forming a solid, comprehensive, impactful, competitive, and holistic data strategy

“Data Scientist is the sexiest job of the 21st century” is old news, “data is the new oil” is repeated ad nauseam, everyone’s collecting data and trying to use it nowadays, yet most companies do not have a data strategy.

Needless to say, that is a problem. Data strategy is the way in which a company organizes and leverages its data and information systems to create value — without this, maintaining said data is an expensive liability. Putting data scientists in place to blindly “do data science” makes it even more expensive. Organizations of all sizes and industries in 2023 need data strategy, and need it fast.

The good news is it is also an opportunity. Now more than ever, companies can and should use data strategy to gain a competitive advantage. Companies that have integrated a good data strategy make their data work for them and are truly data-driven.

In this article, I will outline the steps and things to consider in formulating a comprehensive data strategy, summarizing and expounding on what I’ve learned in the first half of Data Science and Ethical AI at the Asian Institute of Management.

Asking Right Analytics Questions

All strategy starts with a goal in mind: a general form a military strategy to win a war, a student forms a study strategy to ace a test, a coach forms a training and game-day strategy to beat their opponent, and a data science leader forms a data strategy to answer the right analytics questions.

To form this question, one must have intimate knowledge of a company’s specific needs. Beyond “how do we increase revenue and reduce cost”, being in an industry and company long enough, thinking critically about its performance, will very quickly lead to an impactful business question. Here are some examples, generated by my learning team at the Asian Institute of Management in just a few minutes:

• How can we decrease the rate of failed deliveries? (Logistics Industry)
• How do we reduce customer churn? (SaaS industry)
• How can we prevent or reduce the incidence of patient falls in a hospital? (Hospital industry)
• How can we maximize our occupancy rate? (Hotel industry)

Exactly! Good analytics questions come from good business questions. An analyst in an organization cannot “do” analytics without a question that impacts the rest of the organization.

With that original business question, we can now start forming analytics questions. There are three types of analytics questions: (a) descriptive, (b) predictive, and (c) prescriptive.

Descriptive questions bring out patterns in historical data. Going back to the SaaS example “how do we reduce client churn”, some descriptive questions could be “what is the customer profile of our churned customers”, or “what is the feedback of our churned clients?”

Predictive questions on the other hand are questions that are answered by modelling the historical data and seeing where it will go in the future. Back to our example, “what customers are likely to churn?” is a good predictive analytics question.

Finally, prescriptive questions take what is learned through descriptive and predictive analytics (as well as sometimes manipulating the variables in the predictions) and translates it into concrete action. For instance, “how do we choose clients that are a good fit for our product?”

Using descriptive, predictive, and prescriptive analytics questions, a data science team can get to work finding out what they need to answer them.

Mission Articulation

Parallel to the formation of analytics questions in the need to articulate an organization’s mission statement. According to Gordon from Investopedia, a mission statement “is used by a company to explain, in simple and concise terms, its purpose(s) for being.” Here are a couple real-world examples of mission statements:

• Starbucks: “To inspire and nurture the human spirit — one person, one cup, and one neighborhood at a time.”

• Robinsons Supermarket: “To educate and empower its customers on their choice of food and products that promote health living.”

While a company’s objectives may overlap heavily with the objectives of other companies (especially the majority of companies organized, let’s face it, for profit), a solid mission statement gives a company a single calling to move towards. This is most obvious in a company’s products and marketing materials, but also plays a large part in forming a data strategy.

Unintegrated with the company’s mission, a data science and analytics team will have trouble figuring out its priorities. I will demonstrate this in the next section.

From Mission to Tech Workload

The mission statement feeds directly into what are called strategic imperatives, or important objectives that serve the company’s mission. For example, if our hypothetical SaaS company’s mission is to be the household name when it comes to the service they provide, then its strategic imperatives may be to attract clients from all sectors, and to provide a product that is well-priced and universally liked. From attracting clients from all sectors comes the business objective of choosing the sales channels that effectively convert the most customers, which, in turn, is a use case for a simple tool that tracks marketing activities and conversions. You can also see how this informs the forming of analytics questions as well!

When one knows an organization’s mission and strategic imperatives, it becomes much easier to figure out the technical analytics and data science workload needed, rather than working backwards and creating tools for the sake of creating tools (and it’s easier to sell to the higher-ups this way!).

Defensive vs Offensive Data Strategies

When formulating a data strategy, one also has to consider whether to take a defensive or offensive stance. No, this isn’t me taking the coaching metaphor from three sections ago too far; having a defensive data strategy means focusing on data security and data quality, while having an offensive data strategy means focusing on having wider and more flexible access to data. A defensive data strategy leads to reliable data that is well-documented, accessible only to those with permission, and compliant with regulations. Offensive data strategy on the other hand leads to data that is easy to access and well adapted to different business uses. A defensive data structure is the Single Source of Truth (SSOT) where data is kept in one central system, while an offensive data structure is the Multiple Versions of the Truth (MVOT), where data is kept in multiple systems to serve different purposes.

Like many other facts of life, a company should formulate a strategy that strikes a balance between defensive and offensive. What that balance looks like depends on many factors, such as the company’s industry, size, digital maturity, and, indeed, its mission. A retail start-up looking to get into grocery stores nationwide would do well with a more aggressive data strategy to move fast and find its market, while a well-established bank would do better with a more defensive data strategy to ensure it is not recording any mistakes, to be able to detect fraud and to comply with strict regulations.

This balance may change over time as well, and it often does. A company may choose to create an MVOT system based on their current SSOT when expanding internationally, so that data science and analytics capabilities are adapted to each country it opens in. A tech startup like Uber that expanded aggressively and may have started with MVOT for each of its business units on the other hand would’ve had to consolidate them into a SSOT to better protect its growing base of drivers and minimize risks as they were getting more attention from regulators.

Whole-of-Enterprise Approach

The last major thing to consider in forming a data strategy is how to integrate data science into the organization. Ideally, one would favor a whole-of-enterprise approach, where the data team has involvement and buy-in from all units and levels of the organization. A data science team that lives separately from the rest of the company and only comes out once in a while to give reports cannot do its job effectively, as the questions it asks and the models it creates will likely be irrelevant to the actual needs of the company. Data science and analytics are expensive and time-consuming: worst case scenario, it can be defunded if there is no visible return on investment. Data science should be in constant communication with representatives from all business units, as well as the C-suite, in order to make the most impact.

Conclusion and Reflection

In summary, a data team does not work alone, but must be aligned with the needs and interests of the whole company — that much is obvious. What was not obvious to me before learning data strategy is how deep that alignment must go, that one ought to be able to trace a line from a company’s mission statement to the technical data workload. There’s a big difference between that and the seemingly anything goes “just do analytics” culture of my previous workplaces. I now see how a lack of an explicit data strategy leads to a lack of organization and direction for even the strongest of data science analytics teams.

A new data team should work with stakeholders using the preceeding talking points to form a data strategy, and an established data team should work on continually evolving its data strategy as the organization evolves. As a data science leader, I will initiate these talks with stakeholders if they aren’t already happening, and make sure my team stays true to its objective of creating real value for the company.