In a war, for every yard of ground gained by one side, a yard is lost by the other. Yesterday, it was Apple, fighting for user experience and against those shoes you didn’t buy that follow you around the internet, that gained; and Criteo, the foremost provider of those ads, that lost. Writing exclusively for ExchangeWire, Duncan Arthur, managing director, Illuma, explains why Apple’s Intelligent Tracking Prevention (ITP) is actually a good thing for advertising.
Apple updated the Safari browser to reject the tracking ‘cookies’ that enable third parties to track users from site to site. Though Safari accounts for less than 4% of browser use, Criteo lost 27% of its market value.
Apple can do this because they don’t have an advertising business. Chrome, by far the dominant browser, owned by Google, is also making changes to make tracking by, presumably, anyone other than Google more difficult.
Criteo and others are developing what they call ‘workarounds’, but with no sense of whether they will work. But they are trying to get back to the status quo and the status quo isn’t all that great.
For all the talk of how data is informing advertising through ‘machine learning’, most of the automation that underlies ad tech is only capable of very narrow pattern recognition, such as Criteo’s model of observing you looking at a car site and then for weeks serving you car ads wherever you go.
Advertising would do better to do real learning about the person on the other side of the ad server. Both because tracking is about to get much more technically difficult as the browsers elevate user experience, and because of upcoming privacy regulations.
Data’s bigness is separate from its quality
Advertisers will spend big money offline to do audience segmentation work, which offers deep insight about people who buy their product. That is entirely valid until they to make that insight actionable for advertising. Whereupon they discover that almost none of the details their research discovered about your audience can be translated into terms that available 3rd party audience segments can mirror. And even where a DSP shows a segment name that sounds like your consumer, the data inside is likely old, shallow, and inaccurate.
The best data companies have is their own – CRM data gathered from current customers. But depending on this to make media decisions is more of a stop-loss strategy than a growth strategy. Just like retargeting based on site visits or shopping cart abandons, they end up advertising only to people who already know them – and in the case of retargeting, spending money to reach people who may have actively decided not to buy.
One can only get so much growth out of marketing to existing customers (well, really, no growth except for the rare business that can raise prices without losing customers). So advertisers fall back on the old standards, based on very limited machine observation rather than learning, which they must know lead them down a path of diminishing returns.
Growth comes from learning
If advertising changed its mindset from simply being a delivery mechanism to also being a learning process we would be more effective. What do I mean by that?
Developing reach intelligently requires a system that understands both the advertiser’s message and the context in which people encounter the ad. Certain messages may do better in certain contexts than others.
To understand subject matter, it’s not enough to parse keywords, the machinery must be smart enough to understand the dimensionality of the content and then must learn whether it helps or hurts a range of ad messages. This requires a philosophy of constant experimentation.
One must also recognise that effective contexts change all the time. It’s not only the subject matter, but the cultural context of the person’s life, which can change moment to moment as we are buffeted by a constant stream of social media zeitgeist. So it’s not a matter of landing on a handful of effective contexts that can then be applied repeatedly, it’s about constant testing and adjusting to what is effective right now.
So, the limitations browsers are placing on advertising, for all the rending of garments and gyrations of stock prices they will cause, are actually good for advertising. It takes discomfort to move out of the tried (but not true) and towards something better. In this case, out of the world of pattern recognition and lookalike modelling and into a much more effective world of machines actually learning about people.