Multimodal risk assessment for financial services - overview
AI compliance assurance for financial services
The sale of financial products such as loans, insurance and investment products are typically sold on a commission basis. In Australia, many of these products are sold by Financial Advisors who have an obligation to act in the best interest of their clients.
To mitigate any risk of advisors putting commissions ahead of a client’s best interest, financial service providers have been running compliance review programs to identify advisors who need to change their practices. However, these programs are typically retrospective only and cannot scale.
Working together with Promontory, who have expertise in Financial Advice compliance, we are building a tool that analyses “Statements of Advice”. Here we assess whether advisors have met the client’s goals and objectives. This assessment is built on top of business rules and a number of deep learning and machine learning models for:
- Understanding tabular structure within documents
- Identifying goals; recommendations and associated product domains (insurance; superannuation)
- Understanding whether a recommendation supports a goal.
Building these models will allow Financial Service providers to do real time risk assessment of statements of advice and therefore protect their clients.
Also, at IBM Research – Australia we are integrating advances in the psychology of decision-making with cutting edge tools in big data analytics and artificial intelligence with the goal of helping correct confirmation bias in financial decision-making.
Recent advances in psychology and cognitive neuroscience see human decision-making during uncertainty as a process of information accumulation where a decision is made after reaching an information threshold. Built-in processes in the brain, used to assess information and turn it into evidence, can be subject to a number of distortions known as cognitive biases. For instance, “confirmation bias” is the unconscious tendency to give more weight to evidence that supports an existing belief, at the expense of contradicting information.
The main innovation consists of a tool that collates, classifies, and clusters alternative sources of financial data (internet web traffic, Twitter, RSS feeds, weather, satellite data) and contrasts that information with conventional decision-making tools (financial reports, analysis based on fundamentals).