To be Data-Driven is to be Decision-Driven

(Originally published on my LinkedIn page in September 2018.)

I had a great question from Dima Syrotkin, who asked about my key insight from chatting with Cassie Kozyrkov, Chief of Decision Science at Google, at the recent Google Cloud Conference in Helsinki, Finland.

I had several, but if I had to pick a favourite, maybe it was this. Needs context, though.

Experience cannot be measured. Behaviour can.

I like to say "Experience cannot be measured. Behaviour can." to shift the focus of designers on to behaviour. Now this is a bit of a trick principle, because in the end humans don't care about behaviours, they care about experiences. They always place the outcomes of what they do above what they do itself, funny as it sounds.

I've said, for example, that people are not interested in health - they are interested in its benefits. Health is a vanity metric. In the sense that if 'health' didn't help us get the benefits we expect from it, we'd lose interest in health quite soon!

Now, a key insight I got from Cassie is to be decision-driven in order evolve behavioural metrics (essentially objective) that can enable desirable human experiences (essentially subjective). This attests to the critical role of the human sciences in framing those decisional questions to begin with. (It also has deep implications for linking explicit to implicit signals, but that is another discussion.)

See this wonderful slide from Cassie's presentation at the Google Cloud for Finland launch.

Image credit: Slide by Cassie Kozyrkov

(Image Source for Bill Watterson comic : http://irrelevantvoice.blogspot.com/2006/07/happiness-10-cents.html)

This is the reason I've found solutions such as 'Happy or Not' to be inspiring or indicative, rather than actionable. When implemented poorly, such systems can even have the opposite effect, creating more confusion than clarity towards true customer insight.

Here's an example. If a bank is merely looking for 'feel good' evidence, a high positive percentage via 'Happy or Not' works well. But it doesn't reveal much about why different customers chose happy, or not, or any grade in-between the two. Now imagine if the same system was improved to support a decision on impacting wait times. The customer question would be "How was your wait time today - Happy or Not?" But in order to frame that question, the bank would have to do the work on figuring why wait times were critical to customers, vis-a-vis their other 'experiential' priorities.

A positive example of data that helps drive decisions would be the Net Promoter Score or NPS. It categorises respondents as Promoters, Passives, or Detractors. That is actionable insight.

In a world drowning in data, the human sciences can help at every step on the fascinating journey from behaviour to experience.

So the next time you find yourself commissioning data science work or probing data, ask yourself: what's actionable about this insight; what decisions am I seeking to drive from this exploration?