Banking

In the world of credit, he who has the best models wins. Banks are using machine learning to build better models for estimating default probability and loss severity, and for loss forecasting. They are using these models to improve pricing for risk, credit approval, and portfolio management. Building more granular models with automated machine learning makes credit scoring more precise as models learn the nuances of discrete populations.

Fighting financial crime, especially money laundering and fraud, is more important than ever and is getting more challenging as criminals get ever more sophisticated. Using machine learning, banks are learning from their investigational findings and fraud losses and training models to accurately detect suspicious activity and to spot and prevent fraud in real time. And these models continue to get better as they learn over time.

Clients expect banks to know who they are, what they need, and when they need it. Drawing from data on clients in similar situations, banks are using machine learning to predict client needs. Some banks are identifying event triggers which may indicate that a new need has arisen. Banks are learning from client complaints where their attrition risk is highest so they can take action, and they are building predictors of traffic volume (in branch, in contact centers) so that they can staff optimally.

Banks are using machine learning to predict which prospects are likely to become the most profitable clients and are using this ability to prioritize leads and referrals. Banks are learning from their clients to target their offers more precisely, an imperative with digital advertising. And they are creating sophisticated analytics to predict client price sensitivity, tailor their value proposition, and estimate price-volume elasticity.

To improve cash management, banks are using machine learning to predict new loan demand, to predict prepayment speed, and to forecast ATM cash requirements. Banks are using historical data on cash inflows and outflows to build models to predict cash flows. This allows them to have the right amount of cash on hand where and when they need it and to optimize the return on excess cash.

In financial markets, traders are using historical transaction cost analysis (TCA) and execution data to build models that optimize order routing and trade execution strategy. These models evaluate the relative merits of the numerous potential algorithmic trading approaches, venues and counterparties. These support trader decision making and help to minimize market impact and cost while demonstrating and recording the trader’s compliance with best execution requirements.