Philip Morris International [NYSE: PM]

An avalanche of discussions related to AI and Machine Learning took over social media and the news. All of a sudden, journalists and thought leaders feel they need to have an opinion. Meanwhile, organizations of all sectors are under pressure to show they are "using" AI. Nonetheless, implementing Data Analytics and AI in a way that delivers real business value requires discipline and structure. This article discusses the three fundamental elements which are the bedrock of a successful practice: firstly, start with the business priorities and decision-making; secondly, establish a data foundation; and thirdly, build a data product ecosystem.

#1 Start with the business priorities and decision-making.

A fictitious hospitality firm decided to be more data-driven and invested millions of dollars in hiring an army of data scientists and state-of-the-art technology. Unfortunately, after 18 months, half of them had left without access to the data they needed to build their models, while those who remained had been working on some highly complex projects for which the liaison with real business questions was nebulous.

A recurrent misstep in establishing a data strategy is to begin with the data and technology. The starting point should be the business goals and the decision process. Data has no intrinsic value; it only has impact when influencing the way people decide. For example, imagine a supply chain manager responsible for deciding every month the ideal quantity of products to be manufactured. For her, one of the key business questions is: How many units will be sold next month? Hence, she may think of building a demand forecasting model. But does she have the quality data she needs? This question refers to the next pillar.

#2 Establish a strong data foundation.

Another organization (also fictional) decided it needed the best AI tool to generate product recommendations for its consumers. However, after some tests, they realized the model outcomes did not make any sense. The team first checked the tool, but it was working flawlessly. Going back into the data process, they realized that the data ingested into the AI system was not clean and had a lot of duplicates. In addition, they also detected problems in the way they were collecting data in the first place. Therefore, they had just realized their data was not "AI-ready".

Data needs to be structured in a way that is transparent and organized. The firm has to generate clean, secure, reliable, and accessible information, so that authorized users be able to use it confidently to build their analysis, dashboards, reports, and models.

  ​Successfully implementing AI and analytics requires structure and discipline and encompasses three pillars: addressing priority business goals and decisions, setting up a data foundation, and establishing a data product ecosystem.   

Data foundation refers to how the enterprise manages and governs its data across the entire data lifecycle. That comprises how data is extracted from transactional systems and loaded into a data lake, data warehouse or relational database. It also encompasses how data is transformed, cleaned, aggregated, and organized. Finally, and very importantly, it includes Data Governance, which pertains to the policies, processes, and responsibilities through which the company defines how it produces, processes, and uses data and technology. To illustrate this point, we refer to the issue Samsung Electronics faced in 2023. After an accidental leak of sensitive source code by an engineer who uploaded data to ChatGPT, the company subsequently banned the use of generative AI tools. This case reinforces how crucial it is that corporations establish effective policies and procedures to define how they use data and, more specifically, how they utilize AI.

#3 Building a data product ecosystem.

It is time to circle back and address the priority business decisions with a portfolio of data products. These are data artifacts that take raw data, transform it, and display the information in a way that users can utilize to make decisions. They may be built in different forms, including reports, dashboards, and AI models, fit for purpose to specific business processes and users.

Each data product should have a data product owner who is responsible for its entire data pipeline and the front-end used by decision-makers. The data product owner will check for data quality and guarantee that the UI and UX are suitable for users' needs.

A good example of a data product is the Disney World consumer App, available for the Disney parks' visitors, to help them make a myriad of decisions, from booking a restaurant table to finding their next ride. The App helps consumers - who are oftentimes tired and caring for children—to find what they need and to make their decisions quickly and confidently. This data product—and the recommendation systems running in its background—add value to Disney's consumer experience, which is paramount to the company's success in the long run.

In conclusion, successfully implementing AI and analytics requires structure and discipline and encompasses three pillars: addressing priority business goals and decisions setting up a data foundation, and and establishing a data product ecosystem. An enterprise-wide focus on these underpinnings will lead to a consistent pathway towards a more agile and efficient decision process and, hence, a nimbler and stronger organization.