The New York Times ran an interesting piece on the opportunities and risks of data science yesterday. One mention was that IBM, which has been working to integrate its Watson big-data “artificial intelligence” into healthcare, has started to offer Watson Paths, a software program that allows doctors to see “the underlying evidence and inference paths Watson took in making a recommendation.” Along similar lines, it discussed the case for integrating a human element into big data-driven decisions in order to double-check the numbers and try to understand the processes used.
The volume of content online is often such that we have to use data analysis tools to find timely answers to any number of questions – who are the influential figures in your market? Which consumers matter? How should I tailor my product and my messaging to speak to these people? There are tools from Sysomos to Brandwatch, Klout to Traackr, Twitonomy to Google Analytics, and more – all to help you drive your marketing and PR decisions.
Some of these give you access to the raw data they work with. For example, if you pull social content from Sysomos, you’ll have the chance to review all the posts yourself to ensure that they’re vaguely on track (though you’ll still miss whatever’s in the database). Others don’t. If I search for influencers on Traackr, I don’t really know what goes into their database and their algorithms.
The old axiom of Garbage In, Garbage Out still applies
So when working with tools, brands and agencies need to think about what level of granularity they’re going to reach with the data, and to what extent they’re going to trust the tools to get the right answer on their own.
Many are basically black-boxes – and there’s a lot of inaccuracy in there. If you’ve ever tried automated social media sentiment analysis, you’ll see this in action. Where “I hate when I miss my [favorite brand] coffee in the morning” gets tracked as “Negative” for [favorite brand]. People are complicated, and it’s hard to measure the right things without knowing what’s being measured.
Transparency helps, when possible. Perhaps there should be more Watson Paths-style aids for marketers that show users how their tools are coming to their conclusions. It’s much easier to test different outcomes online on a micro scale than it is to make healthcare decisions. Yet broad campaign strategies and approaches still have a lot of momentum and funding invested in them, so in 21st century PR and marketing, we need to ensure we’re working with good data, not just any data.
Image is from the author’s collection.