A dark wave of dismay is sweeping across the data-driven marketing community as it contemplates a future deprived of behavioral data needed to train its algorithms for super-effective marketing automation. The collective dream of right-person-right-message-right-time-every-time seems to be fading into a hangover of broken cookies and declined tracking prompts. “We were just getting started,” bemoaned a client recently, “and now it looks like all our investments in machine learning and data science are up for review.”
“I guess we’ll just be relying on Google, Facebook and Amazon for data and analytics from here on out,” sighed another.
Two bright spots appear through the dread. First, the impending loss of data has accelerated many machine learning initiatives to get in as much training as possible ahead of the drought. If DNNs can find enough patterns in the existing data, maybe they’ll be able to still make useful predictions with much less data. Perhaps they’ll also be able to learn how best to persuade customers to create accounts and opt in to communication features that might reduce dependencies on targeted and retargeted ads.
Second—sorry for burying the lead here—some are realizing their data flows won’t actually be diminishing at all—quite the contrary! For many, the sources and nature of data is merely shifting. Loss of cookies and device IDs is being offset by sharp growth in digital engagement (due in large part to the lasting effects of lock-downs), resulting in a surge of in-session observational data. Meanwhile, the technology of AI inferencing is migrating from the cloud to the edge with intriguing implications.
Much of this is happening in the IoT world. A recent Gartner survey of IT organizations implementing IoT (subscription required) indicated that 73% are running at least some AI inferencing at the edge of the network: in cameras, smartphones, consumer equipment such as smart TVs, smart speakers, and so forth. These new streams of data might not deliver lists of qualified leads, but they will help marketers understand how consumers use products, reveal new insights and open new channels for communication and new digital business opportunities like usage-based subscriptions.
Back on the web, edge migration is shifting the ways AI might be deployed in service to marketing. Google’s FLoC is the prime example: by moving AI from the cloud into the browser, its federated learning model has access to far more browsing behavior than even google.com, as it addresses privacy by only revealing the results of its learning in aggregate form. The lesson is clear: there’s lots of data at the edge to learn from, and if you can put your AI there you don’t need tracking to make real-time decisions.
Unfortunately, this opportunity is neither evenly distributed nor universally available. Most consumer edge products are under the control of the Four and brands that lack direct consumer relationships are still mostly boxed out. For years consumer product makers have experimented with connected appliances and merchandise with mixed results. Even when adopted, use cases for marketing based on edge data can be challenging.
Does this mean marketers should ignore the possibilities of data and AI inferencing at the edge and go back to the five stages of grief? I don’t think so. New, privacy-inspired forms of data collaboration such as data clean rooms and decentralized ecosystems are opening up new possibilities for data-driven marketers to explore while avoiding the pitfalls of privacy and manipulation. I’m looking forward to presenting more details on this at Gartner’s upcoming Marketing Symposium in August. Hope to see you there!