worth the risk as those users are still likely valuable.
Google’s Android has its own version of Apple’s IDFA called “Android Advertising ID” (AAID). Google has slowly been ramping up its user privacy protections on Chrome. It has not shown its hand yet to reveal if it will do the same for Android, but it is largely expected within a few years.
Apple will itself validate when a click generates an install and will share that information with the ad network, platform or MMP through an encrypted postback.
This sounds good. Bad ads claiming credit for fake installs becomes much harder to pull off. Ergo, attribution fraud will be harder for bad actors.
However, less visibility into ad networks with data aggregated at a campaign level will make it easier for fraudsters to sell their fake traffic on their apps and hide their inability to drive business outcomes.
There has been a resistance for mobile marketers to invest in brand advertising. The combination of digital advertising’s measurability and targeting abilities has meant it has largely been adopted as a direct response, sales activation channel to drive immediate outcomes. This has come at the expense of exploring it to drive brand outcomes such as increased brand awareness and favourability which lead to longer term growth.
If mobile advertising’s ability to target people who are likely to buy a product imminently and then track that through to an immediate action at enough scale is harmed, then it will be hard for it to justify its budgets on a media efficiency metric like CPI or ROAS.
Mobile advertisers may be forced into targeting broader audiences who do not have an apparent need or want for the product or app. Advertising will have to justify its investment in a more macro sense, looking at business KPIs (such as net profit) over a longer period of time – lending itself to higher funnel activity.
Levels of readiness and response vary across the digital platforms. Where some, such as Facebook, are very active with solutions and education, others are yet to show their hand.
It is likely that the smaller inventory suppliers will take the lead of Facebook and Google. That is, they will see how Facebook and Google respond and try to align their solution behind them to attempt to make mobile advertising less fragmented.
Facebook has announced it will comply with Apple’s App Tracking Transparency (ATT) framework and SKAdNetwork. But there will be big changes to how campaigns are set up and run, in addition to the already discussed limitations on targeting tactics such as lookalike audiences and retargeting, and on attribution.
For app campaigns specifically, advertisers will be limited to 1 Ad Account per app for campaigns targeting iOS14 users. Those campaigns will need to be standalone and there will be a limit of 9 active campaigns per app and 5 ad sets per campaign of the same optimisation type.
You could use Automated App Ads (AAA) to help consolidate campaigns with its simplified structure of one campaign, one ad set, and one ad. Our experience shows this can deliver improved performance but you’ll generate fewer learnings and insights compared to a traditional App Install campaign.
For web–based campaigns, advertisers will be limited to the use of 8 conversion events per domain. Facebook will initially configure the conversion events they believe are the most relevant to your business based on your activity (last 28 days). If necessary, you will then be able to change your configuration via a now-available (end of January) section in Events Manager.
If more than one event is completed by a user (i.e. “add to cart” and “purchase”), only the higher prioritised event will be reported. Reporting will be partial in Ads Manager and Events Manager for additional events outside of the priority 8 events.
After the changes from Apple, ad sets optimising for an event beyond the 8 that are prioritised will be paused.
When using Value optimisation, each purchase event will count as a minimum of 4 events toward the 8 event optimisation limit. This is because, as touched on before, for advertisers using dozens and hundreds of products and prices, Facebook will automatically create value sets (minimum of 4 e.g. £1-£5, £5-£50, £50-£500, £1000+) but we won’t be able to see the exact granular value with the new setup.
Advertisers will however be able to use events outside of the prioritised 8 events to create website custom audiences (depending on consent being granted).
It is important to note that when an event is changed, it will trigger a 72 hour cool down period before campaigns can deliver using changed events. So we recommend you prioritise your event as soon as possible before the changes come into effect.
As people adopt iOS14, the sizes of app connection audiences and Mobile App Custom Audiences (MACA) will decrease. New iOS14 ad campaigns will not be able to use MACA for inclusion targeting, but may continue using them for exclusion, though these audiences will become less effective at excluding people on iOS14 over time. These new iOS14 campaigns will also no longer be able to use app connections targeting.
Facebook will however continue to support the ability to run app install campaigns using Mobile App Install Optimisation, App Event Optimisation, Value Optimisation, and App Install with Event Optimisation through Automated App Ads (AAA).
If you have previously used Value Optimisation, value sets will be automatically assigned based on historical data and turned on in Events Manager.
From a mobile-web point of view, we expect that Website Custom Audiences (WCA) will also get smaller due to the loss of some events from iOS14 users. Advertisers using Dynamic Ads for retargeting may also see performance and audience size decrease due to the loss of some events from iOS14 users.
Facebook will be deprecating both 28-day lookback windows and its 7-day view window. All data for these windows will remain accessible via the Ads Insights API. The default attribution windows are moving from 7-day click-through and 1-day view-through to 7-day click-through only for all conversions and catalogue sales objective campaigns.
To support the measurement of web events on iOS devices, Facebook will introduce Aggregated Event Measurement (AEM) once Apple requires the ATT prompt. AEM intends to help advertisers measure campaign performance whilst considering consumers’ decisions about their data.
Reporting, like any other reporting using SKAdNetwork, will be slower with installs delayed by a minimum of 24 hours. Reporting will be at the campaign-level only, and ad-level & ad set data will be modeled. There will be no more demographic breakdowns (age, gender, regional, placement, etc.) for app install or post-install metrics. However, you’ll still be able to see the demographic make up in the Facebook audience you serve ads to.
Facebook is also warning brands might see decreases in performance and audience size for retargeting efforts, but that prospecting efforts like lookalikes and Dynamic Ads for Broad Audiences will see minimal impact.
With the limited measurement capability of Apple’s SKAdNetwork API, we expect a drop in performance on all app optimisation campaigns, likely resulting in higher costs per action. Additionally, VO will transition from Ads Manager to Events Manager and value sets will need to be enabled.
Facebook might deprecate FAN as it becomes a much less workable product for advertisers and less valuable for publishers.
Facebook Login will not be available for matching users who have opted out of tracking on iOS14 devices. Facebook will soon be releasing an updated version of the Facebook SDK that will prevent a user from receiving targeted advertising and being accurately attributed by just using the Facebook Login.
Google is as yet less forthcoming with news on developments, but is starting to reveal its response. In late January it released this blog post addressing a few points. In summary:
Apple Search Ads will not be restricted to using SKAdNetwork. Instead, they will use Apple Ads Attribution for deterministic attribution. Some in the industry have complained that this might give Apple an advantage, and we are yet to see if Apple Ads Attribution will be made available on other channels.
There are clearly many things an app needs to change, even aside from implementing SKAdNetwork.
In December 2019, Apple introduced a new privacy information section for each app’s App Store listing page, called the ‘privacy nutrition label’. Apps are required to give people an overview of the developer’s privacy practices. The privacy nutrition labels give people information on how the app plans to use their data – including for tracking and attribution and for sharing data with other parties.
There is some flexibility in how an app presents the message asking for permission to track. There is no specific time an app must present the message, and it can be delayed, but know that you cannot access the device’s IDFA until you have.
The wording of the ATT prompt has come in for criticism. It is viewed as intentionally damaging to the likelihood of consent being granted.
The only part of the prompt that can be customised is the second part. In the example to the right that is the part that says “Use tracking”.
Additionally the background overlay around the white pop-up box can have its colour and opacity adapted.
With control over exactly when the ATT prompt is delivered to a user, app developers and marketers are considering a pre-ATT prompt message that details why their data is valuable with the aim of encouraging consent. Some are also thinking of offering incentives (free trial, in-game bonuses) if consent is given – although it is important to remember that product functionality cannot be withheld if consent is not granted.
The full impact of iOS14 will only be realised when Apple releases it out of beta. But there are things marketers can do to prepare.
If you need any help understanding how your mobile marketing may be affected, drop us a note at email@example.com.
162-168 Regent Street
+44 203 096 5108