Guest Blog: Accelerating Adoption of Machine Learning in the Financial Services Industry

Machine Learning (ML) is such a powerful new set of technologies, opening up so many business possibilities, that it’s sometimes difficult to imagine how best to put it to use. The possibilities are overwhelming, and there is little history to build on. In this post Baran Karlidag ties ML to the specific challenges that financial services companies are facing in today’s economy, and provides a number of concrete ideas that FinServ firms should be considering. He also addresses issues like governance and regulations that often constrain FinServ firms.

—Mark

Guest post by Baran Karlidag, Global Financial Services Partner Solutions Architect, AWS

Baran Karlidag is a Partner Solutions Architect focusing on Financial Services at Amazon Web Services. Baran works with AWS Partners to help Financial Services customers in their cloud adoption journey and build innovative solutions. Prior to joining AWS, Baran held senior technology and leadership roles in HSBC, Royal Bank of Scotland, Abn Amro and Reuters. Baran has a MSc in Electrical and Electronics Engineering.

Adoption of machine learning (ML) in financial services companies is accelerating and changing the way they operate and serve their customers. Sixty-seven percent of UK financial services companies that responded to a recent Bank of England (BoE) survey¹ indicated that they use machine learning in some form across a range of business areas. The challenge facing financial services organisations today is, How can they take these pockets of machine learning and deploy more widely to have a meaningful impact on their businesses, improving their competitive response to agile FinTech disruptors entering their markets?

When making the case for a more widespread program of machine learning across their organisation, CIOs and CTOs are citing the following drivers.

Market-changing disruption by nimble FinTech companies: The market share of traditional financial services companies is coming under sustained attack from agile FinTech firms. A survey of 7,900 retail banking customers in 20 countries revealed that 60 percent of respondents use products from big technology companies/challenger banks or are likely to do so within next three years.² Traditional experiences are no longer enough for customers who want instant and personalised interactions through multiple channels. Gen Y respondents to the survey rated websites and mobile applications as more important than physical branches.The environment that traditional banks operate in is changing rapidly, and they need to move quickly or risk becoming irrelevant.
New legislation driving structural change in traditional markets: Legislation and standards like PSD2 and Open Banking in the UK enable sharing financial information electronically with third parties securely and only under conditions that customers approve of. In effect, the banks’ monopoly over their customers’ data is removed so it’s easier for customers to choose better products. For example, Canopy uses Open Banking to access an individual’s bank account information to prove he can afford his rent without having to provide bank statements or payslips.If banks don’t act quickly, they risk being reduced to the role of utility providers.
Rapidly evolving cyber threats and fraud attempts: Financial organisations of all shapes are facing increasingly sophisticated cyber threats and advanced fraud attempts. Traditional approaches to cyber security often fall short, struggling to act with the speed of new threats and leveraging the ever-increasing volume of data to mount a defence.

Machine Learning Can Help Address the Disruption

The CIO, whether in banking, capital markets, or insurance, has to balance a complex set of goals in today’s disruptive environment. She needs to drive innovation and accelerate the digital transformation agenda while maintaining the existing infrastructure, ensuring security, and keeping the costs down.

Financial services companies can benefit from ML technologies for various use cases to address the challenges described above. Wherever there is complexity and high volumes of data, ML may provide a more powerful solution over the traditional rules-based models. ML models make use of more data points and can be built and adapt more quickly, providing better detection with lower false positive rates.

Financial services companies own massive amounts of data that has accumulated over decades. Today this data can be complemented with new data sources like geospatial data, social media, and other online activities. ML technologies make it possible to analyse very large data sets and identify relationships which would otherwise remain undiscovered. This provides a rich source of insight that can power better personalisation, tailored products, and services—which will help traditional banks retain their clients.

However, financial services companies face a number of challenges in scaling the adoption of ML:

Enterprise governance: Align machine learning application development with existing corporate processes and ensure regulatory compliance
Innovation: Ensure the organization can innovate quickly and leverage new technologies in the machine learning domain
The aftershock of “going live”: Build the organizational capability to support and evolve machine learning systems in production

Enterprise Governance

Financial services is a highly regulated industry with strict security requirements. This means that access to data needs to be carefully controlled and monitored. Teams cannot just download data from production systems to use in ML projects due to its sensitive nature.

Financial services companies need to create the necessary governance, processes, and capabilities for ML to efficiently use and create value from data. Equally, an ML solution needs to be developed in a way which is aligned with the organisation’s IT and security processes.

Software development processes have been stable over decades in financial services: the artefacts they generate and the associated controls are well established. But these processes do not entirely cover the end-to-end ML development lifecycle. A machine learning project has very different inputs, collaboration requirements, and artefacts than a traditional software development project. ML development is a complex, iterative process which becomes harder because there are no integrated tools for the entire machine learning workflow. Organisations need to modify existing processes and tooling to include ML. This makes processes more complex and requires additional training for the teams involved.

Regulators require financial services organisations to have a validation process for their internal financial models. Financial services organizations will need to produce artefacts to demonstrate their ML systems have been designed, trained, and updated in line with their design objectives to yield fair and unbiased results. McKinsey’s model validation arm has added 6 new risk elements and modified 12 existing elements in their model validation framework to address risks specific to machine learning.³

Today, 57 percent of financial services companies polled by BoE say that their applications are governed through their existing risk management framework or enterprise risk function.¹ Several firms highlight the need for their risk management frameworks to evolve given their increasing use of ML.

Innovation

ML technologies offer many innovation opportunities, but it is often challenging for enterprise organizations to identify the best technology for a specific use case. A couple of factors slow down experimentation and adoption of new technologies.

The ML landscape is fast-paced with rapidly evolving frameworks and models. The choice among popular frameworks is not straightforward; some frameworks provide better results for some use cases. Data scientists in the organisation will benefit from experimenting with different frameworks, but this can be quite difficult to achieve in a complex financial services IT organisation. Data science teams will need to navigate different parts of the technology organisation to experiment with new frameworks, and these new technologies will need to be validated against the organisation’s security standards.

Once validated, the IT team will need to acquire the skills to install and maintain the new framework. Machine learning requires a lot of computational power. Most likely, the data science teams will not have access to infrastructure resources to quickly test and validate if the new framework will prove valuable to their use case. Budget constraints and the procurement lead times often reduce organizations’ capability to experiment and hence innovate quickly. It is not feasible to own idle infrastructure just for experimentation, especially when this infrastructure needs to be at the scale to address the computation requirements of modern ML technologies.

Going live…

Deploying and maintaining ML effectively is challenging in a financial services company, and this brings another level of complexity to the IT organisation. The ML model needs to go through change and release management processes and be integrated with the existing tooling. Once deployed, systems need to be monitored, sized for load, and patched. Infrastructure teams need to acquire new skills to effectively perform these tasks.

ML applications in production need to be monitored to make sure they are performing as intended. Organizations need to make sure that the data used for building the application is still relevant. If the ML model has drifted, it has to be evolved and deployed again. This must be done in a manner that allows the organization to respond quickly to changing circumstances without bringing extra overhead to infrastructure teams.

Luckily, there are methods available that can help you manage these interlocking layers of complexity, including one from Amazon Web Services.

Amazon SageMaker Can Help Overcome These Challenges…

The Amazon SageMaker service covers the entire ML lifecycle and can help companies overcome the challenges described above.

SageMaker solves these challenges by providing all of the components used for machine learning in a single toolset so models get to production faster with much less effort and at lower cost. The tools cover the entire ML lifecycle, from initial exploration to managing the production estate. They are seamlessly integrated with other AWS products and security features. Organizations can use this toolset to align their ML development processes with complex internal governance and satisfy most of their security requirements.

CIOs need to transform their organisations to address disruption in financial services, and they need to integrate modern ML technologies in their processes. Enabling experimentation with new technologies and speeding up the development lifecycle of ML projects will help institutions to effectively gain insights and get a competitive advantage. Amazon SageMaker is a very powerful enabler for financial services companies to realise these benefits and adopt machine learning at scale.

More on this topic
The Cloud Era of Financial Services
AWS for Financial Services
AWS Financial Services Webinar Series
AWS Machine Learning for Financial Services
AWS Well-Architected Framework, Financial Services: Artificial Intelligence and Machine Learning

Footnotes
1. Bank of England, “Machine learning in UK Financial Services”
2. Capgemini, “World Retail Banking Report 2019”
3. McKinsey, “Derisking machine learning and artificial intelligence”

Author

AWS Cloud Enterprise Strategy Blog