Warum Sie ein zentralisiertes Data-Science-Team brauchen

People   |   Alan Jacobson   |   Aug 31, 2023 TIME TO READ: 8 MINS
TIME TO READ: 8 MINS

Organizations worldwide and across nearly every industry have been putting data science organizations in place. While this is not new, there are a lot of ways to organize data teams. The role of the CDO is changing from data steward to transformation agent of the organization. With this change in mission came significant changes in the organizational structure required to achieve new goals. So, what structure is right for a data science organization? Should an analytics team be a centralized data science team or decentralized model? Here we will talk a bit about the history of what is happening and where things are heading.

The changing role of the CDO

In the early days, CDOs were focused on data, specifically data governance. Many CDOs were organized within IT, and while centralized, the mandate was quite small – organize and control access to data. With time, the role expanded to delivering analytics and ultimately helping organizations transform to more data-driven processes. The titles started to change as well with the emergence of the Chief Analytic Officer (CAO) or sometimes the combined title of Chief Data & Analytic Officer (CDAO). And as this evolution occurred, the structure and reporting lines tended to change.

As the role was no longer singularly about data and the metrics around how much you had landed, but instead about analytics, the reporting lines started to move away from the IT organization to other areas of the company. This change has continued, and we now found organizations that have put Chief Transformation Officers in place with nearly the same type of remit, helping modernize the way teams work. Fundamentally a large portion of this work tends to be the same focus of using data and analytics to drive efficiency and revenue growth. This transition to focus on value creation for the organization through transforming the way work is done has led to a large percentage of leaders coming from change management backgrounds versus information technology backgrounds.

If you interview some of the top Chief Data & Analytic Officers or Chief Transformation Officers, you will hear them say that the function ideally reports to the CEO or COO, and if not there, it will more likely fit under the CFO than IT. Some have asked if placing data science teams under the CTO would be a fit, based on the technical nature of the work, to which I would suggest that the goal of most modern-day analytic organizations is to operationalize the analytics, not to be known for new research breakthroughs. With this implementation imperative, the closer to the operation the team reports, the better.

Should data teams be centralized?

Another common discussion around the data science organization is around the degree of centralization: should data scientists report into the domains they serve, or should they remain centralized? Here I provide several key reasons why centralization is key to success and a hallmark of the most mature organizations while recognizing the need for data scientists to have depth in a domain:

Key problems that are solved with analytics frequently span multiple domains.

Let’s say you are working to identify correlations to key customer quality issues. Your solution identifies where climactic conditions are resulting in failures and where plant process issues and customer demographics (younger consumers) are having significant problems with your product. Would this analysis be useful for your Manufacturing organization? Or perhaps the solution could be used by Product Development engineers to design improvements? Or possibly the call center could also provide guidance for those young customers;

The data clearly doesn’t know what organization it is a part of, and certainly, the solution belongs in all of these areas. If the goal is to transform the way this organization reacts to quality, a siloed data science approach will likely not reach its full potential and, even worse, will result in separate solutions being built in each domain.

Data scientists need career paths too.

The most important asset of any data science organization is its people. When teams are decentralized and report within each domain area, it is more difficult to properly create career paths for data scientists. If you are a part of a 3-person marketing data analytics team versus the 50-person corporate data science team, the number of opportunities is not only fewer, the support the organization can give is lowered. Are there design reviews with analytic professionals across the company reviewing and guiding key analytic projects? Is the Marketing data scientist valued in the same way by a Marketing organization as she would be by a data science organization?

Within an analytics team, you might have a data engineer building a data warehouse or a business intelligence analyst building dashboards; these professionals need to have career paths to potentially become data scientists. Including the data team structure with the data analysts along with data scientists as team members offers great career growth opportunities. Progressing these analytic professionals from data modeling to more advanced techniques such as machine learning, generative AI, as well as basic statistics will occur more readily as part of a centralized team vs having these professionals decentralized and on an island.

Who will develop the technical training for these data scientists and the overall data science tools and processes to be used? If you view data science as a critical talent for the organization, you likely will need to treat it as a domain with its own skill sets, mentoring, succession planning, talent development and training.

The data science function can serve as a valuable linkage across the business.

I can remember one of my earliest data science problems where two organizations came with seemingly similar problems. The Purchasing team requested analytics to help recognize opportunities to source business to global suppliers. These mega-sources would ship parts worldwide to become large-scale parts providers. At the same time, the manufacturing organization that was responsible for shipping and logistics requested an analysis to identify where shipping could be reduced by localizing suppliers near the plants they were providing parts to. One could see how this could be a great business for the data scientists, as two competing solutions could be developed, and as each customer enjoyed the results, they would want to improve their solution to ‘beat’ the competing analytics that put pressure on their results. Instead, as a centralized data team, the goal would be to provide optimal sourcing that balanced the requirements of both organizations and ultimately delivered the best answer to the corporation. Again, if transformation is desired, centralized work is frequently a powerful elixir.

Analytic methods span business units.

Many analytic solutions that provide breakthroughs for the business are not new algorithms but instead, the application of algorithms that were successfully pioneered in one area and applied to a different domain. Having cross-functional teams with broad experience that work together to solve an organization’s most challenging problems is likely to yield better results than isolated teams working in a single domain.

Your data science team should be the conductor/leader driving organizational change.

Having a centralized team to provide the change management to drive your organization to achieve higher analytic maturity should be a key goal. You don’t want 5 different teams driving this but, instead, one team orchestrating the governance, change management, best practice delivery, and leadership to deliver results to help everyone in the organization become more data literate and analytically savvy.

Developing cross-functional relationships.

While centralization is quite critical, this does not mean that data scientists aren’t spending huge amounts of time embedded with the operations they are working to help. Much like a blackbelt goes to the source of a problem, the data scientist needs to understand the problems they are working on to provide remedies that make sense.
Furthermore, centralization doesn’t mean that people outside the data science function aren’t also doing data science. The goal is to digitally enable everyone in your organization to be data literate and to be able to use data to gain insights, drive actions, and deliver results. When the effort becomes cross-functional, difficult, or critical to the business, the centralized team should likely get the call.

What structure does your data science organization currently employ? What do you wish it would look like? Have you seen results that were related to how the organization was organized? Let us know your thoughts.

Tags
  • BI/Analytics/Data Science
  • Analytics Leader