Successful AI requires the right data architecture – here’s how


For companies that can master it, artificial intelligence (AI) promises cost savings, a competitive advantage and a foothold in the future of business. But while the rate of AI adoption continues to increase, investment levels are often out of step with monetary returns. To be successful with AI, you need the right data architecture. In this article you will learn how.

Currently, only 26% of AI initiatives are scaled from an organization to mainstream production. Unfortunately, this means many companies spend significant time on AI deployments without seeing any tangible ROI.

All companies need to act like a technology company

In a world where every company must behave like a technology company to stay ahead, there is increasing pressure on technical teams and engineering and IT leaders to use data for commercial growth. Especially as spending on cloud storage increases, organizations are striving to improve efficiency and maximize ROI on data that is costly to store. But unfortunately they don’t have the luxury of time.

To meet this demand for quick results, the mapping data architecture can no longer stretch for months without a defined goal. At the same time, the focus on standard data cleansing or business intelligence (BI) reports is regressive.

Tech leaders need to build a data architecture that puts AI at the forefront of their goals.

Otherwise they will have to retrofit it later. In today’s enterprise, data architecture should work towards a defined outcome – and that outcome should include AI applications with clear end-user benefits. This is the key to setting up your business for future success, even if you are not (yet) ready for AI.

Start from scratch? Start with data best practices

Data architecture requires knowledge. There are many tools out there, and how you put them together depends on your business and goals. The starting point is always a literature review to understand what has worked for similar companies, as well as a deep dive into the tools you are considering and their use cases.

Microsoft has a good repository for data models, as well as a lot of literature on data best practices. There are also some great books that can help you take a more strategic, business-centric approach to data architecture.

prediction engines by Ajay Agarwal, Joshua Gans, and Avi Goldfarb is ideal for understanding AI at a more fundamental level, with functional insights into using AI and data for efficient operations. Finally, I recommend for more experienced engineers and technical experts Designing data-intensive applications by Martin Kleppmann. This book gives you the latest in this field with actionable guidance on building data applications, architectures, and strategies.

Three fundamentals for a successful data architecture

Several core principles will help you design a data architecture capable of powering AI applications with ROI. Think of the following as compass points to measure yourself against as you create, format, and organize data:

  • Working towards a goal:

    Keeping the business outcome you are working towards in mind as you build and evolve your data architecture is the basic rule. In particular, I recommend looking at your organization’s short-term goals and aligning your data strategy with them.

    For example, if your business strategy is to achieve $30 million in sales by the end of the year, find out how you can use data to drive it. It doesn’t have to be daunting: Break down the bigger goal into smaller goals and work towards them.

  • Design for quick value:

    While setting a clear goal is crucial, the end solution must always be agile enough to adapt to changing business needs. For example, small projects can become multi-channel projects and you need to keep that in mind. Fixed modeling and fixed rules will only cause more work later.

    Any architecture you design should be able to take in more data as it becomes available and use that data to drive your business’s latest goals. I also recommend automating as much as possible. This will help you quickly and repeatedly deliver valuable business impact with your data strategy over time.

    For example, automate this process from the start if you know you need to deliver monthly reports. That way you only spend time on it in the first month. From there, the impact will be consistently efficient and positive.

  • Know how to test for success:

    To stay on track, it’s important to know if your data architecture is working effectively. Data architecture works when it can (1) support AI and (2) deliver actionable, relevant data to everyone in the organization. By following these guidelines closely, you can ensure your data strategy is fit for purpose and future-proof.

The future of data architecture: innovations you should know about

While these key principles are a good starting point for technical leaders and teams, it’s also important not to get stuck in one particular approach. Otherwise, companies risk missing out on opportunities that could offer even greater value over the long term. Instead, technology leaders need to be constantly engaged with the new technologies that are emerging that can improve their work and deliver better outcomes for their business:

  • Cheaper processing:

    We’re already seeing innovations that make processing more cost-effective. This is crucial as many of the advanced technologies being developed require such high computing power that they only exist in theory. Neural networks are a prime example. But the more the required computing power is feasible, the more sophisticated we can solve problems.

    For example, a data scientist must train each machine learning model. But in the future there is an opportunity to build models that can train other models. Of course, this is still just a theory, but we’ll definitely see innovations like this accelerate as computing power becomes more accessible.

  • Bundled tools:

    When it comes to apps or software that can reduce time-to-value for AI, we’re now in a phase where most available technologies are only good at one thing. The tools required to produce AI—such as storage, machine learning providers, API delivery, and quality control—are being unbundled.

    Today, organizations run the risk of wasting valuable time figuring out what tools they need and how to integrate them. However, technologies are emerging that can help solve multiple use cases for data architectures, as well as databases specialized to support AI applications.

    These more bundled offerings will help companies get AI into production faster. It’s similar to what we’ve seen in the fintech space. Companies initially focused on being the best in a core competency before eventually merging into bundled solutions.

  • Data marts vs data warehouses:

    As we look further into the future, it seems safe to predict that data lakes will become the most important investment in AI and data stacks for all businesses. Data lakes help companies understand predictions and best implement those insights. I see data marts becoming more and more valuable in the future.

    Marts provide every team in an organization with the same data in a format they can understand. For example, marketing and finance teams see the same data in familiar metrics and, most importantly, in a format they can use. The new generation of data marts will have more than dimensions, facts and hierarchies. Not only will they share information, but they will also support decision-making in specific departments.


As technology advances, it is crucial that businesses stay current or they will be left behind. That means technology leaders stay connected with their teams and enable them to bring new innovations to the table.

Even as an organization’s data architecture and AI applications become more robust, it’s important to take the time to experiment, learn, and (ultimately) innovate.

Photo credit: by Polina Zimmerman; Pixel; Thanks very much!

Atul Sharma

Atul co-founded decision intelligence company Peak with Richard Potter and David Leitch in 2015. He has played a pivotal role in shaping Peak’s Decision Intelligence platform, which has emerged as an early leader in what is expected to be the largest technology movement for a generation. Peak’s platform is used by leading brands such as Nike, Pepsico, KFC and Sika. With a mission to change the way the world works, the tech scaleup has grown rapidly over the past seven years and now has over 250 employees worldwide. Regularly recognized as a top place to work in the UK, Peak was awarded 3-star Best Companies accreditation this year, recognizing exceptional levels of employee engagement. Prior to Peak, Atul worked in data architecture and data engineering for over 20 years. He has worked on the design and implementation of data integration and data warehouse projects for global companies such as Morrisons Plc, The Economist, HBOS, Admin Re (part of Swiss Re) and Shell.


About Author

Comments are closed.