This is How You Can Build a Better Data-Driven Organization: by Reducing Data
The Real Pros Do More With Less
There’s this thing we do with social media that gives us small hits of dopamine when it works well. You know the feeling — we all chase it these days — you post and like so you can feel good. We sometimes make the same mistakes with data, building tools that give us that same flush of excitement, but don’t lead us anywhere useful.
A data-driven organization is an aligned organization, an informed one, one that values awareness and functionality and is willing to pay the price to have these things. A data-obsessed organization can be one that never learned to get off the dopamine high, even when things are going well. Dopamine’s great, but so is getting work done, which isn’t always as fun. I want to be data-empowered, not data-obsessed.
Take the advantage, take the wins data gives you, but don’t do it in a way that’s going to distract you from the real work.
How do you do this?
Start with a picture of normal. If normal is happening, even if normal is exciting, leave it alone. Let normal look like a dashboard on the wall that gets updated daily. That was not a flippant comment — daily is good enough for normal. If you have to be reassured what normal looks like this very second, then you’ve either got dopamine addiction problems or more-difficult ones. See if you can reduce your dependence on trivial things.
Normal isn’t quite one thing, and it changes as your organization changes. Pick things that show you what you need to know. Figure out if sales and marketing are working as expected. Figure out if your system is running. Figure out if sales and profits are working. Achieve things with data by your side. See if you can make every metric lead to action if it’s high or low.
Each new day has a different shape to it. Just roll with it.
— Ben Zobrist
Once you have an idea of normal, become conversational in it. Everyone needs to speak about normal — ideally with the same terms. Get the conversation going and see if people use the same name for the same thing. Put the name of a metric prominently on your reports and visualizations. Pick simple and clear terms. Be a stickler if meaning drifts. Be polite, but it’s your job to create clarity and confidence. You can’t do that if people feel confused when they want to talk about the company. As a kinder Mr. T might have said, “pity the poor fools,” by encouraging them to succeed with your data.
If it’s not normal, it’s abnormal. Funny how that works.
The thing that doesn’t fit is the thing that’s the most interesting: the part that doesn’t go according to what you expected.
— Richard Feynman
You want to know about abnormal quickly. If you turn down your data updates, you have to stick some monitors in the mix that can raise the alarm.
Think about this a little when you build your information systems. Say everyone gets a fresh batch of data on their wall in the morning. The system that built it can’t entirely be turned off during the day. The data has to still trickle in somewhere, just not to a distracting place.
That means you’ve got to use transform functions that can work on single records. That means you’ve got to work with streams, even slow-moving ones. That means you choose online versions of your core metrics when you can. (An online algorithm is a fancy way of saying you can produce an answer after every record instead of after you’ve seen all of the records. For example, this is how you calculate mean with an online algorithm:
(‘previous mean’ * ‘(count -1)’) + ‘new value’) / ‘count’
It turns out this isn’t vapid advice; this is difficult and important. You’re making people smarter by building better habits with your data while preserving the right to use up-to-the-minute techniques when you need to. You might find yourself working harder to design the system or working harder to calculate metrics. That’s the job too.
What if the world’s not as simple as I’m making it sound? It never is. Data work can be hard work. It’s hard because it takes leadership to decide to engage or not.
Once, I was called out of a movie theater by the CEO of a company to address a problem. “The system is down, we’re losing millions of dollars, and it’s your fault.” I left my girlfriend in the show, drove to the nearest laptop, and got to work. The system was down, but the CEO’s son had caused it, and someone had fixed it by the time I got to a laptop.
You’ve got to have the right calls to action, even if they don’t always turn out to be emergencies. Create too many, and you erode confidence and commitment. Create too few, and you’re letting your organization fall apart. There is no rule of thumb here, so lean in and lead. I don’t know if the CEO was wrong that day. It was inconvenient, but it might have been the right call. You’ll make decisions, and you’ll learn from them.
Your job as a data person is to create confidence. That is more important than putting people on the right information diet. Use reliable data. Explain your conclusions clearly. Make decisions. Choose data visualizations for their clarity and simplicity. Choose metrics for their clarity and power to incite action. Choose tools you can learn and understand.
Once you’ve built confidence, once you know what normal is, then you can simplify and slow the flow.
As long as you keep going, you’ll keep getting better. And as you get better, you gain more confidence. That alone is success
— Tamara Taylor
So that’s the gist. Build great insights, but share them in a stingy way. It’s not the popular thing to do, and it’s not always the possible course of action — at first. Keep in mind that driving with data means driving intelligent behavior, building strategic advantages that don’t come as cheaply as a data pipeline, model, and dashboard. Sometimes we have to put people in rehab so they can detox and live up to their potential.