When you embrace tribal knowledge with the right management philosophy, you can achieve superior data science capability at little to no additional cost.
Tribal knowledge is often used in a derogatory sense to connote organizational bad practices and, in most cases, this criticism has merit.
I've been called in on several occasions to mitigate a tribal knowledge nightmare. However, tribal knowledge is not categorically a bad practice, especially with a data science team; in fact, when used under the right circumstances, it can be a secret weapon.
The wisdom of tribes
I recently did management consulting at a major utility company that runs underground pipe. Since you cannot physically see these pipes when you're above ground, you would imagine that they would have a sophisticated information system for mapping their pipes and equipment. But when I conducted initial interviews with the organization, I found just the opposite -- the bulk of their knowledge for how to get around and find equipment was inside employees' heads. They knew they were taking great risks with this practice, and that's why I was there. Instead of criticizing this organization for its bad practices, let's explore why they would continue to behave this way.
Tribal knowledge is usually an organization's path of least resistance. An organization cannot exist without knowledge, so it must reside somewhere. The options are on a continuum, ranging from tribal knowledge to a formally documented and governed information and knowledge system. Creating and maintaining a formal documentation system like this takes a lot of work; furthermore, this robust information system must adjust every time leadership goes in a new direction. So, the more you push your knowledge maturity up the scale, the more it will cost the organization in setup and maintenance. This is especially true of a data science organization due to the nature of the knowledge that must be managed.
Contrary to popular belief, tribal knowledge can work in an organization, though only under the right management philosophy. As it relates to management (strategy, program, project, etc.), there are three inter-related variables to constantly consider: scope, time, and effort.
I've created the post, lever, and balance method (PLBM) to help leaders frame their management philosophy around these variables. In short, you must decide which of the three will stand firm (post), which one you will intentionally move (lever), and which one you will allow to adjust based on the post and lever (balance).
Posting on effort infers building a team and then locking that team in for the duration of your strategy (therefore, the effort remains firm). Contrasting philosophies may use effort as a lever, where the leaders are intentionally adding or removing team members to manage their strategic outcomes; or they may use effort as a balance, where they balance team member involvement in real-time based on how the other variables (scope and time) look.
There's no right or wrong philosophy, but trying to execute on a strategy where tribal knowledge is in place while team members are moving in and out of the culture will certainly blow up in your face. Unfortunately, this is an all-too common bomb that I typically have to defuse.
It's all about the team
To protect your strategy of embracing tribal knowledge on your data science team, you should focus on the best practices of building a team. Be very selective about the people you choose for your data science team, because you're making a long-term commitment. Rely more on referrals and recommendations from people you trust over clever screening methods. Also, build a robust plan to retain these resources; make sure you present them with challenging work, and that they're rewarded well in terms of compensation and a healthy work/life balance. The biggest risk you run with tribal knowledge is attrition, so guard heavily against it.
In addition, your data science team must have a good coach who understands that human behavior and team dynamics are critical when tribal knowledge is in place. When posting on effort, your data science team becomes a cohort that will move in unison through the classic stages of team development (forming, storming, norming, performing). This is generally a good thing, but this progression is not linear or predictable; if you don't have a good coach that understands how to align your goals with your team dynamics, members will leave for greener grass and take all of their knowledge with them.
Tribal knowledge is considered a bad business practice in most organizations, yet when used correctly, it can be a strategic advantage. Where other organizations are paying a small fortune to create and maintain sophisticated knowledge and information systems, with the right management philosophy, you can achieve superior data science capability at little to no additional cost.
Build a high-performing team of great data scientists, business experts, analytic managers, and most importantly change leaders, and then do everything in your power to keep this team happy. After all, we've had tribes as long as humans have been on this earth, and they function fine -- even without fancy information systems.