Community Forum – How do you structure your R&D data team?

Resource Type
Survey (Community Forum)
Publish Date
Innovation Research Interchange
Digital Innovation, Knowledge Management
Associated Event

A: How do you structure your R&D data team?

A fast-moving consumer goods R&D department is evaluating how to structure data-focused roles within the department. Currently, their data experts have been associated with individual functions and are full-time, dedicated R&D staff. They and the rest of the R&D team work on a hybrid schedule. They are hoping to build synergies and gain efficiencies by centralizing the data expert roles across R&D.

The results from this week’s survey are below.

Community Responses

*Other responses:

  • Data Science skills can be more fully developed and specialized, not diluted by functional needs
  • Avoids knowledge silos; avoids retreading the same hurdles across multiple functions; strengthens ability to illustrate value with a cohesive set of stories, initiatives, etc. without separating value streams from various functions.

*Other responses:

  • Even in a centralized model, a data scientist can very well pair up with SMEs/R&D by program(s).
  • Centralized model is preferred, but with data science translators who interact with functional areas
  • Leverage to explore more breadth in functional projects; as opposed to centralized prioritization, which may leave some functional teams without priority support.
  • The greatest advantages are the first three choices – this blended with the partnership to an Enterprose Data Science Organization to maintain a common ecosystem, toolset, and communicate data needs is ideal.
  • faster adoption

    5. What are the disadvantages of either system that you have either experienced or heard of?

    • Decentralized is generally resource limited and does not have the other data science teammembers to learn improved or best practices.
    • A centralized model can put a resource burden on time critical programs [many programs and few data scientists pool]; A decentralized model may not best utilize and leverage other data science expertise.
    • My experience trying to work with Enterprise-wide Centralized Resource Center for Data Science in an R&D team that was starting to explore Data Science is that, while they had great technical expertise ni state-of-the-art Data Science and the coroporate IT tools available to execute it, they lacked business accumen and it was dificult bringing them on board for people not having experience working with data scientists. Also, getting resources for R&D projects was a very bureaucratic process, lots of justification and paperwork needed to get resources even for simple projects. All this impacted negatively the early stages of Data Science use and Data Literacy in R&D.
    • We started with a small centralized Center of Excellence which IMO was necessary to understand and build the capability for data engineering, data science, and visualization/user interface. Now we have trained/developed citizen data scientists in different business units to supplement that small core team. Our workforce is primarily engineers and technical experts, so well-suited to the citizen model. I think the shift from centralized to federated MAY make the most sense as we continue to mature. This means we would not be fully decentralized but BUs would get more responsiveness/flexibility in their own teams. Note: recommend talking with current owner of Analytics CoE (not me!) for followup.
    • Centralized model preferred, but with roles that outreach to functions. Disadvantages with centralization include people being moved around and not having strong relationships within functions. Disadvantages of decentralization include not utilizing a center of excellence approach for leveraging resources and best practices, limited development and advancement opportunities for data scientists, and lacking standardization in implementation of analytics tools and systems.
    • Confusion around best practices, misalignment on priorities and deployment tools.
    • If the Data Science group sits in IT, the non-R&D functions will consume all of their time. R&D needs its own Data Science team with a high level of domain knowledge.
    • Centralized model sometimes takes a “one-size-fits-all” approach that misses context and misrepresents the nuances of different business units. A completely decentralized approach, on the other hand, results in multiple sources of truth and makes cross-business unit portfolio management nearly impossible.
    • lack of standardization (and therefore difficulty with governance) if decentralized
    • the hybrid model often leads to the decentralized team struggling to align with the enterprise team not only in best practices but also scope of projects
    • Efficiency of any type of data system depends on data quality which is ensure through rigorous process execution, organizational discipline, and Leadership support
    • duplicate or redundant costs, works, and solutions
    • Lack of awareness and re-inventing the wheel
    • In the decentralized model, resources can be underutilized, or booked with “make work” if real needs don’t exist. However in the centralized model, smaller business groups may have difficulty getting any time allocation as the business benefits are inherently smaller.
    • Performance management is the main issue if data scientists are completely decentralized and work in individual labs. How do they get rated against the other scientists in the lab that are all of the same discipline? We have found it’s better to have an organization of the data scientists that do a mix of running their own projects supported by the business, as well as are resources to other projects in other organizations.
    • Unable to reach critical mass on key projects with a decentralized model.

    6. If you would be willing to discuss your experiences with other IRI members, please include contact information below.

    • IRI members volunteered to share their experiences

    Have a response to add?  Email us!