Abstract

Organizations that invest in developing data science capabilities can opt for, among other choices, embedding data science teams in functional departments. While this structure promises more success in analytics projects by connecting them closely with specific needs, providing dedicated resources, and focusing expertise, challenges are in the way of securing these benefits. We conducted a case study of a data science team embedded in a people analytics department at a human resource (HR) function in a large, multinational technology company. The team is nested within HR and organized as an embedded data science team, working solely on HR-related data science projects focusing on business intelligence and analytics. We found three key advantages of the embedded data science structure: 1) agility in delivering quicker and better tailored analytics, 2) concentrated and cultivated data science expertise, and 3) emerging bottom-up solution ideas. We also identified the biggest challenges that need to be overcome to achieve these benefits. We also discuss the disadvantages of the embedded data science structure together with mitigating actions. Finally, we highlight the key role of the data science manager as a crucial link between the team and the organization. Drawing on our findings and analysis, managers can make better informed decisions regarding investments in data science capabilities.

Keywords: data science teams, data science projects, managing data science, data science, analytics

Media Summary

Companies invest more and more in creating data science teams. Some of them choose to embed a data science team in a specific department; for example, finance or human resources. While this promises more success in data science projects by connecting them closely with specific needs, providing dedicated resources, and focusing expertise, challenges stand in the way of securing these benefits. We conducted a case study of a data science team in a people analytics department at a human resource (HR) function in a large, multinational technology company. The team is nested within HR and organized as an embedded data science team, working solely on HR-related data science projects focusing on business intelligence and analytics. We found three key advantages of the embedded data science structure: 1) agility in delivering quicker and better tailored analytics, 2) concentrated and cultivated data science expertise, and 3) emerging bottom-up solution ideas. We also identified the biggest challenges that need to be overcome to achieve these benefits.

We also uncovered the main disadvantages of embedding a data science team: 1) growing away from the enterprise infrastructure, 2) duplicating data science efforts, and 3) weaker links with strategic impact. Balancing the advantages and disadvantages of the embedded data science structure puts heavy responsibility on the shoulders of data science team managers who need three key aptitudes: implementing appropriate prioritization, maintaining team visibility, and developing horizontal networks with other data science teams. Organizations should carefully consider the importance of data science managers, and train and reward them appropriately. This is crucial with the embedded data science structure, as organizations want to leverage the agility, expertise, and creativity of embedded teams without creating risks or challenges.

1. Introduction

As organizations more and more commonly develop data science capabilities to support their operations, decisions around structuring these units are fraught with complexities (Davenport & Bean, 2018; Joshi et al., 2021). While data science encompasses a range of different endeavors, from developing artificial intelligence and machine learning algorithms, through data engineering, to establishing statistical inference, we focus specifically on business intelligence and analytics (Parmiggiani et al., 2022; Vaast & Pinnsoneault, 2021). This is because it is this part of data science that many organizations begin building their data science capabilities with, and to date this form of data science dominates across industries. Among the structures adopted for data science teams are centralized centers of excellence, where a single team focuses on delivering data science to the whole organization, or embedding data science teams in functional departments (Koch et al., 2021; Stelmaszak, 2022). Embedded data science teams work only for their designated functional department, for example, marketing or human resources (HR), and data scientists have limited interaction with data science teams and projects in other functions.

Organizations with embedded data science teams envisage that this structure may bring more success in analytics projects by connecting them directly with functional needs and aligning them closely with business objectives, providing dedicated resources, and focusing expertise in one area (Koch et al., 2021). A data science team may benefit from this structure as data scientists are usually more involved in the planning and execution of projects; talent and resources can specialize and develop subject matter expertise; and new projects can leverage existing expertise. For these reasons, embedded data science is a popular choice for functions requiring specialized knowledge, such as HR or finance (Gal et al., 2018; Leonardi & Contractor, 2018). It is also often chosen for shorter-duration projects more common in data science endeavors focused on business intelligence and analytics products. While projects may be shorter, the outcome products are often used for extended periods and thus require ongoing support, justifying the close engagement of the embedded data science team.

However, achieving these benefits is not guaranteed: managers need to know how to secure them amidst some of the challenges of the embedded structure (Giermindl et al., 2021; Günther et al., 2017). To investigate the advantages of embedding data science teams and the prevalent challenges in securing them, as well as to identify the actions managers can take to minimize the disadvantages of embedding data scientists, we conducted a case study of a research and data science team consisting of 13 members embedded in a people analytics department (consisting of around 40 data analysts, consultants, and business analysts) at a 2,000-strong HR function in a large, multinational technology company employing over 100,000 employees all over the world. The team is nested within HR and works solely on HR-related data science projects, falling under HR in the organizational structure, and providing the function mainly with business intelligence and analytics capabilities, which include a whole range of data science techniques, from applying artificial intelligence and machine learning algorithms to designing visualization dashboards. The team reports to HR operations and analytics, with two levels of leadership between the team’s manager and the chief HR officer.

At the time of the study, the team consisted of six data scientists (responsible more for data and analytics) and six research scientists (responsible more for defining problems and questions), led by the head of data science (serving as the team manager) supported by the lead data scientist (responsible for managing the team’s workload). We provide more information on the team’s makeup in Table 1. Example projects include modeling attrition pressures using social graph analysis combined with other sources, or measuring worldwide employee satisfaction through surveys and natural language processing. In general, data scientists on the team are charged with analyzing data, building models, and developing data science products, such as dashboards and web applications, while research scientists tend to focus more on identifying suitable research questions, designing data collection, and consulting on data science projects with respect to behavioral and social science methods of data collection and analysis. The team operates with more emphasis on providing data science products that often require ongoing operational support, while it sometimes also engages in consulting-like, limited in scope services. The flexibility to move between these modes of operating is noted by the team as one of the benefits of the embedded structure.

Table 1. Team composition, duties, backgrounds, and interviews conducted.

The case study, conducted in spring and summer 2021, involved 19 semistructured interviews with 10 out of 13 employees, with six of them interviewed more than once, totaling around 1,160 minutes of interviews (details in Table 1). In the interviews, we asked about the advantages and disadvantages of the embedded data science structure as perceived by the team members, the lead data scientist, and the head of the team. Importantly for our study, the organization also has a centralized data science team nested within the IT function. This means that we were able to ask interviewees about their perceptions of advantages and disadvantages of the embedded structure against what they observed regarding the centralized team. We analyzed the data using qualitative coding to synthesize the three key advantages and three key disadvantages of this structure, together with the mechanisms to ensure their respective realization and minimization.

We found that the advantages of the embedded data science structure were: (1) agility in delivering quicker and better tailored analytics, (2) concentrated and cultivated data science expertise, and (3) emerging bottom-up solutions. On the other hand, our study revealed some disadvantages of embedding data scientists in functional departments: (1) growing away from the enterprise infrastructure, (2) duplicating data science efforts, and (3) weakening links with strategic impact. In the rest of the article, we describe each advantage together with actions taken to secure it, and analyze each disadvantage with actions taken to minimize it. We conclude with a discussion of the role of the manager as a crucial link between the embedded data science team, the function, and the wider organization, highlighting the three key skills that the data science manager needs to rely on when at the helm of an embedded data science team. We draw on the analysis and selected direct quotes from the interviewees to support the findings.

Our findings are directly relevant to organizations that are considering establishing or reorganizing data science teams. The article is helpful for business managers evaluating the possibility of implementing embedded data science teams. Data science managers will directly benefit from reading the article and learning about the actions they can take to maximize the opportunities of the embedded structure. Data scientists themselves can also learn from the analysis of the advantages and disadvantages of working in an embedded data science team.

2. Securing the Advantages of the Embedded Data Science Structure

2.1. Agility in Delivering Quicker and Better Tailored Analytics

In the case study we conducted, one of the clearer advantages of the embedded data science structure is agility in delivering quicker and better tailored analytics. The data science team universally mentioned that working as a dedicated team in HR enabled more responsiveness toward HR needs: “we’re able to respond to [internal HR] customers more quickly and more with individual or custom type solutions than we could necessarily do if we were centralized” (Head of Data Science). This means that the embedded team is able to work more closely with the function to identify specific needs, for example, by involving customer representatives in projects, demonstrating progress and obtaining immediate feedback, and offering a quicker turnaround on analytics solutions. This was highlighted in our research:

“I have a meeting next week with two people from the talent planning organization that are literally going to sit down with me and my team for three hours and give us the same training they got because we can’t build a product to help them do their job unless we understand their job at really minute detail. And centralization, I think, it’s hard to get that minute when you’re this big.” (Lead Data Scientist)

For these reasons, within the HR function, having a dedicated data science team is seen as a benefit: “having those embedded teams is going to allow the different functions to have more custom solutions that add more value within those organizations” (Head of Data Science).

However, ensuring that a dedicated data science team can deliver analytics quickly and in an aligned way often requires overcoming specific challenges. Agility and speed of delivery may be hampered because of the limited resource of an embedded team. Unlike in other structures where additional talent may be brought onto projects as and when needed, an embedded team has less flexibility, and delays may have significant knock-on effects on other projects:

“There is that challenge that we might get too swamped with too many requests and too much work. Because all of the requests will be drilled down to us. And eventually, our small data science team will have a huge workload for each one of us. And that means that we won’t be able to take as many projects as we want to and customers might feel some gaps or see us not be able to deliver anything for them.” (Data Scientist)

Tailoring data science solutions and aligning them to the specific needs of the function can also lead to ambiguity and challenges in prioritizing the must-have and could-have functionalities.

To overcome these challenges and secure the agility advantage, data science managers working in embedded teams develop strong prioritization rules and strict project management guidelines. When setting priorities, the process needs to be robust but flexible enough not to erase opportunities for identifying bottom-up project ideas. More than in other data science team structures, the embedded structure requires managers to cross-train data scientists and ensure that there are no team members who are the sole experts in a given tool or technique. To mitigate the risk of any bottlenecks or delays, managers also need to cultivate a collaborative approach:

“Everyone is very willing to do something for somebody else, pick up a piece of a project for somebody else entirely, mentor them on how to get it done, if they want to, like everybody’s very willing to work with each other in a lot of different ways.” (Lead Data Scientist)

This approach is often based on mentoring, through on-the-job training, to pair programming. Ensuring that team members in the embedded structure can cover for and help each other where needed is essential in ensuring agility.

2.2. Concentrated and Cultivated Data Science Expertise

The second advantage emerging clearly from our study is concentrated and cultivated data science expertise, that is, fostering HR-related knowledge in the embedded data science team. Such function-specific knowledge is essential in ensuring that data science and analytics products are tailored to add unique value to the function, and an embedded team builds up and curates such knowledge:

“If you’re embedded, you learn the challenges of that business, you’re embedded in the language, you can build up some credibility by being able to get up to speed quicker on what the problem is, or understanding the political reality.” (Senior Data Scientist)

There are several benefits of specialization for the function:

“When you have specialized teams, they can focus on the specific domain. And they can build expertise in those domains. And also, since they will work on the same type of project year over year, they can improve those projects and products over time. And they can shift their focus on improving things rather than just doing the same thing over and over and not have enough time to improve anything.” (Data Scientist)

Concentrating expertise is also beneficial within the team, as new team members can be onboarded and trained more quickly, and problems can often be solved by talking to team members. In a centralized data science structure, customers either need to know what data scientists are able to do, or data scientists need to be skilled at soliciting project ideas. However, when data scientists have enough subject matter expertise to work hands-on, they can not only be more agile, but also more innovative and impactful with project proposals.

Yet to be able to concentrate and cultivate function-specific data science expertise, managers need to overcome several challenges. One is related to finding and hiring the right talent: when the manager wishes to hire data scientists with specific expertise in HR, the hiring pool narrows, potentially prolonging the search for the right candidate. On the other hand, recruiting a data scientist without an HR background means more time is spent on training. Intense focus on one area of expertise entails narrow specialization for data scientists, who may perceive the lack of career progression in an embedded team as a downside:

“So the difficult part about embedding is the cost to the data scientist, the lack of opportunity for career growth. Embedded teams are probably no bigger than 12 or so, 15 tops. So you don’t have skill, mentorship, you don’t have a ton of other technical opportunities, you don’t have even a very long upward growth path. So that’s a cost of the embedded model directly to the data scientists.” (Lead Data Scientist)

Employee retention is also a challenge, as highly specialized data scientists leaving the team or the organization may create significant knowledge gaps.

It is paramount for data science managers to develop strong hiring practices and knowledge management within the team. It is also essential for cultivating function-specific knowledge to ensure continuing development for team members, both in data science and in functional knowledge: “a good data science team is one that is continually upskilling. One that is learning new techniques, new models, new methods to get work done” (Data Scientist). Managers also need to establish ways for data scientists to be able to grow within their roles and expand on their professional interests, for example, by protecting time for individual research.

2.3. Emerging Bottom-Up Solution Ideas

Third, emerging bottom-up solution ideas are frequently mentioned as an advantage of having an embedded data science team. In the company we studied, representatives of the data science team were often involved in meetings where they could ideate and suggest analytics solutions to problems mentioned. More importantly, because of their function-specific knowledge, they could proactively look for problems that could be solved with their data science skills: “it’s going from ‘are you a person that can solve a problem I have’ to ‘are you a person that can find an interesting problem and then just go solve it?’” (Lead Data Scientist). This was often explicitly seen as data scientists’ role: “being able to see possibility, translate problems into possibilities, and look at the different problems that our company has, and putting them together to see if you can create a solution” (Data Scientist). This advantage of the embedded data science team is a benefit to the function because the resulting analytics projects and products tend to be more closely aligned to functional needs. Data scientists are connected closely enough to the functional professionals they work with that they can learn of the existing pain points and generate wide-ranging solutions. This proximity enables the embedded data science team to identify problems even before customers realize there may be a data science solution.

This entails that data science teams need to be well-connected and constantly liaise with customers to know their work and potential problems. Creating bottom-up solution ideas may involve relationship building and attending meetings that can take away data scientists’ time, and requires specific skills:

“But that’s really difficult for technical people. So what you’ll see is the technical data scientists, they love that going off and having a problem and tackling it. But you need the audience, you need the customer, you need to do that part too.” (Data Scientist).

Emerging bottom-up ideas also require fostering creativity, and data science managers may find it difficult to recruit data scientists with strong functional expertise and customer engagement skills, technical skills, and a creative approach.

We found that to mitigate these challenges and leverage the potential for creating solution ideas bottom-up, data science managers can shape their own role in ways that create visibility for the team while not overburdening data scientists with the need to liaise closely with customers at early stages of ideation. This is best captured in a description of the manager’s role:

“I need my manager to have more than technical skills, I need them to have strong networks and ability to influence and move or remove roadblocks. And the number one thing I need the manager to do is [that] they need to be finding the problems to solve.” (Data Scientist)

In other words, to overcome the barriers to realizing the benefit of emerging bottom-up ideas, data science managers can position themselves as those who act as an interface between the team and the function when it comes to identifying problems, while subsequently the team can brainstorm potential solutions bottom-up. Managers need to strike a balance between managing the intake of projects and empowering the data scientists to engage with customers directly.

3. Minimizing the Disadvantages of Embedding Data Scientists

In our research, we also uncovered several potential disadvantages of the embedded data science structure. It is paramount to be mindful and act against these over the course of the growth of the team.

3.1. Growing Away From the Enterprise Infrastructure

While embedded data science teams are characterized by relative independence from other areas in the organization, including the information technology (IT) area, this can increase their agility, but at the same time the risk of growing away from the enterprise infrastructure emerges. Aligning to enterprise standards and developing analytics solutions compatible with enterprise infrastructure is a constant challenge for embedded teams. Some of the underlying technology choices made in an embedded team may not align with the wider infrastructure, resulting in additional work or the need to redevelop solutions if they are handed over to IT for support and maintenance:

“When we prototype, if it is going to become an enterprise solution, it does still have to go to IT for design. And they have to rebuild it if we didn’t follow the approved architecture, or we used open source library that [the company] isn’t going to approve. Hopefully they would approve it, but if they don’t, then IT actually has to figure out how to rebuild our models and approach. And they essentially just do it again.” (Head of Data Science)

The interoperability of tools and architecture with the enterprise infrastructure needs to be ensured in products that are handed over to IT, which includes those that meet a certain level of usage and have high standards for continuous performance and maintenance. With solutions that remain under the embedded data science team’s auspices, collaboration with the IT area also extends to smaller data science projects where the security and performance of the infrastructure are paramount.

The manager’s role here is to foster continuous alignment, which requires significant work: “the most problematic thing for me is the number of IT stakeholders, who they are, when I need to connect with who, and so on” (Head of Data Science). The added challenge is that many IT areas do not have prior experience of working with embedded data science teams: “our team is just blowing things up, because we are an embedded data science team. And they [IT] haven't really had to deal with the creation of any kind of scaled solutions outside of their groups in the past” (Head of Data Science). Managers then need to have an active role in managing relationships with IT, while the wider organization needs to recognize that embedding data science teams will change some processes within IT areas. A clear division of responsibilities and bounds around the respective remits need to be put in place as well to avoid grey areas, and the data science manager needs to recognize that a good partnership with the IT area is a foundational enabler.

3.2. Duplicating Data Science Efforts

Embedding data science teams in functions is also associated with the risk of creating data science silos and duplicating data science efforts. Embedded teams may find themselves working on similar projects and delivering similar solutions without being aware of this duplication. This entails significant additional costs for the organization, as such efforts could be pooled and streamlined. This challenge can be mitigated by managers by establishing relationships with other embedded data science teams:

“I would hope that [duplication] will be minimized in the future as we build a stronger partnership. So I would hope that there would still be a more mature partnership model, we would be at least close enough to have a sense of the type of work that our other groups are doing. So that we know to ask, this seems like it's within your team scope. Have you done something like this?” (Head of Data Science)

Managers can also initiate regular knowledge-sharing sessions involving other embedded data science teams and, for example, discuss current and future projects to uncover cases of potential overlap or duplication. Encouraging developing informal connections by data scientists may also help. Visibility of work and openness about it is crucial.

3.3. Weaker Links With Strategic Impact

Finally, embedded data science teams face the challenge of weaker links with strategic impact. As their projects and solutions tend to be function-specific, maintaining visibility and a clear alignment with strategic objectives requires ongoing work. Reflecting on past experience, one data scientist explained the problem:

“The team produced some really cool work, the people who received the products were very grateful and got a lot of value out of them. But it wasn't well marketed. It wasn’t necessarily attached to big strategic initiatives, not that everything needs to be. But when you’re a relatively small team, you probably need to be more connected to bigger efforts.” (Data Scientist)

It is then key for the manager to work on ensuring constant visibility and strategic alignment. Data scientists will look up to their managers for ensuring that alignment:

“The manager of the team will be a bridge between the team and customers, especially higher up customers. So the manager can also help develop those relationships between the team and customers, especially major customers and high stakeholders. And become translator between the team and the company’s strategy, and the manager can become a translator of that, how they can get that strategy and translate it into the team so that team members better understand how they work with the organization and the company's overall strategy.” (Data Scientist)

Managers can stave off this challenge by fostering relationships with senior stakeholders and participating in strategic-level meetings. While not all projects need to be or should be strategic, data science managers require mechanisms in place for gauging the scope of impact of each project to ensure that the team does not work only on many small projects with minimal impact, but rather can focus on a smaller number of high impact projects.

4. Conclusions

We studied an embedded data science team focusing its work on business intelligence and analytics in the HR function of a large, multinational technology company. Thus, our findings and conclusions are bound by this context. Embedded data science teams are more closely aligned with functional needs, which means they can bring more agility in delivering quicker and better tailored analytics, but at the same time risk growing away from the enterprise infrastructure. They focus on a specific domain that allows to concentrate and cultivate expertise but risks duplicating data science efforts. Finally, they provide a dedicated resource that can offer bottom-up solution ideas, but at the same time can suffer from weaker links with strategic impact. Balancing the advantages and disadvantages of the embedded data science structure puts heavy responsibility on the shoulders of data science team managers (Giermindl et al., 2021; Leonardi & Contractor, 2018).

This means that choosing and developing the right kind of leadership talent to step into these roles is key for organizations investing in developing data science capabilities. Specifically for embedded data science teams, managers need to have three key aptitudes. Within the team, they need to know the expertise and skills of their data scientists to appropriately prioritize and balance their workload. This may be more important than in-depth technical expertise. Vertically, they need strong interpersonal skills to develop and maintain team visibility within the organization, including at the strategic level, and they need a degree of proactivity to promote the team’s work. Finally, they need to be able to develop horizontal networks with other data science teams.

Data scientists themselves should also think through the advantages and disadvantages of embedded data science teams when choosing their career paths. One of the trade-offs is that embedded data scientists often have to be familiar with a wide range of aspects of data science but can develop deeper subject matter expertise. While there are some challenges of working within this structure, data scientists can trust that a capable manager will be able to navigate around these challenges to create a thriving embedded data science team that is recognized not only within the function, but also in the organization as a whole.

Organizations should also carefully consider the importance of data science managers, and train and reward them appropriately. This is especially crucial when choosing the embedded data science structure, as organizations want to leverage the agility, expertise, and creativity of embedded teams without creating undue risks or challenges. Large organizations, like the one we studied, are likely to be able to invest both in embedded and centralized structures to reap the benefits of both. Smaller organizations may need to decide between the two structures and the findings from our research hold for them as well. Such organizations may consider a compromise between the two and deploy a hybrid structure, where some data scientists work centrally and the rest are embedded, with similar benefits and challenges to those we discussed. There is scope for further research around data science teams that focus on developing artificial intelligence and machine learning algorithms and engage in other forms of data science. Similarly, the applicability of our findings needs to be tested in data science teams embedded in other organizational functions, especially those more closely related to product development. Such contexts may highlight productive and important differences and modifications to our findings.

Disclosure Statement

Marta Stelmaszak and Kelsey Kline have no financial or non-financial disclosures to share for this article.

References

Davenport, T. H., & Bean, R. (2018, February 15). Big companies are embracing analytics, but most still don’t have a data-driven culture. Harvard Business Review. https://hbr.org/2018/02/big-companies-are-embracing-analytics-but-most-still-dont-have-a-data-driven-culture

Gal, U., Jensen, T. B., & Stein, M. K. (2018). People analytics in the age of big data: An agenda for IS research. ICIS 2017 Proceedings: Transforming Society with Digital Innovation. AIS eLibrary. https://aisel.aisnet.org/icis2017/TransformingSociety/Presentations/1/

Giermindl, L. M., Strich, F., Christ, O., Leicht-Deobald, U., & Redzepi, A. (2021). The dark sides of people analytics: Reviewing the perils for organisations and employees. European Journal of Information Systems, 31(3), 410–435. https://doi.org/10.1080/0960085X.2021.1927213

Günther, W. A., Mehrizi, M. H. R., Huysman, M., & Feldberg, F. (2017). Debating big data: A literature review on realizing value from big data. Journal of Strategic Information Systems, 26(3), 191–209. https://doi.org/10.1016/j.jsis.2017.07.003

Joshi, M. P., Su, N., Austin, R. D., & Sundaram, A. K. (2021). Why so many data science projects fail to deliver. MIT Sloan Management Review 62(3). https://sloanreview.mit.edu/article/why-so-many-data-science-projects-fail-to-deliver/

Koch, H., Chipidza, W., & Kayworth, T. R. (2021). Realizing value from shadow analytics: A case study. Journal of Strategic Information Systems, 30(2), Article 101668. https://doi.org/10.1016/j.jsis.2021.101668

Leonardi, P., & Contractor, N. (2018, November 1). Better people analytics. Harvard Business Review. https://hbr.org/2018/11/better-people-analytics

Parmiggiani, E., Østerlie, T., & Almklov, P. G. (2022). In the backrooms of data science. Journal of the Association for Information Systems, 23(1), 139–164. https//doi.org/10.17705/1jais.00718  

Stelmaszak, M. (2022). Inside a data science team: Data crafting in generating strategic value from analytics. ECIS 2022 Research Papers. AIS eLibrary. https://aisel.aisnet.org/ecis2022_rp/

Vaast, E., & Pinsonneault, A. (2021). When digital technologies enable and threaten occupational identity: The delicate balancing act of data scientists. MIS Quarterly, 45(3), 1087–1112.

©2023 Marta Stelmaszak and Kelsey Kline. This article is licensed under a Creative Commons Attribution (CC BY 4.0) International license, except where otherwise indicated with respect to particular material included in the article.