Artificial intelligence (AI) is rapidly transforming the operations of financial institutions, making them more efficient and competitive.
AI can help to accelerate revenue growth, reduce risk and cut costs, delivering tremendous value to financial firms. Its applications in finance are extensive, Yeo Hwee Theng, data scientist at DataRobot, said in a webinar co-hosted with FinanceAsia in June.
In a fully AI-driven organisation, AI does the heavy-lifting, allowing financial institutions to focus on their essential work to better serve clients, Nina Xing, another data scientist at DataRobot, said in the same webinar.
In the case of anti-money laundry (AML) practices, AI can help financial institutions save a substantial amount of capital. The estimated return on investment can range between $2 million and $5 million, according to studies by DataRobot.
AML is important for reducing regulatory risks and increasing brand awareness. However, the rule-based models deployed by most banks now are not the most ideal AML solutions. Machine learning models can be introduced to replace or supplement existing rule-based models.
Companies can build rule-based models across the transaction monitoring process in their AML practices. The models gather customer background information and verify if they are legitimate ones when they onboard customers, Yeo said.
Banks then monitor the transactions. Some of them will be flagged as suspicious by the rule-based models after alerts are set up.
“But the problem with rule-based models is that they are rigid and slow at reacting to customer behaviour changes. And as a result, many of the alerts are actually false alarms. And there will be no suspicious activity reports filed after the investigation,” she said.
Only around 40% of the alerts involve complex situations. Studies have also shown that false rates can be over 90% with rule-based models and huge resources are wasted during the manual investigation process, she said.
“[In comparison], machine learning models can filter out the false alarms and reduce the number of cases that requires manual review. Thus, they increase the operational efficiency and reduce costs,” she said.
When it comes to ensuring that the models are always up to date and relevant, Yeo added that AI models can be retrained periodically with the latest transaction data to capture changes in market condition and customer behaviour. The retraining process, which is fairly straightforward, can be scheduled to run regularly.
AML only demonstrates one of the countless applications of AI in finance. On the buy-side of the business, AI can be leveraged in areas such as asset allocation, factor model build-up, and smart beta strategy discovery, Yeo said.
Sentiment analysis is another good example to showcase how natural language processing, a part of the machine learning techniques, helps to drive business. It can help capture signals that traditional quant techniques find it hard to do, she said.
THE CHALLENGES
It is important for financial institutions to embark on the AI journey as soon as possible, which starts when they establish awareness and acceptance of AI across the organisation.
As they move around the AI maturity curve, they will develop more use cases, driving efficiency and standardisations across the end-to-end pipeline while accelerating their data science capabilities.
When many use cases are generated, organisations will start to “democratise” data scientists, allowing staff like business analysts to work on AI initiatives. These are the key milestones in their journey to becoming fully AI-driven, Xing said.
However, automation is not easy to achieve, especially because people need to trust the model that they built in many cases.
“People may be skeptical about AI and think that they give black-box solutions. Others have concern about ensuring the models stay relevant to the ever-changing business environments and are compliant with global regulators,” she said.
The technical details present other challenges. Financial institutions need to know the model accuracy, compare performances between models, and understand the trade-off between different aspects such as speed versus accuracy.
They should also be able to explain the model structure and understand the kind of data needed in the pre-processing stage. Measures have to be put in place to prevent a drift in data.
“All these are important questions to validate that your AI is not a black box. And we are not relying on unexplained models to make critical business decisions,” she said.
They are difficult yet important steps to take to reap the ultimate benefits of AI and stay ahead of the curve in the rapidly changing business environment.
Find out more by watching the on-demand webinar here.
