Machine data revolution: feeding the machine

The need for a new data architecture

It’s no secret that new data-driven technologies such as cloud, AI and ML have
tremendous opportunities to unleash the digital potential of the banking industry. “AI and ML technologies can create automated, real-time and at-scale decisions, which can dramatically improve results for the banks and their customers,” says René Theunissen, partner at Deloitte Consulting specialised in enterprise technology for banking. “As financial institutions collect lots of data, and have been collecting data for a long time, they are – in theory – well positioned to deploy data-driven technologies.”

Yet at most banks, the enterprise data architecture is not designed to support rapid and consistent development of AI and ML. “Banks often rely on legacy technology and data models. Their data architecture is siloed and designed to support human decisions,” explains Benedikt Kratz, director at Deloitte Consulting with a focus on data strategy, architecture and management in the financial industry. “If banks want to become truly data-driven, they have to implement automated, machine-based decisions in their primary processes,” Kratz continues. “This entails reorganising their data architecture end-to-end.”

Human vs. machine decision-making

Banks currently perform lots of data analytics, but mostly to support human decision-making. The data architecture of banks reflects this way of working. “Data is often organised in manually crafted spreadsheets with clean tables and rows,” explains Kratz. “This is convenient for humans but not optimal for the deployment of ML and AI models.” Moreover, data is stored in legacy systems that have been used for a long time and is siloed in systems that do not interact.

Machines, by contrast, can extract low levels of statistical significance across massive volumes of structured and unstructured data. They work around the clock and can make decisions at scale and in real time. “AI and ML allows banks to automate parts of their primary processes,” says Theunissen. “For instance, AI and ML models can make real-time decisions on whether a loan or a mortgage should be granted. This asks for a new data architecture, in which data is easily available in open-standards across all divisions.”

Unlocking the bank’s data goldmine

Fintech start-ups enter the market without the burden of legacy systems and can optimally profit from cloud, AI and ML technologies. However, they do not have the scale, experience and amount of (historic) data that can provide a deep
understanding of customers and their needs. “Traditional banks are sitting on heaps of potentially valuable data,” says Kratz. “Now, it’s time to unlock their data goldmine by enabling themselves to put AI and ML models at work.”

Reengineering data value chains to support AI and ML’s possibilities is a complex task that can take up years, says Theunissen. “It touches every part of the value chain, including how you capture, store and process data.” New technologies and approaches can support this process, including advanced data capture and structuring capabilities, next-generation cloud-based data stores, and analytics to identify connections among random data. Together, these tools and techniques can help organisations turn growing volumes of data into a valuable asset to support automated decision-making.

Fundamental transformation, step by step

Reorganising the data architecture of banks requires a clear strategy, planning and management. “A first step is to understand what kind of data you have, how it’s stored, what the quality is and how it can be valuable,” says Kratz. “This is already a massive undertaking. For instance, ABN AMRO is in the midst of modernising their data architecture, and they discovered that they are currently deploying more than 3,000 systems.”

The next step is to prioritise: what parts of the data architecture need to be updated first, and why? “You cannot do it all at once, so you have to decide where to start,” says Kratz, who adds that compliance and business value are particularly important in this respect. Subsequently, banks have to define the data and ensure that it can be reused in different platforms.

Last but not least, banks need to ensure that ethics are embedded in their new data architecture and that adequate governance is in place. “AI and ML models work with estimations; their outcomes need to be ‘good enough’,” explains Theunissen. “Although they can make decisions at scale, it doesn’t mean that the results are flawless.” Banks make impactful decisions, such as who gets access to loans and on what terms, and they need to be able to explain and defend their decisions. This can be done implementing these requirements in the design of the new data architecture. 

What’s next

There are lots of examples of banks already deploying AI and ML models. Theunissen mentions an example of a tool to assess whether an SME is eligible for a loan, which reduced the processing time from a week to only 15 minutes. “Tools like these are disrupting the market,” he says. “However, these are mostly small initiatives and often do not transcend the proof-of-concept phase. Now, it’s time to create a futureproof data architecture that enables deploying AI and ML models at scale.”

Reorganising the data value chain to get ready for the AI and ML revolution is a long, complex, but exciting journey. “The current data architecture of banks brings a lot of hidden costs in manual labour,” says Theunissen. “Modernising the data architecture is a massive undertaking. But in the end, it will bring enormous benefits in the form of efficiency, cost reductions and smarter decisions.”