Enterprise AI at Scale: The Data Product Strategy

Leading tech players such as Netflix, Amazon and Uber use data science and machine learning at scale in all of their core business processes, but many other companies still struggle to expand their projects beyond a small pilot scope.

In this article, the authors of Driving Digital Transformation through Data and AI discuss what it takes to bring your machine learning to the next level of maturity and redesign business strategy for the age of AI. 

The current state of Artificial Intelligence (AI) is powerful enough to automate, optimize and redesign a major part of critical decision making and to enable new disruptive business models.

The data product strategy examines how profoundly data and AI will disrupt the market your company operates in and defines the path for a successful digital transformation through data and AI.

Customer expectations, products, distribution channels, supply chains and competitors are changing rapidly. Your business model and strategic positioning in the marketplace must be adapted to these new realities.

The data product strategy is at the heart of AI strategy

In essence, a data product strategy sets the focus of where data products will contribute to your company's success and need to be fully aligned with the vision and priorities set by your business strategy.

Oftentimes, investments in machine learning are too thinly spread over too many business areas and not well aligned with where the business wants to go. Without a data product strategy, one can quickly lose attention spreading resources across projects without creating value and synergies.

The data product strategy provides a high-level view on focus areas or data product domains for data products that should be built by determining the business priorities for the data product portfolio. Data product domains can be tied to particular business functions (e.g. sales and marketing function) or to particular company objectives and key results that are cross-functional by nature (e.g. managing quality).

The data product strategy provides the foundation for the business and AI strategy since it provides a pathway on how business value is generated and towards adopting the new business model and realizing the reinvented business vision.

Data products consist of more than data and machine learning

A lot of organizations develop algorithms that never leave the lab environment. A machine learning model provides little value unless incorporated into a holistic deployable and scalable product that solves an actual business problem or improves customer experience.

Data products are software applications with a strong focus on data science and machine learning and can consist of other frontend and backend software functionalities. The goal of data products is to provide an overall solution to a business problem and therefore there are many feedback loops and interactions with the customers and users when designing and building the product. They can directly help achieve company OKRs of the business vision and strategy, such as achieving a greater customer experience, boosting online sales volumes, driving profitability and changing the cost to sales ratio.

Implementing a data product often requires changes in business processes that utilize the data product. Every data product has a product lifecycle which follows several phases from design, prototyping, development and deployment, operation and maintenance to end of life. After deployment, the data product might be potentially scaled to other business areas.

A rigid product lifecycle is run by agile data product teams

The best way to implement the data product strategy is to set up data product teams that focus on a particular data product domain and consist of the right mix of roles to fulfill the mission of that respective data product domain.

The data product strategy should determine which data product teams are set up and what their mission and objectives should be. Data product teams should be cross-functional, dedicated, self-sufficient, and empowered with a focus on business problems. Data products must be managed in a disciplined way throughout the different phases of the data product lifecycle from design to delivery and operations.

The design of data products starts with identifying data product ideas in the business domains of the data product strategy and prioritizing them based on business value, scalability, feasibility and contribution towards the overall strategy and vision and on their potential to scale across different processes or markets.

The entire process is like a big funnel where a lot of data product idea candidates come in at the top and become fewer and fewer with each step. At the same time, the likelihood rises with each step of a data product idea to actually work and have a sufficiently strong business impact. During the development and deployment phase, the data product is implemented and documented in a series of sprints. After go-live, the data product is operated and maintained in DevOps mode.

Roll out of data products should be a priority for enterprise AI at scale

Often, the value of the data product can be multiplied if it is rolled out to other markets or business scopes. Scaling of data products is therefore a key value driver that requires special attention and resource planning.

When evaluating data product ideas and creating the roadmaps for the data product domains, teams should always look out for the sweet spots: data products that perfectly match your strategy and can be easily implemented, scaled and rolled out to other business scopes to increase business value.

In many cases, there are substantial synergies and value can be generated faster due to similar data models and target variables. This would include the adaptation of data products to fit other parts of the business and expanding the functionality of data products to solve more business problems in the same business domain. For example, data models and the business logic might be different in another production plant or market, and, hence, the model features, machine learning models and data preprocessing pipelines have to be changed accordingly to make the data product usable in the new scope.

Rollout managers can be hired to increase the implementation speed and support change management during the data product scaling phase.

A data product strategy needs foundational capabilities and a data-driven culture

In summary, data product strategy begins with redefining the business strategy to react to the impacts of AI in your industry.

Your data product portfolio is then defined based on your business strategy and managed by corresponding cross-functional and autonomous data product teams with a long-term mission.

Delivering the data product strategy needs further capabilities, in particular, a company-wide data platform and processes for data governance and data quality management.

To reap the benefits of their data product strategy, organizations will need to nourish a data-driven culture and clarify their position on AI ethics to create trust with employees and customers, and make data fun to work with for everyone.

At the end, only strong visionary leadership can make all of this happen by sharing a strong vision and leading by example, providing the right level of education and training, and providing an innovative ground for all employees to own their part of the transformation.