For your data science team to function effectively, you must have all roles clearly defined and governed.
The toll on efficiency that unmanaged roles and responsibilities can take is somewhat misleading. On the surface it may seem, if roles aren't actively managed, something may be missed. This is rarely the case. The group will quickly figure out what's missing and fill the gap in its own informal way. Just below the surface, it may seem there would be duplication of work. You're getting close; this is true. However, it's not the biggest drain on team efficiency.
The biggest problem with under-managing roles on a data science team (or any team for that matter) is how quickly the group dynamic breaks down and takes your productivity with it. Social psychologists have studied this for some time now. The group as a whole has a basic dependency for the all the people in that group to understand what their role is. Until that's reconciled, the group will have a difficult time functioning.
The first step in avoiding group dysfunction from role ambiguity is getting all the functions and roles straight, and then assigning expected involvement to each function/role combination. This is typically done in a Roles and Responsibilities Matrix - commonly known as a RACI chart. Once done, you're in a good position to assign people to roles. This will help clarify who's doing what, as long as it's controlled properly. We'll cover that in a bit; let's first talk about setting up roles and responsibilities.
For every function a data science team needs to perform, there are a number of ways a person or role can be involved. If you've ever been on a project team, you probably know what a roles and responsibilities matrix (RACI) is. In brief, this is a matrix with team functions (e.g., leadership, management, analytics, business experts) as rows, and roles (e.g., business leader, analytic manager, lead data scientist) as columns. The following role assignments - or methods of involvement - complete the cells of the matrix:
- Responsible: The person or people who are executing the function.
- Accountable: The person who is governing the function.
- Consulted: The people who provide input and advice.
- Informed: The people who need to receive updates.
I recommend adding a Support category to the traditional RACI, making it a RASCI matrix:
- Support: The people helping the Responsible person.
Adding this type of assignment to your matrix implies there should only be one person assigned as Responsible for each function the team performs.
Once the framework is in place, we can start adding in the pieces. The primary function of any data science team is of course analytics. If you have one data scientist, then you're job is easy; however, there may be multiple data scientists on your team. Data science can be broken down into sub-functions if you have a large enough team.
For instance, you may need both qualitative and quantitative analysis if you're taking a mixed-method approach. Or, you might run parallel streams for a while then converge to synthesize best findings. Helping data scientists understand the business need are business experts. Again, depending on how your organization is structured, you may have different business experts that represent different lines of business or functions.
Data scientists and business experts won't manage themselves, so you need leadership and management functions as well. The leadership function is responsible for dealing with change and the management function is responsible for dealing with complexity. The leaders in your group will set the vision, navigate uncertainties and adjust when necessary, manage the group dynamic and motivate the team, and help the rest of the organization adopt a more analytic culture. The managers in your group will build plans and schedules, clarify objectives, manage risk, ensure the team is delivering what's expected, and yes, make sure people understand their roles and are performing their functions as prescribed.
It's very important for the management function to control for this. This is where the physicians heal themselves with their own medicine. When constructing the RASCI, list role disambiguation as a sub-function of the management function and make sure you have roles and role assignments (i.e., RASCI) to cover the function. There must be at least one role responsible and one role accountable (they could be the same role) and it's best to have support and consultation assigned as well. The function should be performed on a regular basis: weekly, bi-weekly, or monthly. The responsible and supporting people should survey the team for who's actually doing what, and if everyone's clear on their role.
Oddly enough, although the person responsible should be part of the management team, they may need some consultation or support from the leadership team. If simple controlling measures don't correct the problem (e.g., reminding people what their roles and responsibilities are), you may need a cultural intervention. Members from the leadership team who understand group culture would be best suited to advise on this; it's not fair to expect managers to have competence in this area.
Although the concept of documenting roles and responsibilities has been around for a long time, role ambiguity still runs rampant, and it can destroy the productivity of your data science team. Take some time to clearly articulate: 1) what functions your data science team will need; 2) what roles will cover those functions; 3) what method of involvement each role will have in each function; and finally 4) who will be assigned to each role. Once in place, make sure you constantly monitor the team to make sure everybody continues to understand and perform their role (their whole role and nothing but their role). Otherwise, instead of building great analytic solutions, your team will be busy trying to figure out who's on first and what's on second.
- Connect with your customers even when you don't know who they are
- Balance your big data analytics goals to enable breakthroughs