Finding the Right Mix for Your Analytics Team
An analytics team is a group of people working together to solve a business problem using data. Hence, some level of business understanding is needed to be successful in any analytics team. Although technical skills are important, soft skills (communication, presentation, interpersonal, and curiosity) are equally critical. I have seen many teams failing due to teams being created without right mix or any thought put toward it. Bringing a random group of analytics talent will only obstruct an organization’s data journey.
Where should the data team sit in your organization?
What constitutes the right mix of team members?
How do you create a dream data team to produce value?
Where do you fit in the data team?
Why is your data team not functioning effectively?
More than two decades ago, IT owned everything, and there was no need to maintain a separate data department. Business teams relied on IT to create report on a monthly or quarterly basis to serve their needs. Using data as part of daily business workflows is a relatively new concept, and the shape of data teams is still evolving.
Much is made of hiring talented data team members and investing in modern machine learning tools, but in the process, businesses inadvertently overlook the human aspect of analytics. To extract real value from data, technical advancement needs to meet—and mesh with—the human side of data. Zillow, a leading online real estate company, is a recent example of a business that blamed its data science teams for losing billions on house prediction algorithms. But although data science contributed to some degree, this major failure also came down to human factors such as sentiment and behavior being missed.
Most of the questions we look to answer with data are human-centric: How can we acquire more subscribers, improve customer engagement, and improve patient experience? Hence, adding people to the data process is essential to drive analytics adoption.
This leads us to the next question of what your data team should look like. What skills should it have? What kind of data structure will work best for your organization, leading to more effective collaboration and better value?
When pondering the answer to these questions, always consider the people in the equation. In other words, use data to assist in decision making—and not as a decision maker.
The placement of a data team in an organization often reflects that organization’s attitude and mindset toward data. There are all kinds of companies that see data as an asset, data as a cost center, data as an afterthought, or data as a necessity. Although every organization has its own rationale, this mindset can change over time, leading companies to try different data team structures and team placements.
As companies rely increasingly on data for decision making, data-related jobs are on the rise.42 Organizations of all sizes are constantly searching for that dream team with the needed data skills. Gone are the days when some companies were technology companies and others were businesses specializing in, for example, a supply chain. Today, even a supply chain company needs to think data to understand the logistics challenges borne of COVID lockdowns. Only supply chain companies that can leverage data to improve visibility about demand and inventory and anticipate potential disruptions can survive.
To uncover creative ways to organize people, data, and infrastructure, a business strategy to increase data agility and impactful outcomes is critical.
Data agility describes a perfect synchronization of technology and data to provide value for data investment. In the following diagram, all pieces complement each other, and one without the other will slow down value gain through data use and fail to produce business results.
Components to drive forward your data agility
Start asking some fundamental questions to solidify the structure, required data roles, and placement of the data team within your organization. No two companies have the same data structure, and it’s often necessary to reorganize data teams every few years to attain full operational efficiency.
1. Start by defining your purpose in having a data function. Define the near-term value you are trying to create using data. For example, do you hope to
• Consolidate fragmented data products and unify building trusted data—Save Money.
• Drive innovation and identify new opportunities—Bring Money.
• Improve operational efficiency by leveraging data to identify pain points—Save Time.
Sourcing, team skills, and aligning your team structure depends on your near time goals (Save Money, Bring Money, or Save Time) and values you want to achieve.
2. Are you adding data to your organization as a new job family, or are you rearranging your IT department to fulfill data requests as needed?
Your answer to this question will help you determine the kind of roles and rough reporting structure you need. This is also the time to assess siloed data projects, and the users of such data, across organizations.
3. Within what industry will your data team work? Health care, for example, has HIPAA and other privacy regulations that oversee the sharing—or, more accurately, not sharing—of patient information. In contrast, supply chain organizations can share vendor information much more freely.
4. Will your data team assume the role of services/help operations or product innovator? Often organizational leaders will set up a data team as a services department to create reports as per business requirements—all the while indirectly hoping that the team will drive transformation and the creation of the next generation of data products. This leads to confusion and conflicting expectations on the data team rather than the production of value.
5. Which is the higher priority for your data team: speed or quality? Although both speed and quality can be important, it is better to acknowledge which has more significance for your company. Is it more critical to get the insight fast to make a business decision, or to uncover missing or incorrect data to improve overall quality?
6. What is your organization’s collective data strength? Do you understand the collective data your company owns? Do you have a good sense of the power of connected data across teams, or at least have procedures in place to manage data sharing between cross-functional teams? These questions apply to both internal and external data sets. Do you have a process to supplement your own data strength with external data as needed? For example, you may be collecting patient data as a team (internal data) and getting physician data form a national data source (external data). But as an organization you do not connect both patient and physician data in meaningful ways to answer questions like: Is the same patient shared between physicians? Is there a group of physicians who work closely together?
To gain value and impact from your data, it is essential to effectively structure your data department. Here, I discuss four different team structure models for achieving different kinds of data results.
Model 1: Blend With Software Development
Because software technology has been around longer than data, a blended team is the most logical starting point for many companies. When the volume of data that could be collected began to grow, software engineers were the first to step up to implement data solutions initially and hence leading to the word “data engineers.”
Some aspects of software engineering require backend processing of data, so there is tightly glued commingling of technologies between software development and data. It is not surprising to note that about 42 percent of data engineers—including me—once held a software engineering role.43 There are generalists who can work on both software and data needs on an as-needed basis. It is not possible to be an expert on all the related technologies, but these jacks-of-all-trades know enough to support companies’ software and data needs.
Small companies may benefit from this model when they do not handle enough data work to maintain a dedicated data team. Often these blended departments will evolve until there is a need for dedicated team.
A common mistake software engineers make while performing cross-function work or transitioning into a data role is trying to start with the most technical, buzz-worthy topic areas, for example machine learning. But it is not possible to dive into these fields without understanding the fundamentals of data itself. Data cleaning and understanding data are the two most important of these, and data teams spend most of their time in data prep and cleaning. Hence, it is best to begin with a narrow focus on learning how to clean data, validate data, summarize data, and build iteratively on fragments of knowledge. The following diagram is one example of a software engineer’s journey to data:
Data journey of a software engineer
Model 2: Central Data Team
A central data team maintains a separate data function that is crucial to the organization and supports all its data needs. It provides a unified reporting structure of data talent under one overarching umbrella, irrespective of the products or departments they support. A centralized team may be split into several focus areas, or towers, such as the following:
• Platform teams focus on foundational aspects of a data framework.
• Client success and insight teams provide white-glove service to new and existing clients.
• Data governance manages controls around data access, creating trusted data sets.
• Data labs and innovation teams lead the creation of next-generation products and solutions through experimentation. This model is well suited for organizations with large volumes of data-related work and products built by a central team, which can benefit various groups across organizations.
Another way to arrange a data team is to embed it within different business areas. In this model, there is data practice within marketing, finance, audit, and other departments. This arrangement sometimes includes a Center of Excellence (CoE) in addition to embedded data teams.
I often get the question, “What is a CoE, and what are the benefits of having one?”
Think of a CoE as a knowledge center that can be used by any team in the organization to learn general answers and best practices. It helps to streamline analytics efforts across the organization by taking on the advisory or consultative role, while individual data teams handle implementation within their business units. A CoE is helpful in large organizations with diverse portfolios of products because it offers a central place for training and the adoption of new tools. It is also helpful in companies that wish to acquire other businesses and add them seamlessly to existing organizational operations.
In this team structure, each business function has its own data talent, data pods, and priorities.
Tap Into the Benefits of Inner Sourcing
With embedded data teams, you can build expertise in decentralized pockets of the organization, so it is vital to share knowledge and grow your community of data practitioners. Inner sourcing is the practice of either borrowing team members across functions or sharing the code base to solve problems collaboratively. Inner sourcing ensures that no team is reinventing the wheel within the organization—instead they can tap into a foundation of already-existing company knowledge.
Model 4: Crowdsourcing
You might have heard the words open source and outsource already. But what is this new breed called crowdsourcing? In straightforward terms, crowdsourcing is simply the process of outsourcing an activity to a crowd. Crowdsourcing, however, does not equal open sourcing.
Have you been part of a company that has ideas for a data product but does not have the money, talent, or time to build it? Or a company that leverages community to generate ideas for product innovation? Try to think of two examples of companies you can resonate using crowdsourcing. Here’s one: we all are familiar with LEGO building blocks. Only a few people are aware, however, of the large community of about million users who can submit ideas on the LEGO website. Call it crowdsourcing or not, but LEGO used collective community knowledge to drive product innovation.
Another example of a data-crowdsourced project is Missing Maps,44 whose mission is to add vulnerable communities of the world to the map. This helps to provide services for the people in need.
Now let’s imagine a project in your organization where you need to create business glossary of common terms or perform a data labeling
exercise. Such a project you can crowdsource by collaborating with universities or technical communities. So, while crowdsourcing is not a team structure per se, it is something to consider when you want to achieve certain data goals.
Most organizations arrange their data teams by recreating, tweaking, or customizing one of the aforementioned models. There are also organizations that spread their data talent across every department with no overarching unification. We see this sometimes in parent organizations that acquire companies with no roadmap to integrate them, or in companies that do not treat data with importance. But while some companies do feature this type of disintegrated data talent across the organization, it should be considered the exception to the rule than the standard.
Advantages and Disadvantages of Team Structures
There are advantages and disadvantages of all four aforementioned team structures. It is important to understand the advantages and shortcomings of each model before you pick the model best suited for your organization.
Think you know data team structures? Think again
Data Team Model | Advantages | Disadvantages |
Blended with Software Development | • Doesn’t add too many team members—ideal for small companies | • Inefficient in large companies with heavy data backlogs • Cannot support data-driven strategy as team members are generalists with minimal specialization |
Central Data Team | • With no siloed groups, it offers better support if someone quits the team • Better suited to large organizations delivering cross-functional data solutions and sharing ideas across departments | • Requires a strong data leader (often missing) • Can delay delivery of business needs if not paired with clear priorities and a clear data strategy |
Embedded Data Teams | • Helps achieve business outcomes faster | • Inner sourcing may not always be supported to help other teams • May be difficult to adopt data analytics as a key area due to the difficulty of achieving alignment across an enterprise |
Crowdsourcing (not a team structure but an option for some ad hoc data requests) | • Helps businesses tap into mass intelligence • Applicable to small and low-budget initiatives | • Higher risk of plagiarism and breaches of confidentiality • Quality of work may suffer if there are insufficient guidelines or acceptance criteria |
Different Data Team Roles
Let’s use the analogy of a professional kitchen to describe the different roles that exist within data teams. Consider data as an ingredient and setting up teams as different roles in kitchen to create a finished meal to consumers’ satisfaction. Organizations need not fill every role, and some roles are not needed for an organization’s entire life cycle. In addition, titles are often used interchangeably and could mean different things for different companies. But you will need to consider every role and responsibility in order to build the team that works best for your organization.
Data Leader
A data leader is like a head chef who oversees the operation of the entire kitchen or organization. Depending on the company size, they could be one individual or multiple people working together to successfully manage and use data. A data leader could be a chief data officer (CDO), data strategist, or a director or manager of data function. The data leader is the visionary, the one who has an idea of their organization’s data direction in upcoming years. They have a thorough business understanding and focus on using data to solve problems.
A data leader champions the need for organizational data capability, which helps the rest of the organization follow along. These members should set the highest-priority items for the enterprise in order to direct analytics talent toward the most critical priorities, break department silos, and facilitate open communication.
Data Engineer
This is a critical and foundational role that is often overshadowed by the data scientist role. A data engineer processes large volumes of data, builds data platforms, plans and enacts disaster recovery, and creates data pipelines for both internal and external data sources that can be used by data analysts and data scientists. Finding a good data engineer is tough because the role requires not only technical coding skills but also broader industry or domain knowledge. I have met several data engineers who possess good technical skills but struggle with problem solving, understanding their data to set up appropriate data flow. Although data engineers will not be analyzing data like a data analyst, they are still required to have a strong overall understanding of company’s data.
When building a data team, your data engineer should be the first person you hire after your data leader.
A data engineer needs a wide range of skills, so at most organizations, it is not an entry-level role. And, with more organizations adopting DevOps and DataOps, a data engineer’s role is evolving constantly. As a result, it is better suited to someone adaptable and curious—a problem solver. They are the sous chefs in the kitchen, always ensuring the flow of data, while the other roles described as follows are the station chefs responsible for specific parts of the meal.
Data Analyst
This is one of the most misinterpreted roles, and consequently the data analyst is often treated as a second-class citizen. In some organizations, data analyst is a technical role requiring strong SQL, Python, and other coding tools knowledge, while at other places it is a business role requiring the ability to interpret data anomalies using the business logic in place. Some companies will also require an analyst with knowledge of their business domain—health care or finance, for example—to accelerate the analysis process.
A good data analyst can be an asset for the organization as they bring different angles to the data and bring up the issues in data. A good analyst can see issues in the data, which were overlooked or not traceable by others. Data analysis is a good starting role for someone who wants to work with data to build immense data knowledge in terms of both company’s data and overall data analysis skills required. A good data analyst does not fear the unknown and loves solving problems as efficiently and quickly as they can.
Visual Designer
This is a relatively new role in data teams, and not all companies have a visual designer. Have you been part of an organization where there is no consistency across a team’s visualizations built by different members of the teams? Have you seen dashboards that are hard to interpret, with too much information and too little clarity and simplicity? Have you worked with visualization teams who are good at retrieving the necessary data but unable to present it in an engaging way? If you answered yes to any of these questions, visual designers are the solution. The core responsibility of the visual designer is to work in conjunction with visualization analyst to present data in rich, simple, engaging ways, ensuring a consistent and accurate information experience across the organization.
This role is well suited for someone with creative ideas, collaborative skills, and the ability to communicate and influence overall product design. They are the decorateur chef, ensuring the best aesthetics for the meal or the products being built.
Data Visualization Analyst
This role is for someone with both business understanding and knowledge of visualization tools such as Tableau and Power BI. But hiring for this role should be based on tool-agnostic and focus instead on business understanding. The data visualization analyst acts as a facilitator or enabler of data-driven decision making by providing the necessary insights and highlighting potential opportunities. Users do not know all the data that is available, and they don’t provide clear requirements for what they want to see. Visualization analysts should fill the gap between what users want and what they need, and supplement this with visualization best practices to produce a report that people want to use—rather than a report that scares them.
In my experience working with visual analysts, I have noticed that even people strong with tools knowledge may struggle to present data in meaningful ways for business or executive users. In Chapter 7, I explain data storytelling—the art of persuasion and design thinking.
Data Scientist
The role of the data scientist is to create next-generation data products using predictive models. With the advances in AI automation, there is a newer role called Citizen Data Scientist (CDS). Gartner® defines “A citizen data scientist is a person who creates or generates models that leverage predictive or prescriptive analytics, but whose primary job function is outside of the field of statistics and analytics.”45 They can perform both simple and moderately sophisticated analytical tasks that would previously have required more expertise of a data scientist. CDS can be considered as a generalist who can supplement data scientists by building solutions using off-the-shelf AI tools like Dataiku, and so on, and not perform the complex functionalities of data modeling.
A CDS performs the easier, more repetitive, or more conventional analytics tasks, leaving complex challenging task of finding hidden insights to the data scientists. It is a role that more efficiently allocates both funds and skills, since data scientists are expensive and tough to find.
Data Literacy Specialist (DLS)
The DLS is another new role emerging and evolving across organizations. DLSs are champions advocating the use of the data and the benefits associated with it. They may also create training material for rolling out a data literacy program or documentation that captures different metrics of an organization’s data literacy. They can also mentor less experienced teams in data use. Some background in leading change management initiatives will help a DLS to succeed in their role. DLS should be able to increase a company’s awareness and adoption of the available data.
It is not enough for organizations to want their personnel to use data—they must provide the right environment and tools to enable data literacy. I have outlined the challenges of data literacy in later chapters, but a proactive DLS can go a long way in helping your company overcome challenges and gain traction toward enterprise data literacy.
Data Steward
Does your company collect a lot of data but not possess sufficient trust or confidence in the data? Do definitions (e.g., “patient ID”) change from one system to another system within the organization? Let’s say you are creating a report to show inventory levels at various coffee shops the organization owns, but no one manages to inactivate stores in the database when a store closes. Users will lose confidence in the report’s inventory numbers as it continues to show inventory of closed stores. These are the kind of scenarios a data steward can help with. The core functions of a data steward are to manage quality of data, perform data-quality profiling, and maintain metadata with a complete understanding of data flow and system integrations. If your organization has a data catalog tool, the data steward will spend their time to maintain it. Depending on the needs of the company, a team may include both technical and business stewards to maintain high-quality data.
Conclusion
Regardless of the data structure or team skills an organization chooses, it is essential to measure the journey and incorporate assessment metrics. When building a data team, do not simply pick people with strong technical skills. The human aspect of data literacy is equally important if you want to gain value from your data. People from diverse backgrounds will bring different perspectives that can help with effective problem solving. And don’t always prioritize outside talent with the latest certification and training for your team. Consider internal talent—people who already possess strong, experiential business knowledge to fill the roles of a data team. Create a path to upskill existing team members hoping to gain new data skills.
Here are some key takeaways to consider when building an efficient data team:
• Plan a team structure. Your team should showcase more than just technical skill and hence should not be made up of only technical team members. Humans respond to emotions, personal motives, and social influences. Numbers cannot be interpreted as mathematical calculations solely and true data literacy requires more than reading numbers. Teams must consist of members with a strong understanding of people’s social behavior. Well-rounded, diverse teams will not only see data as computed but also see the reasoning behind it.
• Be prepared to evolve. No team structure will be perfect or last forever. Companies should start with what makes logical sense, define and measure success, and be prepared to change as needed.
• Define value proposition. Oftentimes, value is treated as abstract with a consensus that data-driven model will add value. It’s better to specifically identify your value proposition for the next six months, one year, and five years. Define value gains prior to the start of the next data initiative, and check in frequently to ensure that nothing gets derailed along the way.
• Celebrate the small wins. Keep communication channels open and celebrate small wins, such as when your data team tries something new or solves a pesky problem. No one enjoys being unappreciated or unacknowledged. Celebration and recognition will motivate others to join your community of data users and bring different perspectives.
• Aim for collective data strength. Gone are the days where the focus was on collecting data and determining better ways to store it. Now the focus is connecting various silos of data to derive meaning. The team structure you implement should help connect data and determine the collective strength of the data the organization holds. This includes both connecting internal data and supplementing this with external data.
• Cultivate curiosity about data adoption. Instead of rolling out team structures as part of a mundane reorganization, creating chaos and fear, explain the reasoning behind team changes. Stoke curiosity about what next year will look like for both individuals and organizations. Create new job openings that lead to better opportunities.