What Should be the Analytics Organization Structure?
What are the advantages and disadvantage of different structures
Analytics organizations tend to be setup in very different ways dependent on the specific companies that set them up, be it in terms of reporting lines, segmented or central teams, or in terms of overall focus (project vs. product/business based). Each of these attributes brings its own set of tradeoff that need to be understood and managed.
Over the years as an analytics professional, I had the chance to work in a variety of industry and businesses. Sometimes analytics was placed under business or products, other time finance and sometimes technology. Amazon for instance tended to place analytics and business intelligence under Finance while the data engineering was placed under the technology department, while at Facebook there was a central analytics team with sub-segments detached to the different product pillars.
The reporting lines dictate to a large extent the priorities that end up being set forth on the analytics team. Finance having a focus on control, product a focus on prioritization and business cases and technology on building data-flows.
One of the other impact of reporting line, is on the hiring selection process for data professionals, especially in areas that cannot staff a full data team.
Segmented or Centralized Analytics Team
There are different advantages and disadvantages of Segmented teams and Central teams that need to be considered. The structure has impact on hiring, knowledge sharing, collaboration, career path, focus and objectivity of the organization. Each of these tradeoffs will need to be managed in order to provide for a successful organization.
Centralized Analytics team tend to bring about a more structured approach hiring than segmented teams. They are able to set and enforce hiring standards in terms of the functions they managed. They generally have a standard interview loop and are better able to get interviewers from different analytics specialization. Segmented teams would in general have less emphasis and ability to test the analytical knowledge of the different candidates and may thus value more highly different set of skills.
Another advantage in terms of hiring is the ability to more easily place good candidate in a different domain function. For instance if a good candidate was interviewing for an analytics position that became filled in the meantime, s/he would have to potentially re-interview for a different analytics position at a company with a segmented analytics team while at one with a central team, so long as there is overall headcount at company level s/he could directly be offered the job.
Knowledge Sharing & Collaboration
An aspect that distinguishing segmented vs. central team is the approach towards knowledge sharing and collaboration within the analytics teams. In central analytics teams knowledge sharing can happen in a variety of ways, through working on the same project together when the needs come, through project rotation and hand overs, through mentoring, or typical presentations or code reviews. In segmented analytics teams, the knowledge sharing within data will need to be pushed at a higher rate, segmented teams tend to benefit from higher involvement with the specific business they are in, at the expense of wider analytics involvement.
Given that a large part of growth within data roles come from working hands on on projects with other data professionals and being handed stretch projects in certain analytical domains, the segmented data team format hinders growth within the analytics career ladder. More often than not, for people having done their career in segmented data teams, the natural career progression is towards business or product roles rather than progression towards more senior analytics roles.
Segmented analytics team tend to hire more generalists than central analytics teams. Central teams benefit for economics of scale and specialization and can therefore more easily divide responsibilities between different job descriptions such as analysts, data engineers, data-scientists, … While in segmented teams there might only be 1 or 2 data person supporting a given business leaving no scope for specialization. The data person in these teams need to be a jack of all trade.
Segmented data teams by virtue of focusing on a single business area tend to have a deeper focus than central analytics team that may more easily rotate across project and business areas. The deeper focus has advantage in enabling the analytics practitioner to more easily know what kind of behavior is expected and what are the kind of anomalies that you would see in the data, it also allows the analytics practitioner to better avoid the different pitfalls in the data. The broader focus offered by central analytics team however bring also their own set of advantages such as being able to leverage best practices and be able to more easily port methods from other part of the business.
While segmented teams reporting to finance may have a certain independence bringing a degree of objectivity, this is not always the case when the data persons report to the business. Analytics practitioners might be pressured to report numbers in a certain way, or making leaps of thoughts they are not comfortable with. A central analytics team help bring a certain independence in these cases.
Matrix organization help blend the benefits of a segmented analytics team and the benefit of a central team by detaching their members to the specific business cause. Analytics practitioner are able to get a deeper business understanding while retaining all the advantages of a central analytics team in terms of hiring, career path, knowledge sharing and objectivity.
Project or product/business line focus
In some organizations beside the direct line management there are also questions as to how to staff analytics practitioners. Whether it is by project or by assigning them directly to a business or product line.
A lot of the value of analytics is obtained by setting a performance measurement process and in putting processes in a state of control. As such it is important to have at least an analytics owner that is available to monitor the performance and take actions such as refreshing a machine learning model or directly deep dive into the datasets to correct a data or identify a business issue in need of fixing.
Nevertheless project focus can have value for certain analytics tasks, whether it is to provide a temporary relief in case of peak workload or in needing to first prove value for investing in certain areas. These types of project based work can also be useful for analytics practitioners to discover different domain areas.
There are tradeoffs that need to be considered when setting up an Analytics Organization, these usually are simplified into what’s the degree of data focus versus business focus. There are multiple aspect to it however and ways to mitigate certain tradeoffs. Dedicated central analytics organization can, for instance, be setup in a matrix like structure to help them get more into the specific business mindset and priorities.