Why your AI project is also a data project

This global evolution is reflected in the Belgian business landscape too. Although only 6% of Benelux companies so far have fully integrated GenAI into their processes, 87% are planning to invest in it. In the top 20 of the Beltug Priorities Compass 2024, no fewer than eight priorities are linked to AI and data governance. “CIOs are clearly aware of the importance and complexity associated with AI and data governance: good data quality and a thoughtful IT architecture allow companies to use data to its full potential,” Yashfeen says.

The rapid proliferation of often very public AI cases creates a sense of urgency. Yashfeen: “In my opinion, this is where a major pitfall immediately lies and it seems like businesses think that there is no time to waste. That sentiment risks companies making a mistake. The key is to find the right use cases. AI is never an end in itself, but is just a means to achieve business objectives. Those objectives can focus on generating revenue growth, rationalizing processes or improving service delivery and customer experience. The ideal scenario involves design thinking, where business and IT shape ideas together.”

“Artificial intelligence is not just about the implementation of an AI model. I see it as a story of infrastructure, data, applications, platforms, AI services and people.” Before starting an AI project, therefore, some crucial questions arise:

- What data is available in our company?

- Through which AI models can we train our data?

- Is our infrastructure ready for AI?

- How do we integrate an AI application within our business ecosystem?

- How can we structure and secure our data?

- How do we ensure good governance and compliance, especially in case of sector-specific rules?
- …

“Companies have to make the choice of developing an AI application in-house or to use an off-the-shelf solution available on the market today. Of course, you would do well to pick the low-hanging fruit first, but even then it can feel very complex. A proprietary solution naturally poses even more of a challenge,” Yashfeen says.

The following four principles then come to the fore: automation, scalability, governance and ownership.

1. Without automation, an IT infrastructure would not evolve quickly enough with the fast-changing business environment.

2. Scalability is needed so that the IT infrastructure is future-proof and remains efficient, regardless of data volumes.

3. Governance and security, in turn, protect against cyber threats, ensure regulatory compliance around AI and guard against ethical AI.

4. Finally, data ownership promotes internal collaboration and accountability: essential for delivering accurate insights through AI.

The dynamics in generative AI are very different from those in classic AI projects. Thanks to tools like ChatGPT and Copilot, AI is no longer the exclusive domain of the data scientist and the choice between an off-the-shelf AI solution or in-house development also depends, of course, on how adept you are as a company in terms of data and AI.

‘We see three maturity levels in the market. Firstly, you have those companies that are starting a first AI project and have no previous experience in selecting the necessary use cases and tools to fall back on. Then you have the companies that have already defined their AI case and where collecting, structuring and integrating large data volumes is on the agenda. Finally, you have the companies with AI projects of advanced status, where the key is to generate reliable results and align them with data governance, IT security, ethics and legal requirements.’

Robust data management is considered a crucial prerequisite for project success at any maturity level. Structured, secure, scalable and reliable data are crucial. There are three transformation steps to getting there.

1. The first priority is setting up the right IT infrastructure and making data accessible and usable for AI tools.

2. Next, you need a data platform to enable AI integration that meets cybersecurity, data and privacy protection and all compliance requirements.

3. Only when those conditions are met does the development of AI services in any form begin, ranging from GenAI to forecasting models.

“Success requires a wide range of employee profiles: from design thinkers, to data wizards, to architects and adoption specialists, and those very profiles are the ones that are the most difficult to find. That means that co-creation often is the way to go to achieve a successful project,” Yashfeen says.

“Regardless of the business case you develop around an AI application, I recommend including the legal aspect in your thinking from the start. By applying the principles of responsible AI, the AI application immediately meets the requirements defined by/(or: set by) the European AI Act. Focus in responsible AI projects is naturally on transparency, inclusiveness, fairness, security and privacy.”

“Artificial intelligence is making its way through all sectors at an impressive rate and impacting very diverse business activities. So the chances are considerable that your business too will benefit from using AI but it is also important to take a measured approach. This implies a clear definition of the business case, quality data management and attention to all legislative and security aspects,” Yashfeen concludes.