11 building blocks for a successful data strategy

Data strategy has become one of the many buzzwords used and misused in the data & analytics space. To some it stands for building a data lake, to others it’s creating a data catalog or training employees to use dashboards. These may be components needed for a data-driven organization but that’s not strategy.?

#Strategy is a plan to achieve one or more long-term goals under conditions of uncertainty. It’s a broad formula for how a business is going to compete, what its goals should be, and what policies will be needed to carry out those goals. With that in mind, #DataStrategy is a plan with:

• clear goals of how to become data-driven
• actions required to get there
• resources needed to execute on these actions?

So, what is a successful data strategy, with success defined here as being an organization that uses data for business decisions to the fullest extent possible? Like with many other complex topics, it helps to break it down into smaller components. Understanding all these individual components helps also in developing and executing data strategy as well as communicating about data with non-data colleagues and executives.?

The following eleven building blocks of data strategy is a pragmatic approach in defining data strategy based on my experience across few companies and industries. These building blocks keep always coming up as success drivers.

1. Executive sponsorship

The fish stinks from the head. Any failures in leadership work their way down the rest of the company. But the opposite is true as well: change begins from the top. Data strategy is a great example. A company cannot become data-driven if there is no executive sponsorship.?

Executive sponsorship requires not only talk but also action. Adding lofty goals about using data for business decisions to strategy slides won’t buy you much if you don’t put the money where the mouth is. Executive sponsorship includes providing resources to build analytics capabilities, supporting change management efforts required throughout this data transformation, reviewing progress towards the goal, and probably most important, rejecting any investment requests not backed up by data. Executives need to ensure that data strategy is closely linked with corporate strategy and that corporate strategy is informed by data.

2. Business needs

What is #AI for? Or #BI? Like any other utility or function in a business setting, data & analytics enable the business to achieve its goals. For a for-profit business, it’s increasing revenues, reducing costs, and ideally both, all that while addressing customers’, employees’, and other stakeholders’ needs. None of the activities in the data & analytics area should happen in a vacuum disconnected from business objectives.?

So, how do you align data & analytics investments with business objectives? Ideally, business wants would be translated into data products. However, business colleagues and executives may not have the full visibility and understanding of what is possible to accomplish with data. Instead of relying on business wants, the data & analytics teams should collaboratively evaluate with business colleagues their needs, driven of course by business objectives. This consultative engagement ensures that business questions and hypotheses are fleshed out, business colleagues are involved in data product development leading to better acceptance, and data products are contributing at the end to business goals.

3. Descriptive analytics

Reporting is dead, long live reporting! Or dashboarding or visual analytics or whatever you want to call it. With the current hype around big data and AI you may get the impression that the panacea for a data-driven organization is predicting the future, i.e., predictive analytics. But analyzing what has happened in the past, i.e., descriptive analytics, is the starting point of any analyses.?

Descriptive analytics is not necessarily about creating dashboards, although you may get that impression based on how dashboarding is becoming a buzzword on its own. It’s about evaluating what has happened in the past, identifying trends, correlations, and causations. Dashboards may be a way to visualize what has happened, but infographics, a chart or two lines in an email may do the job too.?

If a dashboard is the means to communicate what has happened or what is currently happening, it’s important to keep the audience in mind and create the dashboard based on business needs. Would you expect the average airplane passenger to understand the airplane cockpit dashboard? No, you wouldn’t. But you would expect the average airplane passenger to be able to navigate the entertainment screen in front of her.?

Same is true in business. Descriptive analytics needs to focus on digesting relevant insight for the business user and presenting it in an intuitive and easily digestible format. And if a dashboard is the chosen method to visualize that insight, think of a Tesla dashboard not Boeing.

4. Data privacy alignment

1.6 billion Euros. This is the amount of fines and penalties imposed under the General Data Protection Regulation (#GDPR) on companies operating in the European Union as of January 2022.?

Any organization using data that can identify a person needs to ensure that data privacy regulations in the countries it operates in are met. These regulations are still an evolving field. For example, GDPR, probably the currently strictest data privacy protection in the world, became law in the European Union “only” in May 2018. GDPR has replaced previous data protection rules across Europe, provided greater protection and rights to individuals, and defined large fines for organizations not abiding by these rules. Some of its principles include data minimization, storage limitation, integrity and confidentiality, and accountability.?

Like with any law, it leaves room for interpretation. What is minimal data, for example? This may look very different for a pharma company only starting to leverage personal data for business decisions vs. an e-commerce business built on customer data. According to GDPR, personal data shall be “limited to what is necessary in relation to the purposes for which they are processed”. Pharma and e-commerce will have different viewpoints of what is “necessary”.?

Here is where the alignment with data privacy comes in. The main job of legal and data privacy professionals in an organization is to minimize risk to the business. The best way to avoid risk associated with processing personal data is to not process any data at all. This of course collides with the objective of a data-driven organization. There needs to be a trade-off between risk and reward.

A successful data strategy requires strong cooperation and alignment between the data & analytics teams and data privacy experts. Both sides need to understand each other’s objectives and they need to have a collaborative and transparent partnership. The trade-off between risk and reward needs to be clearly defined and it must be backed up by executive sponsorship.

5. Talent

So far, the first four building blocks of the data strategy could have been accomplished without internal data teams. With executive sponsorship and clearly defined business needs, a consulting company could have analyzed the past and a software provider could have created intuitive dashboards, all that while ensuring alignment with data privacy.?

If you want to move to the next level, you need to start building internal data teams. While the hype around data scientists may indicate that hiring data scientists will make an AI company, a successful data strategy is built along the entire #DataAnalyticsValueChain. This value chain consists of several steps and starts with defining a business hypothesis. It’s followed by identifying the data assets needed to address this hypothesis and acquiring & cleansing new data if not available in-house. After alignment with data privacy comes preparation of data models, data analysis, and review & validation of insights. This often is an iterative process before the results are published and turned into business decisions.?

You need different skill sets along this data analytics value chain. You need data translators to derive business needs, data stewards to identify relevant data, data engineers to integrate several data sources and prepare data models, data analysts to analyze the past, data scientists to develop machine learning models, ML engineers to automate modeling pipelines, and visualization experts to, well, visualize insights. Building these data teams won’t happen overnight and the mix of these roles will depend on the data maturity of the organization and will evolve over time.?

When you are ready to start hiring, pay attention to the needs of your future employees. Data people have in general a #GrowthMindset (as opposed to a fixed mindset). They want to learn, they want to grow, they want to be challenged. And many times, they also want to see a purpose behind what they do. It’s not only about money.

Once you have the people on board, retention becomes crucial. To support growth, be ready to offer promotional steps that will recognize the growing experience and contribution of the person. Offering only two levels in the HR job leveling such as “junior” and “senior” for individual contributors (since not everybody wants to be a people manager) may not cut it. And since data people want to learn and grow, you may want to deviate from always requiring direct business impact from data teams. Limited “play time” for blue skies projects where direct business benefits may not be immediately apparent may be a great motivator for growth-minded data people.

6. Data mindset

What is the biggest inhibitor of becoming a data-driven organization? For some organizations it may be the data mindset, or the lack of. And I’m not talking here about the data mindset of analysts, data scientists, or data engineers. It should be given that they have it. I’m talking here about the data mindset of non-data employees.?

But what is data mindset? A mindset is a way of thinking or a frame of mind. Data mindset defines then how you use data in your daily decisions. #DataLiteracy is probably a more hands-on term to describe the ability to use data, which can be taught. According to MIT, data literacy is the ability to read, work with, analyze, and argue with data.?

One approach to improve data mindset is a data literacy program targeted to all employees. This is, however, not a quick band aid. A program like this takes three to four years to bear fruits. It needs to be customized to the current state of the organization and should include several components such as formal learning courses, informal lunch & learn sessions, data consultation hours, short informational videos, podcasts, social media posts, posters, fliers, etc. It should be given that the data literacy program needs to be fully supported though executive sponsorship, ideally including direct executives’ involvement.?

In addition to reading, working with, analyzing, and arguing with data, data literacy includes understanding of how every employee generates data. For example, in a B2B setting, a sales rep can communicate with the customer through an email system that captures opens and clicks, in which case every interaction with the customer improves the understanding of that customer. On the other hand, if this communication happens through Outlook email, there is no systematic and structured way to capture the customer engagement and improve the understanding of the customer. Since this is usually a choice of the sales rep, data generation is a crucial component of data mindset.?

Every data-driven company needs data savvy employees. Data literacy is not just a skill. It’s a mindset. A data literacy program needs to be built bottom-up and improved continuously through experiments. Think here more of #DataEvangelism than corporate training.

7. Data governance

If data is the new gold, do you know where your gold is? Or do you know who knows where it is? If you don’t have it in-house, do you know where you can get it? Once you get it, do you know how to integrate it with your existing gold? Do you have a catalog where you can look up where your gold is, what it consists of, and what its purity level is? And do you have a taxonomy that describes all that so when two people talk about it, they talk the same language??

Yes, many questions. But you need to have clear answers if you are serious about using gold … I mean data … for business decisions. All these questions should be answered by a well-defined data governance.?

#DataGovernance is a data management concept that addresses data availability, quality, usability, and integrity. It consists of processes, policies, standards, metrics, and roles that enable the effective and efficient use of data across the organization. It also establishes clear accountability to ensure compliance with data privacy requirements. A data steward is a role that defines the data governance processes, manages it, and is the go-to person for any questions about your data assets.

Data governance leads to decreased data management costs and to increased access to data for relevant stakeholders, helping also to grow the data mindset.

8. Data backbone

Data needs to be stored somewhere, ideally secure and meeting data privacy requirements. Appropriate IT infrastructure and tools need to enable the data practices that have been defined through data governance’s policies and standards. And the IT infrastructure needs to address basic requirements such as sufficient data bandwidth.?

As customers of IT services, the data & analytics teams need to have access to data that is of high quality, usability, and integrity. They need IT systems that allow easy integration of new data sources. The most important part of a sound data backbone, however, is the consolidation of the relevant data in central data repositories based on use cases. For example, for customer analysis in a B2B setting, a central data repository needs to consolidate all data relevant for every customer, which may come through several engagement points and sources. In a manufacturing environment, a central data repository needs to include all data captured through the manufacturing process and its machines. And in a retail setting, a central data repository should combine all transactional data associated with procuring, storing, and selling products.?

A well-managed data backbone needs to deliver all this.

9. Advanced analytics

Artificial Intelligence (AI) is not the panacea for all business challenges, but advanced analytics is the next level for an organization to become data-savvy, leading to higher business impact as well as higher analytics sophistication. The basic insights extracted so far through descriptive analytics can be now enhanced, even using the same data, through new methods delivering decision support and even decision automation.?

Advanced analytics consists of areas such as predictive and prescriptive analytics, optimization, and simulation. #Statistics and #MachineLearning methods are used for data mining, segmentation, pattern detection, image recognition, and probability predictions, among others. All these can lead to higher revenues, lower costs, and optimized investments. These solutions can range from a proof of concept where an idea is being explored to fully automated real-time machine learning engines.

Let’s ignore the hype of Terminator-level general AI, which doesn’t exist yet, even though this may be the public and sometimes even business perception. Unfortunately, this misconception is often reflected in discussions on this topic where the number of times somebody mentions “AI” is inversely proportional to the knowledge of that person on this topic. Let’s also not expect that simply hiring data scientists will do the trick to become a data-driven company.?

And speaking of data scientists, this is where their knowledge of statistics and machine learning concepts comes in. Data scientists are part of a data team with different skill sets along the data analytics value chain. By the time you get to invest in advanced analytics you should have many of the other roles and functions operational.?

10. Production

Most companies’ approach to data & analytics, especially to advanced analytics, is to start with a proof of concept (POC), pilot, minimum viable product (MVP) or whatever you want to call it. This is a perfectly viable beginning. Without clearly defined data flows in place, without machine learning pipelines available, and without the infrastructure to manage all this, starting small makes perfect sense.?

At some point in time, however, demands for data products will move from one-off solutions to frequent updates, replication for other entities, other customer segments, or other but similar data sources, and real-time data analysis. At this point in time, data flows and machine learning pipelines will need to be automated, machine learning models optimized for performance and scalability, and the models deployed into production to ensure repeatable and dependable operation.?

This is where #MLOps comes in. It’s a function consisting of machine learning engineers who automate the pipelines and optimize the models. This team consists also of software engineers who simplify the code and make it more performant. They will develop continuous integration (CI) and continuous deployment (CD) pipelines to automate building, testing, and deploying machine learning models.

Whether an advanced analytics solution requires automation will depend on the business requirement. Any time data is processed and analyzed real-time whether it’s transactional data in retail, phone calls in telecom, or production data in manufacturing, machine learning pipelines need to be automated. This is also required for models that are refreshed “only” daily or weekly. But even models with less frequent updates will benefit from code simplification and optimization.

11. External partnerships

Do you have sufficient data to drive your business decisions? With a successful data governance you should know what data you have available internally, but that should also give you some ideas what potential other data, and the associated insights, is available out there. This is where partnering with other companies or organizations comes in.

Let’s take healthcare as an example. A physician may treat one thousand patients per year, so she builds experience from several thousands of patients over the years. Wouldn’t the physician experience and hence patient treatments be improved if the physician could leverage the experience from treating millions of patients? Data & insights partnerships can achieve that. Partnerships in healthcare are of course challenging due to the siloed approach not only separating data across the three key stakeholders patients, providers, and payers, but also with basically no data sharing within each of the stakeholders. But if there is a will, there is a way.?

How can this be accomplished? The easiest way would be to simply share data. Since data is the new gold and hence must have some monetary value associated with it, data sharing would need to be compensated. This is a perfect solution in many environments where data involves products, services, or processes. Cooperation with other companies in these areas may improve data-driven decisions for all participants.

But when it comes to personally identifiable data, there are limits driven especially by regulations such as GDPR in the EU. However, that’s not necessarily a showstopper. At the end, you don’t really need the data, you need the insights available in the data. Methodologies such as #FederatedMachineLearning, differential privacy, and privacy-preserving record linkage allow for sharing insights contained in the data without sharing the data itself. For example, a physician could get the insights on how other physicians have treated a disease from the insurance companies who capture the treatment data from all physicians. Other approaches such as #NFTs (non-fungible tokens) could enable patients to share their personal health information transparently and efficiently, which could then be aggregated and used for treatment decisions for other patients, all for the broader good.?

Think of external partnerships as an approach to capture additional external insights to improve data-driven decisions.

Data strategy on a page

Bringing these eleven building blocks together leads to a data strategy on a page. This is meant to be an easy lookup of what is needed to become a successful data-driven company. The implementation of course is not easy, and the sequence of implementation will depend on where the organization is today. For some who are just starting to realize the value of data there are usually many low-hanging fruits to pick. For others who have organically grown their analytics capabilities over the years there are several choices to move to the next level.

Implementation and data roadmap are topics on their own, but for now here is the combined data strategy on a page: