Today I want to share some thoughts around data-driven decisions. Rather than present some evangelical-sounding argument, I think it might help if I speak from my personal experience:
- I care about data because it is the only way I know if I’m getting better.
- It also helps me to think about what truly matters.
Seriously, I apply this to everything: from my work, to my workouts, to my gardening, to my…
Enough of that. Let’s take a real example: let’s say our goal is
“Make the company culture better”
The first obvious question is what does “better” mean? This is always worth a proper discussion. For our purposes now, let’s say we’ve agreed for our purposes it means that people enjoy being at work, and are productive.
So how would we know if we were changing that? That’s when it gets difficult, and very important, to measure: since then we can see what things (experiments) we do that impact this.
On the “people enjoying being at work” topic, we might look at some quantitative measures from People data such as “rate at which people leave the company” or “hiring ability”, together with some softer, qualitative, conversation-based data. We might even want to construct things, like a culture survey or exit interviews, to understand some specific aspects of what matters to us.
Then in terms of “productivity” (or some measure of value delivered), we might want to look at the customer benefit delivered each week, in £. This might be too hard (indeed I’m not sure anyone has cracked it ), so we’re looking at throughput (a count of stories done in a time interval) instead as a proxy-measure. If you have any ideas of how we could do that better I’d be really keen to hear them!
Following scientific method:
- First, we want a baseline: to capture what the current status of these things is.
- Then, we want to conduct smart, hypothesis-driven experiments, so we can see if these genuinely do impact things: being sure to take repeatable measures.
Unfortunately for our example, this interactive human behaviour is almost the definition of a Complex Adaptive System , so it won’t be easy to show causation – we’d need to do some Principle Component Analysis  (or similar) and that would require “long” and ideally also “wide” data, together with experiments and consciously-held-out subsets. In a business environment, this is hardly pragmatic.
We can though at least show that something important has changed, even if we can’t easily prove why it changed.
By gathering data that matters, you too can be data-driven! Experiment with this on something simple to start with, and see whether it works for you. It might be what time of day to commute to work, your own wellbeing, or the quality of your code. Let us know in the comments below what you try and how it goes!