Strong cultures are designed while weak cultures emerge. If you want to support data-driven decision making with a strong data culture, it has to be by design.
Humans have evolved as communal thinkers. From an evolutionary perspective, it is the thing that sets humans apart because we are able to share knowledge and cooperate both flexibly and in very large numbers. Our brains evolved to specialize, which then allowed each of us to develop our own talents. Paradoxically, this made others appear more foreign so we developed rituals, metaphors, ceremonies, stories and catchphrases. These, in turn, are what allow strangers to cohere.
We experience this as culture. Culture exists so we can access others’ knowledge. Culture tells us where to put our own knowledge. It is the mechanism by which we build a community of mind.
Culture as community; community as knowledge
There’s a simple way to show how our knowledge comes from a community of mind. Try this exercise:
- How well do you think you know how a toilet works? 1=not at all, 10=very well
- Describe how you think it works
- Revise your assessment. How well do you think you know how it works now? 1=not at all, 10=very well
In numerous studies, people give a higher answer to 1 than to 3, by a significant margin.
What’s going on here? As Steven Sloman and Philip Fernbach explain in The Knowledge Illusion, human knowledge is communal which gives us a sense that we know more than we do. When we are asked to describe how something works, the act of deliberation punctures the illusion and we realize that we know less than we thought. This is a powerful insight because it helps us understand that we need others in order to have a place for our own knowledge and for us to access knowledge that we might feel we already have. This makes us dependent on culture—having a sense of belonging, shared intent and vulnerability. Ultimately we feel safe when we belong.
The key to culture is being able to understand others’ intent. We help each other out by sharing our knowledge which, in turn, means that we have evolved to rely on shared knowledge. The problem is that we fail to distinguish the knowledge that’s in our own heads from knowledge in someone else’s head. Since thinking is done as communities—making us cognitive team players—it doesn’t matter who knows it, it just matters that someone does. This means that access matters.
Machines as part of our community
So what happens when machines and data are part of decision-making? Up until very recently, machines were not part of our community. But with advances in AI, machines are increasingly able to discern our intent and to personalize their knowledge and actions towards us. The case can be made that machines—and their data resources—are becoming part of our community. In fact, people also feel they understand something when they believe that a machine has the knowledge too, as long as they believe it’s accessible to them.
The challenge with big data and machine learning is that the base unit of evidence for our knowledge is now data that is only machine-readable. Machines hold knowledge that humans may be unable to access because its scale, speed and scope are beyond our intuitions. And data lacks the core tenets of human culture: data has no vulnerability, no belonging, no inherent sense of purpose. Technology doesn’t share our intentionality; it doesn’t share our common goals. The machine can’t empathize with us and truly understand what we want or what we’re looking for. It can do a clever job of imitating a human but that cleverness is in the programmer, not the machine itself. So we aren’t there yet—data remains largely foreign and it’s one of the reasons why a data culture isn’t easy for most companies.
What does it mean, therefore, to design a culture that enables humans to more effectively use machines and data? The key to incorporating machines into our teams is dividing the cognitive load of the team by understanding what we as humans are good at and what machines are good at. Humans excel at the unpredictable and in understanding context. This enables humans to see patterns from previous experience and apply knowledge in new situations, but we must continually learn from data-based feedback for our intuitions to be reliable. Machines can think in multiple dimensions, are fast and repeatable but work best in situations where past data is reliably representative of the future state.
A data culture that works is one where humans are included; where humans use data to update their intuitions and where people are able to access machine knowledge in ways that are user-friendly and intuitive. Tools and systems need to bridge the gap between how machines learn and how humans learn.
We recommend a two-step process for defining a new data culture culture.
Step 1: define your “drafting” versus “breakaway” culture
As a first step, we encourage teams to use our “breakaway” versus “drafting” structure to bring out ideas for strong data cultures (think of drafting and breakaway in cycling terms—drafting with the pack vs breaking away from the pack). Start by describing the culture that will emerge if things continue as-is. This “drafting” culture is weak and will emerge if there is no conscious effort to manage culture change. Many companies describe a drafting data culture in terms such as “decisions made by politics not merit” or “data is tribal not open” or “we learn by instruction not experimentation.”
Next, describe the “breakaway” culture; the strong culture that you want to design. A strong data culture might value:
- embracing “radical access”; who, when, why, where and how is everyone included irrespective of their technical skills?
- thinking probabilistically: use data to manage uncertainty and complexity; are multiple hypotheses entertained?
- data storytelling; use the power of stories to communicate causal relationships and to keep asking “how might we be wrong?”
- embracing failure: seeing failure as an inevitable outcome of progress and an indication that there is experimentation in the business
- curiosity as a meta-goal: curiosity as an objective can enable other learning goals, how is it rewarded?
- challenging the idea: questioning data and humans on the same level by understanding the differences between human and machine biases
- objective feedback: create data explicitly for feedback and understand the cost/benefit of data
Step 2: describe your “headwinds” and “tailwinds”
These are conditions that help or hinder the design and implementation of a working data culture. These could be headwinds such as lack of access to user-friendly tools for people, or tailwinds such as a high level of executive commitment to developing a data culture.
This exercise works well in a virtual whiteboard. Use our templates to prompt your ideas and place sticky notes with each idea in the appropriate places on a canvas such as Miro, in MS teams or a Google doc. Reach out if you’re interested in working with our templates directly in Miro.
PS: if you, like most people, thought you had a thorough understanding of how a toilet works but then realized perhaps you didn’t, this is how one works:
When the toilet is flushed, the water flows from the tank quickly into the bowl, raising the water level in the bowl above the highest curve of the trapway. This purges the trapway of air, filling it with water. As soon as the trapway fills, a siphon effect is created that sucks water out of the bowl and sends it through the trapway to the drain. This action stops when the water level in the bowl is lower than the first bend of the trapway, allowing air to interrupt the process.