 |
| All Hail The Criteo Overlords |
In a recent interview on NPR's Fresh Air, Bill McKibben spoke about his latest book, where he outlined threats to humanity. (Mostly, spoiler alert, climate change.) At the tail end of the talk, McKibben also noted the threat posed by Artificial Intelligence (AI), loosely defined as computers making decisions based on a virtuous learning loop built on data accumulation and analysis.
McKibben did so with a fairly famous thought exercise known as the Paperclip Problem, or to get more high faluting, instrumental convergence. In this, an AI robot with a seemingly good but unbounded mission (say, the most efficient manufacture of paperclips) would quickly move to terminate its human masters, since they would likely shut off the robot at some future point, and thereby prevent paperclips from being made.
Now, at this point, you might be wondering how we're getting to digital advertising challenges. And with that, I give you the Criteo Box, which is a term that some in our field use to describe template retargeting ads made (in)famous by the dominant player in the space, Criteo. (Example above.)
Criteo boxes are loathed by many design and brand marketing professionals, because they are machine and data driven utilitarian shopping bots that seem to eliminate the need for design. The challenge becomes all about the dynamic product recommendations shown in the ads, because by whatever analytic standard is being used to determine good ads from bad, the data has driven you to this, the final plateau of performance.
An inelegant bare bones box with as many recs as you can fit, Because Data.
Which might lead you to think that design doesn't matter, because it's been solved by AI. Like betting that you will win in chess against IBM's Watson, it's a losing proposition. Just accept the box and move on, with the small possible caveat that it's only solved for remarketing and not acquisition. (But will also likely be solved at some point for acqusition, again, Because Data.)
But here's where I'd like to hold out hope for humanity's continued presence in my life's work, while still being OK with analytics. My belief is that the Criteo Box is only dominant due to an over-reliance on short term goal events.
If you are judging only by clicks, an ad with multiple entry points and good dynamic SKUs might always win over something more brand related. (Side note: please don't use clicks as your goal event, as it's really a bad idea due to bad actors and fat fingering on mobile, and it's not 2001. Tangent over.)
But what if you were looking at, say, purchases? Or the lifetime value of the consumer? Or the margins driven from that value? Or...
Well, you get the point.
The reason why we don't judge ads by these longer funnel approaches is because no advertiser is going to run just display ads. They are also going to follow up with email, have a social and native presence, perform work in search engine optimization for paid and native, and upsell the user on site. All of which will have impact on the performance of the ads, and possibly not an equal one.
In addition, advertisers are going to rise or fall based on customer service, their offline presence, print and broadcast and podcast and heaven knows what else. (Oh, and a side note? Advertisers don't exist in a vacuum without competition, and if everyone in your space is making nothing but Criteo Boxes, your non-Criteo Box ad is likely going to stand out. And, perhaps, perform better.)
Because life is about a lot more than paperclips. Strong performance practices are rarely so cut and dried as to be about a single factor or a single metric. Things that you think you know probably need to be re-tested, and re-thought, rather than assumed to be settled law.
People who design ads without consulting the data are, I believe, acting in an irresponsible manner to their clients.
But so are the ones that act only from data, rather than be inspired by it.
(Also, beating Criteo Box controls? Not a new trick for me. Reach out and let's talk.)
No comments:
Post a Comment