How to Define and Execute Your Data and AI Strategy

Written by DAIN Co-founders Ulla Kruhse-Lehtonen & Dirk Hofmann

To date, many companies are investing in Data and Artificial Intelligence (AI). As the terminology varies, the activities may be called Artificial Intelligence, Advanced Analytics, Data Science, or Machine Learning, but the goals are the same across companies: Increase revenues and efficiency in current business, and develop new data-enabled offerings. It is well understood that to stay competitive in the digital economy, the company’s internal processes and products need to be smart – and smartness comes from Data and AI. Successful Digital transformation is not possible without Data and AI.

In our experience, the best way to assess a company’s Data and AI maturity, is to start by focusing on what companies do with their data and analytics people. Less mature companies miss certain roles, for example, they may have Data Scientists but no Data Engineers, which means that the data environment is not properly productized. Or data and analytics experts are scattered around the organization, but not in a systematic manner indicating that the hiring decisions were made by individual visionary business leaders instead of aligning with a company-wide strategy.

Over the past three years, our company DAIN Studios has been involved in more than 40 Data and AI initiatives in different companies and industries in Finland, Germany, Austria, Switzerland, and the Netherlands. Our clients are typically large, publicly listed companies. In our work, we have defined Data and AI strategies, evaluated AI execution projects, and advised companies on topics such as data governance, organization, and operating model. We have also built cloud infrastructures, engineered data, and developed scalable machine-learning models. We have advised dozens of business leaders on how to become data driven and use AI to their benefit. This article reviews some of the findings that we have made and proposes best practices for going forward.

When we started our company in early 2016, the term Artificial Intelligence had just begun its resurrection from the 1960s. Before 2016, the cool term was “Big Data” (which now sounds hopelessly outdated) and before that, “Advanced Analytics” and “Data Science” (still used!). While many digitally native companies were applying advanced Data and AI methods in their business in 2016, many older companies were not. Digitalization and the resulting requirements for Data and AI have taken many established companies by surprise by disrupting their business models. Industries experiencing competition from digital-native companies, such as media and retail, have had to transform themselves and rapidly adopt data utilization. In contrast, many manufacturing companies are only in the first phases of their digital and data transformation.

As a result of increased Data and AI awareness, many established companies have commenced targeted Data and AI programs with big expectations to turn around the business and attract star talent. However, a couple of years into the programs, many show signs of fatigue and unmet expectations from senior managers and leaders unhappy with the speed of progress. Pilots have been made in selected areas and even data-enabled products may have been launched, but the desired large-scale business transformation has not taken place. Data and AI are still niche activities, not the premise for business. In some cases, people whisper about “the project which must not be named.” As a result, management grows increasingly impatient and wonders how to get out of the rut.

The reality is that there are no shortcuts. Amazon, Google, Apple and Facebook all used very different business strategies to gain their current market dominance and global influence, but their common success is arguably their foresight in understanding the value of data and positioning themselves early. They worked from the inside out, placing continuous emphasis on capability building, alongside developing, testing and deploying the top technologies internally, so that they could offer the best to their customers. For established, non-digital companies the road is even rockier. Old companies have established ways of working, digitally immature people, and legacy infrastructure. Overcoming those matters calls for strong determination and persistence from the company leadership. It means bringing Data and AI into the core of all aspects of decision making – from strategy to operations, supported by Key Performance Indicators that align data-driven decision making. Such action usually manifests itself with a focus on Data and AI capability development seen on the agenda of leadership meetings – from the Board, to C-suite, to senior managers. It is usually on the agenda of forward-looking Human Resource managers who understand that digital talent is of key importance to the company.

Based on our experience, business leaders need to be highly involved in all aspects of the execution of data and AI strategies and the capabilities that the supporting initiatives involve. We observe that fully committed leadership has been one of the common denominators for success in digital transformation and becoming a data-driven company.

Sometimes the leadership understands the importance of becoming Data and AI driven, but feels inadequate in their own knowledge about the subject matter. That is a good sign. Many universities and consultancies offer Data and AI training for business leaders. An effective way is to custom-tailor a Data and AI workshop as part of the leadership strategy days. A word of warning: Sometimes business leaders make the mistake to focus on statistics, computer science, and coding in their desire to enhance their understanding of AI. While coding is a critical skill for Data Scientists and Data Engineers, business leaders are better off putting their efforts into creating an effective company environment for Data and AI. That means setting business goals, hiring the right people, educating the workforce, committing to investments, and implementing an effective operating model and organization for Data and AI. This is best done by setting clear goals and incentives for the organization and following up on them.

In the next section, we go deeper into the topics that should be accounted for when defining and executing a Data and AI Strategy.

Setting the Data and AI Vision

The premise for successful Data and AI Strategy is to know your business goals. What are your must-win battles? Where do you need to succeed in the future? Obviously, access to data will help in the definition of business priorities, but it is important to remember that Data and AI will not solve your issues in business models, products, and services. Data and AI will help you make more informed decisions, obtain information faster, automate processes, and enable delivery faster than a human mind – but it will not construct or replace the lack of business vision and ideas.

AI priorities are derived from business priorities. Since Data and AI will make different contributions in different areas, when assessing where to focus a company’s Data and AI efforts, one should consider the business case for each business area as well as AI’s relative importance to the case.

For example, putting Data and AI into use in sales and marketing will typically yield results quickly, while employing them in product development takes longer but can eventually result in large impactful outcomes. Often it makes sense to start with process optimization cases. An improvement of one percent in efficiency or avoided downtime may mean saving millions of euros. Calculating business cases for cost-saving cases is easier than for new business. Early wins are important to communicate to obtain buy-in from the organization, as well as increase general understanding on demonstrated AI benefits.

Gaining support and buy-in for a Data and AI vision is equally important. The conventional way to do this is to make a business case for Data and AI showing the baseline Internal Rate of Return (IRR) on planned investments (“current state”), and compare it to the IRR of investments in Data and AI (“future state”). Strategizing along these lines is a good exercise for understanding the big picture, however, it needs to be kept in mind that for many digital products, services, and businesses, the option of “not doing Data and AI” is not feasible. Would Google be Google without Data and AI?

The natural starting point for Data and AI utilization is the optimization of current business processes: Business models, products, services, internal processes and functions (e.g. marketing, HR). Once you have a solid understanding of the Data and AI use cases helping your current business, new data-driven business opportunities should also be investigated. These include Data Monetization (e.g. selling data) and Data Partnerships (where new offerings are created by pooling data from several organizations). Neither topic is easy, but the opportunities are worth looking into.

Data Asset Management / Data Governance

The availability of high-quality data is the foundation for successful, productized AI. Data can be called an Asset if it is structured according to the FAIR principles (Findable – Accessible – Interoperable – Reusable) as suggested by the EU Horizon 2020 Program (2016). Data that resides in various systems, in different formats and ontologies, misses key attributes (such as unique identifiers), is not an Asset. If the Data Asset is not reusable, every data-science/AI activity will be a separate, possibly large IT exercise. The principle of “Build Once – Use Many” is pivotal for maximizing the value of Data Assets. For example, for the personalization of an online service, you might want to use behavioral data from the online and mobile channels, CRM data, and consumer online and offline transactions – not only data from the online service itself. The goal of a productized Data Asset is to support all use cases.

When building your company’s Data Asset, start with the data needed for the prioritized business opportunities/use cases. This sounds self-evident, but in many companies, there is an organizational disconnect between the IT teams that engineer data and the business functions. In the worst case, the Data Engineering teams are busy building a Data Asset integrating various data sources into a common data environment, but these are not the data sources the business would need. As a result, both sides end up frustrated.

One good practical way to start building a Data Asset is to take stock of the current data assessing how “FAIR” the data is. The process is called Data Due Diligence or Data Inventory. A Data Due Diligence responds to questions such as the following: What data exists? Where does it reside? How can it be accessed? What is its quality? Can it be linked to other data?  How much effort does its retrieval take?, and Do we miss some obvious data sources (considering the use cases)? Once the current state of the Data Asset has been identified, a roadmap for its development can be made.

Solution Architecture and Technology

Solution Architecture and Technology refer to the technical side of the Data Asset. Apart from digital native companies, existing companies typically have plenty of legacy infrastructure. One of the first tasks after the definition of the Business & AI vision and Data Due Diligence is to have an experienced Data and Solution Architect take a critical look into the current technical architecture and define the target architecture and its development roadmap. This task, too, should follow the end-to-end use case logic accounting for data collection from operating systems (e.g. CRM, ERP), data warehouses, cloud environments, analytical environments, and business-interfacing systems. Traditionally, coming from the BI world, many data solution architectures stop at the data warehouse level solving reporting use cases. However, automated Machine-Learning/AI solutions need to be linked back to operative systems meaning that operative systems need to be an integral part of the data and solution architecture. For example, to use your Consumer Data Asset (including individual micro-segments, next-best offers, and other consumer scores) in real time as part of a modern, omnichannel Marketing Automation system, you need to set up end-to-end architecture. Sometimes the marketing department oversees Marketing Technology while IT is responsible for backend systems. This may lead to Marketing only using marketing data (e.g. online, email) and discarding many other interesting data sources, because they are not aware of their existence. In the worst-case scenario, Data Science teams, which should create the company-wide ML/AI algorithms, are not involved in either activity.

Managing the transition from traditional IT systems into the digital world is often a lengthy process. While Automation and AI will eventually drive costs down, during the transition time, costs are likely to increase as new and old solutions live side by side. Furthermore, a typical IT department’s budgets are tied up with the operating and maintenance of current systems while development budgets are modest. New technical solutions require new investments.

Data Protection and Privacy

Data Protection and Privacy is of key interest to consumers and those with access to consumer data. Data protection relates to data collection, processing, and utilization. According to the General Data Protection Regulation (GDPR) of the European Union, the legitimate interest of data processing must be defined, and the user informed about the collection, processing, and combination of their data. The user must be offered mechanisms to opt out and object data processing. The level of user identification between data flows between different data-processing systems must be defined.

When using data and ensuring compliance with the privacy legislation, it is critical to write the privacy policies according to the desired state of the AI use cases, not the current cases. The point is to avoid a situation where your privacy policy allows you to use your data only for reporting purposes while you want to develop hyper-targeted, personalized AI models.

A good team for setting up the company-wide privacy policies consists of a combination of Business Owners, Privacy Lawyers, and AI Strategists (or Data Scientists). AI Strategists with a technical background will help translate the business use cases into Data and AI requirements and discuss the interpretations of different options (including User Experience) with Privacy Lawyers.

Human Skills

The Data and AI journey requires new roles in an organization. While the exact role terminology varies, Data and AI roles are needed for four different levels of business processes:

  • Business Units (P/L) and Business Functions (e.g. Sales, Marketing, Finance)
  • Data Science (and Business Intelligence)
  • Data Asset Management
  • Data Platforms and Technical Solutions

The business use cases come from the business level (#1). In addition to actual business people, the role of the AI Strategist resides here. The AI Strategist translates the business vision and goals into Data and AI requirements, oversees project execution, and ensures that project outcomes are taken into use by business processes. Most companies do not have this role, but we see it as one of the most critical roles in the successful execution of Data and AI projects. Without an AI Strategist, the communication distance between people with a business/engineering background and the Data Scientists is often too wide and can take some time to align. A good background for an AI Strategist is that of a Senior Data Scientist who wants to develop himself into the business and managerial talent. Over time, the AI Strategist will develop responsibility for AI Product Ownership tasks.

Some consulting firms present this role in the form of an “Analytics Translator”, and while this is similar to our definition, we place an emphasis on the role of the AI Strategist as a driver of business impact.

In addition to AI Strategists, business leaders themselves also need to have an understanding of the opportunities of Data and AI in order to drive the topic forward and integrate the AI outcomes into their respective business processes.

While most companies lack the role of the AI Strategist, many have hired Data Scientists (#2). Data Scientists come in various forms, with different backgrounds. As an educational background, many have studied quantitative methods such as computer science, mathematics, statistics, physics, or engineering. It makes sense to have a Data Science team with different types of educational backgrounds. For example, people with a Statistics or Econometrics background are good in statistical inference while people with a Computer Science background are proficient in machine-learning techniques and coding. Physicists are trained to work with physical phenomena and models and thinking outside the box. Sometimes data-savvy Sociologists, Psychologists, or Biologists can bring different perspectives to the team.

Since Machine Learning/AI is a new field, there is a large demand for experienced Data Scientists. We recommend recruiting a senior Data Scientist or an AI Strategist as the first hire and let them build a balanced team consisting of experienced people and promising young talent.

A common mistake is to hire only Data Scientists and not fill the technical roles such as Data Engineers and Data Architects (#3) or Platform Engineers and Solution Architects (#4). This leads to a high level of frustration among Data Scientists as they must retrieve data from the source systems and build the databases themselves. In practice, this often means that the Data Asset and Data Infrastructure will not be built properly. Data Science teams will do pilots and build point solutions, but a scalable data foundation will be out of reach. Data Scientists are trained to build machine-learning models, not to do Extract Transform and Load (ETL), databases, and cloud solutions. The frustrations will spill over to the management side as the pilots do not scale.

Nowadays, many people call themselves Data Scientists or Data Engineers. It may be difficult to distinguish walk from talk. In addition to looking at the education and past work experience of potential recruits in this field, we recommend using an assessment test in recruiting. Depending on the role at hand, it can be defined to emphasize various aspects of the work. In our own recruitment process, we use assessment tests, which reveal a whole lot more about the candidate than interviews only. You can define the assessment in a fun and challenging way so that the candidates enjoy and appreciate doing it.

A good start is to hire a Chief Data and AI Officer with experience in business, data, data science, and technology to hire the talent and set up the teams. In addition to subject-matter expertise, this person should have excellent leadership and communication skills as they need to communicate effectively with different levels of people in the organization.

Data and AI Organization

The optimal Data and AI organization structure depends on the overall company size and organization, culture, the level of AI maturity, and the type of Data/AI tasks.

Generally, to get things going, establishing a Center of Excellence (CoE) brings focus into the topic. Depending on where the CoE sits in a company, it will be responsible for different areas. The CoE may consist of Data Science & BI teams only while the technical teams (Data Engineering, Platforms) reside in IT. Alternatively, the CoE may cover the technology side while the Data Scientists sit in business units. The optimal setup needs to be carefully investigated. In our experience, most companies will benefit from a common technical infrastructure and Data Asset, as well as some type of a centralized Data Science team, which solves the most difficult use cases and creates a scalable AI portfolio for the use of all business units and functions.

The AI Strategists should optimally sit within business units to drive the AI use cases forward, but in the beginning, they can also reside with to the Data Science teams and help business from there.

In a mature Data/AI-driven company, the role of the CoE will become smaller as the whole company uses data in their daily business. At a mature stage, the CoE will continue taking care of common data governance topics such as data quality and integrity, technical systems, ontologies and standards.

Sometimes, to make starting easy, it makes sense to introduce a company-wide AI Program to drive the Data and AI agenda forward – with the premise that the program exists for 2-3 years and will then be dissolved. A program’s benefit is that you do not need to make line-organizational decisions in the early phase, but will learn over time what type of team structure works for your organization.

Operating Model

A closely related topic to Data and AI Organization is the Operating Model between different business units. Prioritized business use cases should drive the Data and AI development. In order to have the Data Experts work on the most important use cases, business leaders should establish an AI Steering Group or include the Data and AI development into the existing leadership team meetings. The Head of the CoE should drive the agenda in the meetings. In addition to a cross-unit steering group, individual use case areas should have their own, operational steering groups.

For the first years of the CoE, Data and AI development budgets should be centralized. Budgets drive prioritization, and without a centralized budget, Data and AI activities will not scale up. Individual business units do not want to carry the costs for company-wide capability building (e.g. common data models, infrastructure, APIs) even if it would be optimal for the whole company. This means that the AI solutions tend to become separate, non-connected islands. Furthermore, without a common roadmap, prioritization and clear governance, businesses that contribute the most financially will demand that they get the resources even if resources would be strategically better used in another area.

It is important to remember that it is not only about the data experts. Business processes and business people are fundamentally impacted by the utilization of Data and AI, for example, in Marketing, to drive trigger-based, personalized marketing, data and targeting models need to be available, but so do marketing content production, customer treatment models, channel strategies, front-end systems, and so on. Always-on, data-driven marketing requires different skills and capabilities from marketers than traditional marketing. Similarly for process automation: If Data Scientists build a predictive service maintenance model, business impact will only be obtained if the service fleet and technical systems are enabled to perform timely intervention to respond to the predictions.

A smart way to increase the business impact is to give everyone included in a Data/AI project the same incentives. For example, if the goal is to increase the marketing campaign lift by 20 percent with AI-driven targeting, this target should be given to Marketers, Data Scientists, and Data Engineers. This is likely to prompt some objections but will eventually lead to the best results for the company.

Data Science and Machine Learning/AI Algorithms

Like the Data Asset, algorithms can also be treated as an Algorithm Asset. That means that over time, the portfolio of machine-learning/AI algorithms will become FAIR. Every new analytical modeling exercise does not need to start from scratch but builds on top of tested code. This will make the Data Science team more efficient over time. Like software coding teams, it requires the Data Science team to use common code repositories and standards.

It is also important to establish maintenance processes for the Data and Algorithm Assets. If maintenance processes remain undeployed, development teams remain in a state of stagnation as their efforts go into keeping the current Assets in production. By applying maintenance processes to data and algorithm portfolios, new solutions can be discovered and developed.

Final Words

To summarize, the following steps are needed to execute Data and AI successfully:

  1. Formulate your Data and AI use cases, based on your business priorities;
  2. Understand the current state of your Data and AI projects and enablers;
  3. Define your Data and AI vision and the execution roadmap, including investments;
  4. Execute the first use cases aiming at production readiness;
  5. Scale up operations.

Sometimes people think that the highest level of an AI-driven company is when all decision-making and business processes are done by automated algorithms. That is, however, a misconception. Automation and AI do not make smart business decisions by themselves. The highest level of AI maturity is when the whole company goes into one direction, silos are dissolved, and Data and AI are used by everyone as part of their daily business. Everything that can be automated will be automated, and humans need to ensure that automation is done in a smart way.

 

Ulla Kruhse-Lehtonen is the CEO, Co-founder and Partner of DAIN Studios in Finland.

Dirk Hofmann is the CEO, Co-founder and Partner of DAIN Studios in Germany.