Magazine article Information Today

On Small Data, Part 2: Thinking across Data

Magazine article Information Today

On Small Data, Part 2: Thinking across Data

Article excerpt

A funny thing happened on the way to Big Data: People began to realize they still had to think--there's no escaping that the messiness of data inputs is, well, messy. And, as we heard last month from the "father of machine learning," Michael Jordan (a University of California-Berkeley professor), the levels of quality of inference from one set of data input to the next are by no means related to the volume of input data. There are no error bars in Big Data outputs, as he so rightly observed.

We are, it appears, living in the very early days of data science, when humans are essentially training algorithms and hoping, more or less, for what frontier sharpshooters and modern-day snipers call "Kentucky windage": enough seat-of-the-pants guesswork to muddle things through, despite a crosswind or two affecting your work.

Targeted Ad Campaigns

This is not true in advertising, where the money is. In the cut and thrust of "moneyball" advertising spend strategies, probabilistic data interpretation is now effectively aggregated across multiple devices. That means tracking intent to a very high degree of probable utility. They know where you are and what you might be about to decide, based on correlated Big Data. How? With login data and geolocation coordinates, aggregated across multiple devices. It's the holy grail of retail surveillance--understanding human behaviors continuously.

What's the upshot? Big Data is growing more deterministic, and because the volumes of data points have become so large, the hits--for instance, the valuable correlates of "intending to vacation," "loves Nicaraguan coffee," and "worried about climate change" add up to an ad for ecotourism in Nicaragua's coffee highlands--don't have to be perfect, just close. Probabilistically, you'll generate enough wins to justify the ad placement by pitching the tagline, "Hug a Nicaraguan coffee tree this winter."

Or so the theory goes. The holy grail of 360 attribution is still a pipe dream, even for Facebook, Google, and Amazon. (They don't talk to each other--yet.) We're fine-tuning our algorithms all the time, but there's still a niggling problem for Big Data: the human factor.

Relevancy in Brand Storytelling

Human beings aren't consumer machines. Impulses, whims, poor choices, inspired choices, emotional roller coasters, and the sheer chaos of living an ordinary, difficult life are arguably big factors in a decision. Content marketers learned this lesson the hard way. Branded content--meaning storytelling based on brand attributes--makes terrific theoretical sense, but the threshold for audience attention is still great storytelling.

Heads up, library marketers (and municipal marketers too). It turns out that three things govern what people believe in brand storytelling: relevancy, authenticity, and provenance. Is what you're telling me relevant? Do I really care that your reference librarian is retiring? I may well do--and I may share the news on Facebook and retweet it on Twitter, but only if I care. Otherwise, it's all about you, not me. And that's the killer of all manner of brand communications: I don't care, and you can't make me. But hit on a relevancy, and things warm up (emotionally and otherwise) in brand storytelling, independent of data points.

Assuming your newly written library story is interesting on its face (a great big if), then it might be (perhaps) relevant to the person who is interacting with your media. But that story might be relevant and yet not resonate, because it doesn't pass the sniff test of being about me and what I want in a user experience with your library communication. If it smells funny, it's inauthentic. And there's no quicker turnoff of attention span than the inauthentic. Authenticity is the biggest editor on the web.

Hence, provenance: Where's this story coming from? Is it just about the library? Then why should I care? What's in it for me? …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.