To centralize or not, that is the question
Before jumping in to answer this question, let’s start with some definitions to make sure we are on the same page.
What is a team?
Oxford defines it as “Two or more people working together”. The verb “to team up” means to “Come together as a team to achieve a common goal”. Key things here: working together and common goal.
If you are part of a team and are not working together, are you really a team? What are some famous teams that come to mind? I’m a basketball fan so of course the Dream Team comes to mind. Michael Jordan, the greatest of all time in my opinion, once said “talent wins games, but teamwork and intelligence wins championships”. Successful sports teams win because they can work together. There is some chemistry that just makes them work together so well. Now, you can go on your favorite search engine and type in teamwork quotes and you will find countless quotes from famous athletes to top leaders to legendary coaches. They all say the same thing: teamwork is important and a team that works together succeeds.
Teams work on a common goal. They are not a group of individuals that are doing their own thing and report under the same organizational structure. That, by definition is just a group. In groups, everyone does their own thing, there is little to no collaboration between the individuals, and little to no impact on each other.
How do we see that play out in data teams? For example, let’s take data analysts since it’s the most common use case seen. There is a group of 10 analysts. Each analyst is aligned to a specific business function and there’s very little overlap between the metrics and/or reporting needs. Now, in this case I would say that unless there’s a team lead that has a vision for their team and all these analysts are aware of the goal that they are achieving together, then this is just a group of analysts and not a team.
Now that we’ve talked about what it means to be a team, let's talk a bit more about what I mean about centralized vs decentralized teams. Decentralized teams may sound like an oxymoron, but remember that a team is a group of two or more people working together towards a common goal.
What are centralized teams?
Centralized teams are organized whereby all team members are reporting to the same person. I don’t mean that they are reporting to the CTO, I mean they’re direct reports to a manager. For the case of data engineering, this could be a tech lead, engineering manager, or in some cases VP of Data Engineering. For data science, this could be all data scientists report to a central Lead Data Scientist which may report to the Head of Data and Analytics or someone along that level.
What are decentralized teams?
Decentralized teams are the opposite. They are really not teams because they don’t work together for a common goal. Decentralized teams are ones that I have seen living in silos. Each business unit has their own “data team” and they may not necessarily be sharing the same tools, methodology, or measurement system.
This sounds a bit eerie and you may be thinking that I’m exaggerating. Think about your company and how your data teams are structured. Maybe you are wondering which one is better? Centralized or decentralized teams?
Which one is good or better?
The answer is that it depends. It depends sounds like a cop-out answer but the fact is that there are few factors to take into account before deciding which is one “better” for your organization. Let’s walk through those factors through a quick questionnaire.
Questionnaire
Take out a pen and paper or just your favorite notes app on your phone and ask yourself these questions and write your answers down:
Does your company have strict compliance or regulatory rules it must closely follow? Like HIPAA.
What is the size and age of your organization?
What’s the state of your data? More precisely, if you were to rank it on cleanest data ever (one would dream of this) to we have a data swamp ( data is different places and everyone is scrappy about finding the state of their business). Where would your data be on that spectrum?
Do you know how your data teams are organized today?
If for question 1, you said it has strict regulations it must abide by, then centralized data teams are your best bet. That’s because data governance has to be a key goal for an organization that is guided by rules such as HIPAA. If you said that there are no regulations, let's continue with the questions.
If for question 2, you said that your organization is small and new, you know that reality is that you have to be scrappy. Which means all hands on deck on getting the ship moving. BUT, this does not mean that you cannot have people in your organization that are focused on ensuring that the data is correct, that it’s reliable. After all you want to make sure that you are measuring your performance accurately and consistently. Ultimately you want to see growth.
If your organization is small but has been around for a decade, a centralized team will be beneficial in ensuring that people are being efficient in their jobs. You don’t want to have people spending time doing the same analysis as someone else. You can’t afford to waste resources like that. You don’t have that luxury. If your organization is large and has been around for at least 10 years, then let's continue with the questionnaire because we need to understand more about the organization to determine the best fit.
For question 3, the state of your data. This could be a tricky one, because maybe there is no plan or strategy around your data to begin with. If you are in a “data swamp” where data lives everywhere and no one wants to jump in and clean it up, defining whether you need a centralized or decentralized team is the least of your worries. You need to first figure out what’s the vision for data at your company. If you want a refresher on that, go back to and listen to Episode 1 of the Podcast or read the data-driven culture blog post where I talk about data driven cultures and why it’s important to have a vision and strategy.
If your data is in a “good place”; people know where data lives, how it got there, then either team structure would work. I say this because, in cases where the data is so broadly used in an organization that people know how to find the data and the state of that data, then you don’t need to have a centralized team to control the measurements and reporting. You still need to have dedicated teams that are building out the data pipelines and are building data models, but the analytics piece is just self-service.
For question 4, how are your existing analysts, data engineers, data scientists structured in your organization today? Are they part of individual groups, do they even exist in your organization?A follow up to this question, does the existing team structure work? Are the teams making progress, putting out insights, delivering on new features and/or analysis? If so, then keep that structure the way it is because it’s working. If it’s not working, well then, there’s some work to be done. If it’s a centralized team and it’s not working, then this could be a result of lack clarity in the vision or strategy for that team. If it’s a decentralized team structure, then the issues could be do to the fragmented nature of this team. It could mean that for the size and age of your organization, a decentralized team does not work. Now, switching drastically to a centralized team from one day to the next is harder to do, unless that direction is coming from the top and even if that’s the case, it’s also hard. Before jumping in and turning the world upside down, meet people where they are at. What I mean by that is that, issues of underperforming teams, regardless of size, stem from lack of communication, a lack of clarity of job responsibilities, a lack of understanding what others in the organization are doing, and very key piece is how they are contributing to the organization.
You see, determining whether to centralize or not is harder after the teams are formed. It is always easier once you start. But, even after you start, don’t pigeonhole yourself. Organizations change and grow. What you say today may not be the case 2 years from now or even 5 years. Keep and open mind and be aware of how your data teams are performing in the structure that they exist today because it may need to change in the future.
Share your thoughts and experiences on your experience in centralized or de-centralized teams right here on the site or on Twitter @thedataplaybook.
Resources
Definition of team: https://www.lexico.com/en/definition/team
High performing teams: https://www.mckinsey.com/business-functions/organization/our-insights/high-performing-teams-a-timeless-leadership-topic
Building an effective analytics organization: https://www.mckinsey.com/industries/financial-services/our-insights/building-an-effective-analytics-organization
Benefits of a centralized analytics team: https://www.cio.com/article/3203364/the-big-time-benefits-of-a-centralized-data-analytics-team.html