This article originally appeared in the June edition of our quarterly magazine. To get the full experience, you can read the issue online here, or subscribe to receive a physical copy here.
Let’s call it the Cookie Problem. With the advent of digital advertising, marketers could suddenly track users around the internet, retargeting them with ads for products they had viewed but not bought, identifying them based on browsing habits and accessing a wealth of information that simply wasn’t available to them before.
Now, mobile offers the same sort of tracking, using device identifiers and similar methods, but the question has become how to join up these two ecosystems. This year, mobile will surpass desktop and laptop PCs in terms of the amount of time spent online by UK adults, having grown almost 500 per cent in the past four years. With such a profound shift towards mobile, marketers are asking how they can unify their efforts across the two increasingly equal channels, and also how to bring offline channels – TV, outdoor, print – into the mix.
“Existing marketers’ digital footprints and consumer marketing strategies need to adjust to a cross-device, people-centric world, where true user-level reach and frequency can be controlled across devices, and valuable marketer data can be used to inform messaging, not just in terms of content, but context,” says Gareth Davies, CEO of data intelligence company Adbrain.
The problem of cross-device attribution – being able to identify a user across desktop, mobile and offline activity – has become one of the major debates in mobile marketing, and is often held up as one of the primary obstacles in getting more advertisers and brands to embrace mobile as an advertising channel. In a survey by Nielsen, 30 per cent of brands said the ability to use the same metrics to evaluate reach on mobile that they use for offline channels would increase their use of mobile, while 55 per cent of advertisers are “doubtful or unconvinced of publishers’ targeting claims and about mobile advertising’s effectiveness”.
“It’s one of the most critical challenges faced by the industry,” says Chuck Moxley, chief marketing officer at mobile ad platform 4INFO. “Consumers are moving across screens constantly, but advertisers are playing catch up. Right now, big brands don’t feel confident in the available solutions, and it’s holding them back. Mobile is the Wild West with no standard currently for tying a mobile device to a household.”
[caption id="attachment_55021" align="alignright" width="200"] Gareth Davies, CEO of Adbrain[/caption]
A tale of two solutions
The answer to the problem of cross-channel attribution currently breaks down into two different camps: deterministic and probabilistic. They are the kind of terms that can sound very intimidating to someone coming to this area for the first time, but in truth, the difference between the two is relatively straightforward.
Deterministic methods rely on log-ins that transcend mobile and desktop, primarily Facebook and Google, though others such as Yahoo and eBay make claims for being included in this select group. The data that social networks gather on their users is channel agnostic – it doesn’t matter if you’re liking a post on desktop, laptop, tablet or mobile, the data all goes to the same place, and helps construct the same profile.
The power of deterministic platforms such as Facebook to link together digital and mobile advertising is the reason why we have seen the growth of social logins across various other sites. If a brand can connect its app to your Facebook profile, it opens up a whole wealth of information it can use to target you with more relevant messaging.
Facebook is by far the most popular platform for social logins, representing 45 per cent of all social login use, according to a report by identity management firm Janrain, while Google takes second place with 37 per cent. There’s a considerable drop to the next most represented social network, with Twitter used just five per cent of the time, while Yahoo, LinkedIn, Microsoft and various others scramble over the remainder.
No privacy in the walled garden
In many ways, deterministic is the easy solution for marketers looking for a cross-channel solution, particularly as Facebook and Google partner with big data firms like DataLogix and Axciom in an effort to not only bridge digital and mobile activity, but offline too, adding another layer of insight. But the deterministic method has big problems – most publishers don’t have the cash or scale to build consumer platforms the size of Facebook or Google.
“For most publishers, you are limited to a ‘walled garden’, limited to their registered user database and owned and operated properties, which significantly limits the total audience you can reach,” says 4INFO’s Moxley. “Furthermore, most have only an e-mail address from their registered users, and the match rate from e-mail to home address can be as low as 10 per cent, further limiting the size of the audience you can match to past-purchase or CRM data both for targeting or measurement.”
The relative monopoly that Facebook and Google hold on social login means that to make effective use of deterministic targeting on a wide scale, you need to partner with one or preferably both to get access to their data, adding an SDK into apps so that they can understand what users are browsing and where they’re going in order to target ads.
There are also privacy concerns surrounding deterministic targeting. Both users and regulators have questions about whether it’s acceptable for advertisers to track users across devices using personal information like usernames or addresses. Finally, while data gathered using social logins is supposedly useful across devices, some do face roadblocks in certain places. For example, Google struggles to serve ads on iOS devices, while Apple has issues with those who browse using Chrome.
Social logins are not the be-all and end-all of deterministic, however. Performance marketing technology company, Criteo, has developed its own deterministic solution which relies on customer IDs or other unique data points pooled by its advertiser clients. The company works with over 7,000 advertisers (and 10,000 publishers) and has invited its advertisers to pool these customer IDs and then share the pooled data, in order to identify the same user accessing the same website via multiple devices. This allows Criteo to gather user intent from one device and serve relevant advertising on another.
“Advertisers give us anonymised customer IDs which are encrypted via MD5 and sit in the cloud and are then turned into a key to allow matching across devices,” says Jon Buss, Criteo’s MD for Northern Europe. “The advertisers are pooling their data; they put in a small amount of unique insight and get out a huge amount of value.”
[caption id="attachment_55022" align="alignleft" width="200"] Nimeshh Patel, vice president for EMEA at Drawbridge[/caption]
Big data, bigger scope
So, if deterministic isn’t your solution of choice, what about probabilistic? Rather than relying on confirmed links between two devices, probabilistic methods (sometimes called statistical or predictive identification) use a wealth of first-, second- and third- party data along with algorithms and machine learning to predict links between two devices. Similar modelling methods are used in weather forecasting, pharmaceutical research and even in financial markets to predict trends, and grow more accurate every day.
“We process roughly 40bn ad requests on a daily basis," says Nimeshh Patel, vice president for EMEA at Drawbridge, which operates probably the best-known probabilistic platform. "In those ad requests, because we’re a programmatic platform, we see device IDs and cookie IDs, and we use big data to triangulate those. It could be something as simple as seeing two devices on the same wi-fi network and we can say there’s a statistical likelihood that they belong to the same user, but then we’ll wait and build up multiple observations to the point where we have a very high accuracy rate with the identities we build.”
The power of probabilistic is in its openness and scale. A probabilistic platform can take thousands of otherwise meaningless data points and construct them into a portrait of a consumer that can not only be used to track them across devices, but also, to give insights into their spending habits, demographic profile and interests. In addition, it doesn’t have the limitation of deterministic models, where the popularity of the social login forms a hard ceiling on what you can match.
However, as the name implies, matches on probabilistic platforms are only estimations, based on data, and most solutions only offer accuracy of around 85 per cent, while some are as low as 55 per cent, although these figures are always improving as algorithms and artificial intelligence grow more advanced.
In fact, Drawbridge recently collaborated with Nielsen to conduct a study of its platform and found that its model was 97.3 per cent accurate in its ability to find a relationship between two devices.
Drawbridge has also recently added offline into the attribution mix by partnering with third-party data providers and data onboarding platforms. Technology partners supporting the solution include Datalogix, which connects online ads to in-store sales; and LiveRamp, an Acxiom company that provides data onboarding services.
Meanwhile, AOL recently revealed that it had developed a probabilistic attribution solution covering exposure to ads not just on mobile and desktop, but on TV too. The AOL solution can also cope with the final purchase being made in store rather than via a digital channel. The AOL solution has been validated by research from Comscore as having a 93 per cent match rate, though when pressed by Mobile Marketing, AOL conceded that the match rate when offline exposure and offline purchasing were brought into the mix was much lower, at 40 per cent.
Black box solutions
Both forms of attribution have faced criticism for how they treat data, and the relatively opaque nature of how they match up devices. For deterministic attribution, brands are often forced to give up their hard-won CRM data to third parties like Facebook or Google for the matching process, which they have no access to. For probabilistic methods, agencies are often unclear on how matches are arrived at, and exactly how accurate they are.
“Marketers and agencies really need to get underneath the hood to understand what percentage of the audience is deterministic versus probabilistic, and for the portion that is derived using probabilistic modelling, exactly how they are matching mobile devices to users,” says Moxley. “Many, you will find, use a single data point – the last time they connected to a wi-fi network, for example – which can be horribly inaccurate.”
So, with both solutions flawed in some manner, is there any scope for the two to work together? Many probabilistic platforms already submit data to deterministic models in order to check how accurate their matches are, using the more concrete deterministic data as their yardstick. “If you’re a probabilistic partner, you can't not verify as to how accurate you are and how much coverage you have,” said Nimeshh Patel. “We have to do that. Everybody should be looking to do that.”
Beyond that, there are hybrid models that aim to combine both approaches into a single identity solution, but these lead to problems of their own, with the potential for such platforms to cause even more privacy headaches than deterministic data on its own.
“The enhancement of deterministic data with a probabilistic solution has the opportunity to significantly increase reach and scale,” says Adbrain’s Davies. “We could see a world where the tech giants open-source their IDs, but it makes little sense for these players to allow their valuable user data to leak out or be shared with others unless ads are being bought or sold via their platforms, so expect any such solution to be more ‘data in’ than ‘data out’, meaning probabilistic solutions are critical to augment and enrich marketers’ existing data.”
He who clicks last, wins
Cross-channel attribution has the potential to shake up the advertising industry in two different ways. The first is the Holy Grail of marketing – the closed loop, where marketers are able to track every interaction a consumer has with a brand, linking up initial brand messaging, retargeting and conversion across digital and physical spaces to give a complete portrait of the customer journey. While there are several barriers to this, including the difficulty of integrating bricks-and-mortar stores into this ecosystem, these should be overcome with the help of developments like mobile payment and location tracking.
“Cross-channel attribution is the most transformative opportunity in marketing today,” says Davies. “Thanks to the digitalisation of offline purchasing I believe that within the next five years marketers will be able to measure the complete path to purchase, bridging both online and offline. For this to happen, we’ll see closely integrated technology stacks whereby ad tech companies bridge proprietary IDs with a brand’s CRM system, effectively connecting online with the offline world.”
The second fundamental change that cross-channel tracking and identification could bring is an end to last-click attribution. This is where the benefit for converting the user to a sale is given to the last ad or search result they clicked on before they made the purchase, even if they had carried out weeks or months of research across various devices, clicking on a variety of search results in the process.
While consumers are getting more and more comfortable with making purchases via smartphones, many still prefer to research on mobile then make the actual purchase on desktop. With a more accurate picture of how a consumer is interacting with a brand across devices, the time spent on mobile can be recognised as equally important in influencing purchase decisions, and rewarded as such.
“If you look at who wins, on a last click basis, it’s the people who show an ad to a user on a very high level of frequency,” says Drawbridge’s Patel. “What cross-device does is make the whole thing far more efficient, because for the first time, you can offer a service that says we won’t do that. We’ll allow you to follow that user from desktop to mobile, and we’ll allow you to frequency cap across devices.
“Attribution models, and the way in which people are compensated, need to become far more attuned to protecting a brand’s reputation, and part of the answer to that is identity solutions that enable that frequency-capping across devices. You protect brand reputation, and marketers become far more efficient with their spend.”
A cross-device future
As mobile becomes more and more central to consumers’ purchasing habits, cross-device attribution will move from an attractive proposition that can aid in retargeting to an essential component of every marketer’s toolkit. The question isn’t when it will be implemented, but what method will become dominant, and what changes it will go on to drive.
“In the end, marketers need to make sure they and their agencies are aligned – from both a strategy and compensation standpoint – to ensure they are demanding that their cross-channel campaigns are measured based on what really matters: actual sales lift at the cash register,” says 4INFO’s Moxley.
“More and more CEOs are expecting and demanding hard return on ad spend (ROAS) metrics to justify further advertising expenditures in digital media, and the era of measuring clicks, page views and even store visits will come to a close. The future will be about measuring cross-channel campaign ROAS based on actual sales lift, regardless of where the transaction occurs.”
What cross-device attribution really represents is the realisation of mobile’s initial promise – a truly complete marketing ecosystem that uplifts every aspect of the industry that it touches upon. However it is implemented, and whoever it is that finally cracks it, a complete attribution solution is likely to transform the marketing world in a profound way.