The growth of Data Sciences in financial services

By Philippe Meyer, head of innovation at Avaloq

In today’s highly competitive, fast-growing digital economy, banks and wealth managers need seamless access to innovative solutions that will keep them at the forefront of the digital curve. This is even more necessary in an era of Open Banking with its heightened need for transparency and its new breed of regulations such as the Second Payment Services Directive and GDPR.

As such, the transition of Data Sciences (encompassing predictive analytics, machine learning and artificial intelligence) from academic niche to mainstream business discipline is now at the top of the banking agenda. Fuelled by the sheer quantity of data being generated, the ability to manage and utilise data assets has become key to a bank’s success. Data scientists are bringing the potential of machine learning and artificial intelligence alive, presenting numerous opportunities and challenges.

There is a clear need for horizontal platforms to overcome some of the hurdles created by fragmented implementations, to ease deployment and fulfil the expectations of banks and regulators about their usage. In addition, the maintenance of machine learning processes is complex, and may prove more expensive than deploying them. Models need constant fine-tuning and training to give consistent results over time.

The underlying data challenge

Data consistency, lineage and governance are key to the growth of Data Science. However, many banks struggle with being consistent across existing reporting structures due to multiple systems and silos, and duplication of data accountability. Other data activities such as the infrastructure build, data collection and data verification are likely to consume significant time, effort and resources, limiting the time banks have to focus on their customers.

An additional challenge is the speed at which data is calculated. Some data science solutions, particularly in fraud detection or high-velocity trading, have been engineered to deal with real-time streaming data but this is the not the norm across the industry. In fact, the power of machine learning in banking can only be truly realised when enterprise wide and event-driven solutions come into play.

Towards data science platforms – as a Service

Many of the current industry concerns are simply a reflection of the relative immaturity of the current machine learning market. There is not yet an agreed upon single consistent and layered approach, and there is no consolidation around standards yet. There is also a mix of different players in the market, ranging from some big names in the platform market (e.g. IBM) to a number of smaller-scale businesses, all with different client demands to meet.

But what would such Data Science platforms need to do in order to be useful in finance? Typically, such platforms need to define a clear financial data model with consistent definitions, and cover off the issues of data governance and lineage.

There needs to be a common statistical package, with standard algorithms, which can be used to build dedicated analytics, as many of the required base mathematical functions and statistical processes are similar. Machine learning companies and banks must also come to understanding over what they want to achieve. Although machine learning companies want to differentiate their individual user experiences, internal banking process owners often value consistency and clarity of data representation the most.

An industry on the verge

Avaloq believes that the industry is on the verge of change and real Data Science developments. The power of machine learning, while welcomed for its benefits, still leads to some overriding concerns for the operators and managers of banks. These will need to be overcome for it to reach its full potential.

For example, the maintenance of machine learning processes is complex, and may prove more expensive than deploying them. Models also need to be updated constantly and training of the relevant staff is also essential to achieving consistent results over time. Expert data science resources will inevitably be required beyond the bounds of the initial implementation project.

Because the interpretability of machine learning response models is critical in a highly regulated environment but also difficult to navigate, it is a topic of research and concern. Interpretability of machine learning response models comes into play when the size of the data pool is not great enough to provide a reliable or optimal model. Interoperability is the idea that a Machine Learning model that has been trained on one limited data set can be combined with a model that has been trained on another limited data set.

Assuming these data sets are from the same domain, e.g. payments data, then the combined model will benefit from the aggregated knowledge derived from the two data sets. This is particularly beneficial for small and medium-sized banks who cannot combine their data with competitors but still need to benefit from Machine Learning and a large data lake.

Conclusion

With Open Banking and its accompanying regulations changing the face of the industry, financial institutions increasingly need access to the best fintech solutions and firms that are building the next generation of applications and software. This will help them thrive in this new era, one in which clients want to interact with their banks and wealth managers in wholly digital ways. Although Data Science has presented a number of innovative solutions to meet such requirements, the industry still has a long way to go but the distance to the end is decreasing rapidly.

In overcoming these challenges, the industry will be able to give customers unparalleled insights and opportunities to compete in this new digital landscape.