With Google’s Topics announcement, the digital marketing industry has now started a new year much as we started off 2021: focused on the future of identity. We are facing the threat of a broken model for targeting and measuring ad campaigns. Data shows that relying on users to opt in to share their data simply will not yield sufficient insights at scale to keep the model from breaking. 

Digital ad targeting was built on the ability to target efficiently and measure ROI across sites and apps, greatly assisted by the availability of third-party user data, third-party cookies and mobile IDs. But, next year, Google will stop using third-party cookies to track users in Chrome, unless they opt in to sharing data. Apple has greatly reduced the number of mobile identifiers that can be used by advertisers — again, unless the user intentionally opts in. The industry has been looking at first-party data as a winning bet for individual businesses and to augment alternative identifiers — but you need an unrealistic amount of opt-ins to make up the gap we’re losing in traditional identifiers. 

This is a problem. Attribution, lift studies, retargeting, the ability to build lookalike audiences — these processes and more are dependent on a scale of deterministic data we are unlikely to ever see again. We need to re-invent our approach to targeting and measurement and to recognize the centrality of machine learning to solving these problems. 

Marketers are faced with a crowded field of would-be solutions. There are proposed universal identifiers, including UID 2.0. There are numerous webinars and blog posts laying out strategies for maximizing first-party data. Investors are looking at new models for data clean rooms to match these data sets. But the core problems remain: siloed data, siloed identifiers, and a face-off between walled gardens and the open web. And the biggest hurdle of all, of course, is scale. Simply put, people just will not opt in at the same rate in which they historically have not opted out.

The proposed solutions depend on volumes of first-party and opt-in data that are meaningful and actionable. You need a hashed email for UID. You need registered log-ins to collect first-party data. A shift from default to intentional opt-ins is essentially a shift from a data free-for-all to relying on data donations. Years ago when telcos ran opt-in programs for subscribers, 20-30% opt-in rates were considered an amazing achievement. But, at those rates in our current system of identifiers, the model breaks.

We have proof of how challenging it will be to rely on user opt-ins. We’ve heard about how losing IDFA data is impacting mobile campaigns, but Apple didn’t even “deprecate” IDFAs: it just switched to an opt-in model. At the beginning of June 2021 — barely a month after Apple switched to opt-in IDFA tracking — 68% of Apple programmatic inventory had IDFAs, and by the end of the month, it was down to 35%, where the share stabilized, according to proprietary Emodo analysis.

This is not sustainable. What many in the industry are hoping for from alternative identifiers and first-party data is to make up for the loss in traditional IDs. So, of the 33% we lost — in going from 68% to 35% — how much can we get back via different methods. Unfortunately, there is no realistic way to make up for the lost data. It’s a pipe dream. Ultimately, the industry stands to lose loads of categories and cohorts used in targeting and to face close scrutiny over compliance with privacy standards.

On the whole, new identifiers and thorough first-party data strategies will be significant elements of digital advertising’s future. But there’s another major element, which can provide the scale and efficiency the industry is accustomed to: machine learning (ML) and artificial intelligence (AI). 

Smaller, high-quality, first-party and opt-in data sets are not ends unto themselves — they are the basis for training machine learning models. Through the combination of first-party and contextual data, including metadata about when and how an ad slot is available, well-trained ML models can predict things like the segment in which the user likely belongs, or the user’s likelihood to take action or convert. And ML has proven to be as efficient as deterministic targeting at similar levels of scale as we’re used to – while improving location accuracy. No identifier is required to train these models.

ML can help make up the difference from the identifiers the industry is losing. And it starts with the best training data. Businesses with the best — not just the most — data will have the advantage. But most businesses shouldn’t expect their own data will be sufficient. Instead, they should look to their digital business partners. Publishers and platforms need to share data resources. It’s important to explore alternative data sources, as well, especially now that AI makes quality, not quantity, the primary factor. For instance, mobile carriers are uniquely able to generate opt-ins via their strong and direct relationships with consumers; this type of partnership might make sense for mobile advertisers looking for accuracy and scale. In the end, companies need to consider the types of data, and their actual value, for use as training data – not just the volume of available data.

The digital industry is short on time to “solve” identity, but there’s no flip-the-switch solution. The future won’t be a mirror image of the past. It’s time to add value from technology, as we lose value from data. Machine learning will see a boom this year as it provides efficiency and reach. As any advertiser or publisher takes ownership of their first-party data, they must consider its role not just in targeting, but in training the algorithms that will help campaigns meet goals in the future.

Alistair Goodman is the CEO of Ericsson Emodo.