From lab to factory: how can AI solutions be scaled effectively in banking? 

Artificial intelligence (AI) is a key technology for the future of banking. As with other important digital technologies, the gains made in terms of speed and efficiency by implementing initial use cases have quickly raised the question of how scalable AI solutions are. Agile methods such as DevOps and focused AI labs offer a means of developing solution strategies and addressing uncertainties arising from internal factors, regulatory requirements and ethical considerations.

The rollout and use of AI in banks undoubtedly represent a major technological and organisational paradigm shift. In compliance and customer identification, for example, AI can help to increase speed and efficiency and improve the quality of reporting by pinpointing the data that are relevant to processes and decisions and finding patterns in these data. However, human oversight over AI solutions and their output, including random-sample checks of that output by staff, are vital to the concept of responsible AI.

Mastering the transformation and challenges

Besides the challenges with regard to customers, data (protection), technology, staff skills and regulation, the scaling of AI in business processes, products and services is difficult both to predict and to control, but it can play a decisive role in successful project and implementation outcomes. There is a great deal of uncertainty surrounding the feasibility of and the correct approach for each AI solution, and the limited availability of some resources and skills is a barrier to productive deployment. On top of this, banks’ efforts to scale up AI must take account of ever-changing customer preferences and expectations, frequently inadequate data quality and availability due to in-house data silos, and legacy data and analytics architectures.

Data and AI governance as a success factor

In this context, the guidelines on handling data in day-to-day business published by the SBA shed light on the legal and regulatory requirements that must be observed when deploying AI solutions. The conception of AI projects and in particular the data used can give rise to additional risks such as bias, uncontrollable/inexplicable results (“black box” behaviour), a lack of robustness when exposed to new data and cyber security risks. These must be identified, assessed, audited and embedded organisationally within data and AI governance frameworks that are coordinated with each other before a bank rolls out AI solutions on a large scale. As ever, therefore, a balance must be struck between innovation and risk tolerance when deploying these technologies.

Banks can master these challenges by making the most of their know-how and capabilities with regard to identifying customer preferences, adding value through data (internally and within ecosystems), creating a modern data architecture (reference architecture) and collaborating. Good governance models can help here by ensuring high quality, legally binding rules and organisational integration.

From lab to factory – start small and scale up quickly

Uncertainties regarding the feasibility of and approach to planned AI solutions can be minimised by setting up AI labs. These offer a safe, focused environment for developing selected bleeding-edge use cases that are new, innovative and a high priority for the institution or the sector through research and experimental projects and for building initial prototypes. AI labs are all about conducting empirical, cross-functional and case-specific research to reduce the unknown quantities associated with new projects and technologies.

They also offer the opportunity to take a closer look at the scalability and industrialisation potential of specific use cases and the related technologies, thus providing initial quality assurance for mass production. Industrialising use cases tested in the lab and scaling them up to a bank’s day-to-day business in the real world require the ability to scale limited resources and work in an agile manner. This is the only way to guarantee that the desired efficiency gains can actually feed through into the business. An industrialised, agile model makes it possible to realise AI solutions more comprehensively and more naturally. An agile, scalable collaboration model calls on interdisciplinary feature teams working within a DevOps framework with a local or global delivery concept to focus on transforming the existing business on the one hand and continually refining AI solutions and ensuring that they operate stably on the other.

Three prerequisites for deploying AI successfully

AI will change our society, our economy and also our banks forever. Banks have the ability and the will to harness the huge potential of AI, for example in terms of speed and efficiency, to offer faster processes as well as products and services that are more precisely geared to their customers’ needs. At the same time, AI presents them with technological, organisational and regulatory challenges. They should address these challenges with a mixture of good governance, novel methods and new collaboration models to deploy AI successfully now and in the future.

About the author

Christoph Schrills is Lead Consultant Financial Services & Digitization at NTT DATA in Zurich. He specialises in data, analytics and artificial intelligence.

Blog series on handling dataDigital Finance & Cybersecurity