Is AI included in the data organization or is it separate? by Richard Benjamins

First of all, data is the basis for analytics and is fuelling many AI applications. Indeed, the dominant AI paradigm in business today is data-driven AI, also known as machine learning. Machine learning takes as input large amounts of historical data and generates a program or model that’s able to make predictions on what will happen, for instance, or perform a classification based on new data that the model has never seen before. It might be used to determine which customers will switch to the competition, what components will break down in an industrial plant, what movie to recommend, or whether a particular patient is suffer- ing from a certain disease. Machine learning is therefore dependent on the availability of quality data.

AI is broader than just machine learning. There is NLP, Knowledge Representation & Reasoning, Planning and Robotics. And while several of these areas have enjoyed breakthroughs in the past few years thanks to machine learning, they aren’t only about machine learning.

For the machine learning type of AI, it’s important that the AI department is not too far from the data department (the CDO), especially at the start of the data journey, when frequent interactions are necessary. In this sense, the same principles behind the choices for data and IT dis- cussed in the previous chapter also apply to how to organize data, analytics and AI. That means that these three areas should report into the same senior executive, and preferably have the same direct manager. The reason is simple: if there are issues between the different areas, the escalation process is straightforward and fast, speeding up resolution. Conversely, having the data, analytics and AI in organizations that report to different senior executives is a recipe for problems and even failure. Different senior execs likely have different tangible objectives, and what might be essential for the CMO might be less relevant for the CIO at a particular time. That might put differing emphasis on the priorities of one team or the other, and they may be out of sync. Solving such a problem requires escalating it all the way up to the board, which takes time and effort, generates frustration and delays progress on the digital transformation.

Of course, if the focus of an AI organization is not on machine learning, but on the other, non-data intensive areas, it doesn’t hurt to have it in a separate organization, as there is less dependency. As soon as data starts becomIng the main driver, though, they should be put together.

Another reason these three areas (data, analytics, AI) should be close together is that these technologies are still relatively new in the business world, and organizations lack experience in how they can work together. As such, mistakes may be made in each distinct area, requiring backtracking on earlier decisions; this backtracking has an impact on what was communicated earlier to the other areas.

As we will see, for many decisions, the data maturity of an organization is an important factor in making the best decision. In other words, depending on how data-mature an organization is, how a particular decision is made is less or more relevant. For relatively immature organizations — those that have recently started this journey — it’s crucial to have the AI and data departments close together, as we have argued here. However, for more mature organizations with significant experience, this is less of an issue. Data-mature organizations have data management and governance in place, ensuring a certain quality and frequency of the data. If the data is good, AI and analytics areas can work on their own. In more data-immature organizations, data availability and quality are not well-organized processes, and much faster, more agile interactions are needed to obtain results.

DO YOU NEED DATA SCIENTISTS OR DATA ENGINEERS?

The rise of big data has been accompanied by the ascent of the data scientist, which has been called the sexiest job of the 21st century (Davenport and Patil, 2012). This motivated many data professionals to profile themselves as data scientists, and has led to the hiring of an abundance of data scientists by organizations looking to strengthen their data team. In reality, however, the majority of effort in a data project is dedicated to accessing and understand- ing the data and verifying its quality, with a much smaller part dedicated to the analytics or machine learning.

This is a logical consequence of the overall low data maturity of organizations in the past few years. Only when data is fully managed as an asset can full attention be devoted to value creation with analytics and machine learning. But today, such organizations (mostly big tech companies) rep- resent only a small percentage. This trend has led to many professionals hired as data scientists working at accessing and manipulating data for the sake of data, rather than creating value, and this has led to much frustration.

The lesson learned here is that, in practice, organizations need to hire the right balance of data engineers and data scientists. At the beginning of the data journey, there should be more data engineers than data scientists, and with increasing data maturity the balance can shift in favour of the data scientists … but not earlier!

This phenomenon also has important implications for groups within the different data departments. For organizations starting their data and AI journey, it would be a mistake to separate the data and analytics/AI areas. The data department would do its best to collect, store, organize and make data available, but given the early stage, this process will likely take a long time and generate doubtful data quality. This leads to frustration in the analytics/AI team: they’re waiting for data, and once it is there, it has a lot of problems and they need to refer back to the data department. The further away the departments are in the organizational structure, the more frustration there will be, which will negatively impact the collaboration. As discussed in the previous chapter, the optimal positioning is to have the departments report to the same line man- ager. If this isn’t possible, one of the alternate solutions we suggested may make sense, such as co-location or dotted line reporting.