In 2009, my good friend Tom Foremski said that all companies need to be media companies. He was really talking about the need for companies to be great storytellers rather than relying upon the marketers pitch as sanitized by the legal department.
That idea makes a lot of sense to someone who is steeped in media. But by and large, it has been much harder to get companies onto that same way of thinking. Regardless, and with a tsunami of technology competing for marketing dollars, one of the catchphrases that certainly resonated is that of ‘personalized experiences’ in the context of available and deliverable media.
As night follows day, analysts of all persuasions found a comfortable area into which they could pitch all manner of fluffy stuff. ‘Personalized experiences’ fits neatly into certain goods like car buying, where the options available to the individual and presented online mean you can ‘build’ your dream car relatively easily. The same goes for other goods, but what about services, where the key requirement is information?
Here there are numerous problems and despite what vendors will tell you about the wonders of some AI related topics like machine/deep learning, the reality is that the state of the art is nowhere near good enough, except to meet the needs of a very tiny fraction of an available market. I know. I’ve failed in that regard this last year but learned a great deal along the way.
Let’s start with a basic premise: in order to personalize any content to any degree, the system needs to either present you with a shit ton of options which are then acted upon in the future or, it has to find a way of understanding your preferences through your activity stream. In June, 2015, Jarno Koponen wrote that:
Personalization promises to modify your digital experience based on your personal interests and preferences. Simultaneously, personalization is used to shape you, to influence you and guide your everyday choices and actions. Inaccessible and incomprehensible algorithms make autonomous decisions on your behalf. They reduce the amount of visible choices, thus restricting your personal agency.
Because of the personalization gaps and internal paradox, personalization remains unfulfilling and incomplete. It leaves us with a feeling that it serves someone else’s interests better than our own.
For me, a big part of that problem comes from what Koponen describes as the ‘data gap,’ where there just isn’t enough information about your or my interests to allow for a good personalized fit to whatever content is available. In theory we should see patterns of activity emerge that provide enough clues such that algorithms can infer what a person is interested in. Some vendors promise this to be true, even for anonymous visitors who eventually convert to something.That only works to a point. Take music as an example. The fact I recently played a string of Depeche Mode songs doesn’t necessarily mean I am that interested in Kraftwerk, or, for that matter, want more Depeche at that point in time. But, I might be interested in the next available Depeche concert in my area. Or I might not. You get the problem…it’s called being human and to the best of my knowledge, compute resources are not going to acquire those sentient skills that predict my next behavior. At least not in my lifetime.
More important, the claims of magically taking anonymous data and then converting that to a personalized experience upon receipt of an email address is, quite frankly, fanciful. Dynamic IP addressing and IP addresses that are fixed to a telco will nix that idea straightaway.
Where does this leave us?
What we are rapidly discovering is that in an attention economy where attention is at a premium, the best you can likely expect is to ‘own’ a person’s attention for a limited period except in very exceptional circumstances or, where the niche you live in is sufficiently fluid, that people have to pay attention for long periods of time.
Does that mean personalization is a total waste?
In an ideal world, automated personalization engines designed to drive content should provide enough allure to keep your visitors coming back. At least for a limited period and/or sporadically. Based upon what we’ve seen, the engines only kind of achieve that and only part of the time. To some extent, the results almost appear random but it would be a mistake to fall into the trap of assuming that just because a pattern isn’t emerging in an obvious manner, that it doesn’t exist.
Where I see this as falling apart is in the personalized greeting designed to encourage information consumption. So far, we have not found any system that can calculate what needs to be said and then semantically represent that onto the page. Instead, the current crop of automation engines default to boilerplate, augmented by what the algorithm discovers has been the most popular content. This is especially irritating when content is distributed via email, that tried, trusted and far-from-dead method of engagement. Why?
I sense that most people hate the boilerplate email header because it smacks of yet-another-newsletter that I might, or might not choose to read, regardless of my interest level. That seems to be true for at least 80% of those with whom we already regularly interact. It’s not that they don’t necessarily open the email – there we can do extraordinarily well. It’s that the click rates are often shockingly low.
You can argue that if content is insufficiently attractive then people won’t click. Or, you can argue there’s enough content in the email that people don’t need to click. Either way, there’s a question mark over ambiguous activity that is hard to solve. And that’s before you start wondering about how people’s consumption methods change over time, whether mandated by their work environment or simply because they found a better way to consume content or because their interests changed. Or, for that matter whether your content consumers can readily fit into neat categories…which they mostly can’t.
Factor in changing needs and it is hardly surprising to discover that even with the best laid plans, delivering content to an otherwise loyal readership requires a significant degree of hand crafted work. I know this as well because based upon a series of experiments I’ve been running since Easter, I’ve found that I can consistently outrun or at least equal the results of automated systems.
Paradoxically, the technologies we’re using have helped me figure that out while, at the same time figuring where the strengths and weaknesses lay in our approach to content. It has come at a not inconsiderable cost but then I am less concerned about that aspect than the learnings achieved along the way.
I guess then this is a case where the robots ain’t going to be winning out any time soon.
Image credit - © Tatiana Shepeleva - Fotolia.com