Qualitative Data – the Other Stuff

We’ve become obsessed with big data and analyzing it, though sometimes I think we get in our own way. The issue is that we preferentially collect quantitative data as if it were the only thing worth the time. In fact, quantitative data is but one kind of data, and the information it provides gives a one-dimensional view of the world. It’s not wrong information — just incomplete.

The obsession with quantitative data that can be stored in spreadsheets and worked on with heavy duty math reminds me of what the economist Sumantra Ghoshal once termed “physics envy,” because he believed that economists were too fixated on turning economics, an inherently social science, into a hard science.

Do you know how many kinds of data exist? There are two kinds of quantitative data: the aptly named “quantitative” type that we’re all familiar with and that we can do math on; and “interval” data, which might have numbers — but upon which performing math makes no sense. What’s the square root of your phone number? You get this.

There are two kinds of qualitative data too. Hair or eye colors are examples of data that has no order — blond and blue-eyed might be your preferences, but there’s nothing ordinal about them. There’s no ranking for them, so they are termed “nominal.” On the other hand, letter grades are ordinal, because they have an order. A digital photo file also can be considered qualitative data that is ordinal in that the order of the data’s rendering, i.e. transformation into a picture, matters quite a bit.

It’s an Open-Ended Universe

So the data universe is generally larger than what we collect, and consequently we leave out important potential input when we collect only quantitative data. So, for instance, in an extreme situation a disaffected customer might give a 9 or 10 on a net promoter score, indicating a propensity to recommend a brand. However, the intent actually might be to report bad service to others.

More realistically, a customer might have feelings about a vendor or a brand that don’t get captured but could be instructive. That’s why asking open-ended questions about feelings and emotions is so important — especially if you’re looking past the current transaction and trying to capture lifetime value.

You capture qualitative data in a variety of ways, but often through things like communities — and this is the rub. It’s a Hobson’s choice between gathering a limited set of quantitative data in an automated tool like a survey, or capturing customer input that needs to be massaged into meaning. Qualitative data is more costly.

Automation is getting to be pretty good, though, and machine learning is making the most of the limited data it collects. There’s a post on Harvard Business Review that illustrates my point. In “How to Make Your Company Machine Learning Ready,” James Hobson takes us through a short primer on how machine learning can help your organization, making four points with which I heartily agree:

1. Catalogue your business processes.
2. Focus on simple problems.
3. Don’t use machine learning where standard business logic will suffice.
4. If a process is complicated, use machine learning to create decision support systems.

All of this is eminently practical, so what’s not to like? Well, it seems rather high-level, in that the focus is on “simple problems,” but how often do we deal with simple problems?

Where We Shine

I kept reading this piece waiting for a conclusion that it would be easier to just ask the customer or the employee or some human with direct knowledge of a predicament. The alternative is to buy software, rent hardware, hire data scientists and develop algorithms, but nobody looks at those costs.

Alternatively, sometimes just asking the customer — either directly or through a plethora of community activities available at low cost — gets the answer faster and reduces the error inherent in interpreting what the customer thought a survey actually meant.

I like the idea behind No. 4 above. Decision support is where ML can do the most good, at least today. If it’s backed up by actual research into how customers think, you can position your ML-based decision support system at the point where customers run astray.

There’s one other overlooked benefit of gathering qualitative data. The act of asking, however you do it, is an attempt at customer engagement. People love to be asked what they think or feel, and the effort has been known to build bonds that result in customer loyalty and predictable revenue increases.

That’s the perfect synergy of machine automation and human intuition. People might not be as good as automation at capturing and recalling facts and figures, but we excel at social interaction, which is all about teasing apart qualitative data and being in the moment with another party.