Machine learning adoption continues at a rapid pace, with AI technology being used to the benefit of sell-side efficiency and the buy-side's #SearchForAlpha. To maintain this progress, how can financial services organizations overcome challenges that still exist around data quality and availability?
- Machine learning adoption is now widespread in financial services, with the buy-side mining new data sources, finding patterns and making predictions in the #SearchForAlpha.
- ‘Off the shelf’ solutions and common coding languages, like scikit-learn in Python, are making it possible to rapidly create and test new machine learning models.
- Poor data quality and availability are the biggest machine learning challenges. No matter how well built the model, its output will only be as accurate as the data that feeds it.
The rapid pace of machine learning adoption means that AI technology is now considered an almost everyday part of life for large organizations in financial services.
Our recent survey of senior personnel in global finance, including c-suite executives and data scientists, highlighted this progress when 90 percent of respondents said they had deployed machine learning in one or more departments. Three-quarters believe it now represents a core part of their business strategy.
Machine learning adoption
The first factor driving machine learning adoption is access to high-quality, robust and adaptable machine learning models.
Until recently, data scientists needed to research and hand-build them from scratch. Now ‘off the shelf’ solutions implemented in common open-source frameworks (e.g. TensorFlow, PyTorch, Scikit-learn) are making it possible to rapidly create and test new models.
Secondly, as the amount of data available has grown exponentially, companies have invested heavily in the infrastructure they need to manage it. This infrastructure — with high performance computing power including GPUs, a wide variety of database choices and low-cost cloud storage — provides fertile ground for the growth of effective machine learning applications.
But without market demand and practical uses, any technology will falter. The final factor therefore is the intense competitive pressure that organizations have faced since the financial crisis.
Sell-side vs buy-side
On the sell-side, the need for scale and efficiency is driving machine learning adoption right across organizations, from the trading desk to the back office. It is typically being applied to improve efficiencies and differentiate by, for example, better risk management.
The buy-side is under pressure to create an edge that sets it apart. In a world where data is more commonly available than ever before, firms need to find new data sets or fresh ways of interrogating or combining existing ones. Machine learning can help to mine new data sources, find patterns and make predictions in the search for alpha. This necessity to innovate explains why our survey puts the buy-side ahead of the sell-side in terms of machine learning deployment.
The sell-side, competing on scale across a portfolio of businesses, have often focused on ensuring their core functions, such as customer service, asset serving and risk management, are as efficient as possible to capitalize on their economies of scale. The buy-side, with a tighter focus on managing and growing institutional assets under management, has to offer an edge in investment performance to compete.
Will anything stop machine learning?
Despite its success, there are still obstacles ahead for machine learning.
Gaining useable insight from machine learning not only takes time, resources and talent — it also requires both technical and business understanding.
Machine learning algorithms detect patterns in data, and from these make predictions and recommendations. While this probability-based approach is highly effective in certain areas such as routing customer calls, it may not be appropriate in others, such as trading decisions.
These nuances may explain the divergence between c-suite and data scientist perspectives in our survey.
While 100 percent of c-suite respondents say they have deployed machine learning and that it is core to their business, 56 percent of data scientists do not see machine learning as being core and 12 percent say they are still experimenting with it.
So, while data teams are more aware of the practicalities, executives are focusing more on the potential.
There are also significant regional disparities, with 95 percent of U.S. organizations making a significant investment in machine learning versus 72 percent in APAC and 64 percent in Europe.
The U.S. could have the edge because of its single domestic market and language, plus a primary focus on one asset class — equities.
As a result, U.S. machine learning operations benefit from less complex data collection and analysis processes, while also enjoying access to high quality research from the region’s top tier educational, technology, and financial institutions.
In contrast, Europe and Asia have more languages and variety of mature and emerging market structures to manage adding friction to machine learning deployment.
Meeting the data challenge
No matter how well built the model, its output will only be as accurate as the data that feeds it.
Our survey identified poor data quality and availability as the key challenges to machine learning adoption. This is where Refinitiv can help.
As one of the world’s largest providers of financial markets data and infrastructure, we provide the depth and breadth of data required for machine learning deployment.
Not only that, we also provide it as standardized, and easier to find and clean, because it’s been through a comprehensive data-validation process.
As machine learning continues to evolve, Refinitiv can help you to take full advantage.