If you want to achieve more with AI and data, you need to start with the basics. With a data architecture you lay the foundation for making optimal use of your data.

Nowadays not a day goes by when we don’t talk about artificial intelligence. New technologies are constantly being announced and industry experts are always talking about the change that (generative) AI will bring to companies. All of this beauty can also prompt your company to do more with AI.

Christophe Robyns, managing partner of Brussels-based Agilytic, calls for people not to blindly jump on the hype. “Companies today have high, sometimes unrealistic expectations of AI. So you will also use AI in projects where it is not actually necessary and simpler alternatives would offer just as much added value. As a result, many projects do not achieve the expected impact.”

Consulting firm McKinsey has insisted for years that the lack of a data strategy and supporting architecture are the biggest obstacles to achieving digital transformation goals. Just as a house cannot stand without a solid foundation, data cannot stand without a good foundation. With the help of Robyns, we’ll discuss what you should pay attention to when building a data architecture.

Blueprint of data

You can think of data architecture as a blueprint of your business data. They record every movement that data makes within your organization, from the capture of raw data to its use in concrete practical situations. Robyns points to artificial intelligence to illustrate the importance of architecture:

“It is possible not to have an architecture for experimentation with prototypes, but if you want to integrate AI deeper into your organization, it is essential.” Models are very sensitive to variations in the data. Therefore, it is necessary to continuously monitor the quality of the data. Developing AI is not that difficult in itself, but keeping it alive is,” says Robyns.

A good data architecture grows with your company, emphasizes Robyns. “The needs of your organization will change over time. Versatility and scalability are critical requirements for a modern architecture so that your data grows with your business.

Developing AI itself is not that difficult, but keeping it alive is.

Chris Robyns, Agilytic

Four dimensions

For Robyns, there are four dimensions you should thoroughly explore before embarking on a data or AI project. The answers to these dimensions will shape architecture. “It all starts with the business goal: What do you want to achieve with your data? “That sounds like a logical starting question, but it’s overlooked more often than you think,” says Robyns. “Often people have something in mind in advance, so alternatives are not considered.”

Only then does the technological picture come. “Now you have to ask yourself what tools you will use and, above all, what you already have at home. There’s no point in replacing something that still works well,” Robyns continues.

He doesn’t shy away from challenging some technology clichés. “There are many different types of data platforms, some more complex than others. A simple platform is often a very good solution when the business strategy supports the data strategy. Companies also tend to think they need real-time data for everything, but that’s not necessarily the case.”

You also have to think outside the technological box. Data Office is also an important part of the data architecture. As a company operating in the European Union, you are subject to strict rules about where and how you store customer data and how long it can be retained after processing. Or as Robyns puts it: “Just because you have the infrastructure doesn’t mean you can just let it run. You must carefully document all data flows.”

Finally, there is the organizational aspect. While this isn’t something you’ll find in every manual, Robyns says it’s necessary to pay enough attention to it. “A data project is not an everyday IT project. Different disciplines often come together. You shouldn’t underestimate the impact on your teams. It’s not enough to be an expert in data, you also need to be an expert in transformation to reduce conflict between teams.”

To recap, these are the four dimensions a data architecture should be built around:

1. Business: What do you want to do with data? What business problems do you want to solve?
2. Technology: What tools do you want to use to achieve this goal and what do you already have in-house?
3. Governance: What are you allowed to do with your data? How should you store and process this data?
4. Organization: How will you organize your teams for the project?

Five layers

Because data architecture is based on specific use cases, there is no single right way to develop it. Every project requires unique architecture. However, some components can be distinguished that almost always recur. Robyns lists a few for us:

• Data pipelines: the process of processing data. Data pipelines retrieve data from the source and deliver it to the destination. These affect different types of data (structured and unstructured) and different storage sources (cloud, on-premise, etc.).
• Storage: the place where you store data at rest.
• Orchestration: The coordination and management of data operations and processes.
• Access and security: The goal of data architecture is to break down silos. A modern data architecture ensures that everyone has sufficient access to the data.
• Connectivity: Platforms must be able to interact with the environment to retrieve existing data or transfer enriched data.

Invisible power

Experts sometimes say that architecture is more important than the data itself. Data architecture is the invisible, silent force of your business. This is precisely why Robyns has found it difficult to convince companies, although he sees the message is slowly getting through. “If you don’t immediately see the immediate added value, it can take a little longer to commit to something. When building a data platform, we also try to set up as many concrete use cases as possible to show the added value for the company.”

Without data there is no AI, but without architecture there is also no usable, reliable data. That’s the lesson we learn from our conversation with Chris Robyns. Or how the saying “look before you leap” always remains relevant, even with the latest technologies.

“A data project is not an everyday IT project. You shouldn’t underestimate the impact on your teams.”

Chris Robyns, Agilytic