Taishin Bank International uses artificial intelligence (AI) to power its new 'Personalized Wealth Management Recommendation System' (PWMRS), which can rapidly calculate suggestions of optimal asset allocation to individual Taiwanese customers. It uses factors such as past habits, risk preferences – which it can further enhance – and so on to improve back office efficiency and automated delivery of data to drive recommendations. It won Outstanding Wealth Management Technology Initiative – Back Office at the virtual PBI Global Wealth Summit & Awards 2020.
The improvement in back office data mastery means the granularity of data has advanced, positively impacting services on the Taiwanese customer portal. The front office productivity of advisors has also improved, as they now have more time to knowledgably talk to customers and advise them expertly. The implicit cost of advisors spending large amounts of time searching for information has been drastically reduced.
The PWMRS AI engine also generates a better list of potential customers, which aren’t on the portal, for the sales team to approach, at the most appropriate time in order to grow the business. Equally, precision targeting can eliminate the annoyance of receiving large amounts of unsolicited marketing.
When a customer, using Taishin’s portal, makes a financial planning request the system immediately analyses the customer's data history, examining the monthly asset allocation situation; payment behaviours; occupation; and so on. PWMRS then calculates the client's risk preference and provides the optimal asset allocation recommendations to the advisor, who doesn’t have to question the client so deeply, to pass on and initiate whatever action is required.
It uses an Artificial Neural Network (ANN) algorithm to handle cross-sectional and time-series data. Aligned with a Long Short Term Memory (LSTM) approach, the system can cope with different dimensional data, and optimise services.
Traditionally, LSTM cannot accurately reflect the characteristics of the future state of a customer's decision at a given moment. That is to say, a customer's securities asset balance in September may be affected by the estimated amount of investment in October. Predictability is enhanced by combining it with the ANN. Indeed, Taishin introduced bi-directional LSTM, so that the characteristics of all data can be better reflected, and the accuracy of the overall financial model improved.
Taishin continues to try to strengthen the model's predictive ability to improve its recommendations. For instance, after comparing a large number of models, it found the predictive ability (AUC) after the simultaneous calculation of the extreme gradient boost (XGBoost) is higher than 0.9. This indicates that this model has an outstanding ability of discrimination, says the bank.
The AI-powered PWMRS was officially launched in Q3 2019, after continuous testing and enhancements such as that described above. It offers the following key benefits:
- Advisor productivity rises as the ease and efficiency of data retrieval is improved.
- Less time collecting data = more time communicating = more sales.
- By using ANN, the bank only needs one model to integrate data. This negates the need for amalgamating multiple scenarios into one coherent picture, saving cost and maintenance activities.
- Since its Q3 2019 introduction, the transaction rate of each product has increased. For example, the transaction rate of funds has increased 2.5 times, and traditional insurance has risen 3 times.
- Better risk identification and classification for clients has ultimately been possible, especially for the majority digital customer base at Taishin, who already have rich customer profiles with the bank. These can now more easily be ‘mined’ for useful, actionable data.
Future plans include enhancing the predictability of the AI-driven system as more data and variables are fed in. Specific examples in the pipeline include:
- Using customer feedback for future voice recognition services.
- Using emotional indicators to improve analysis.
- Text exploration technology will be deployed to process a large number of corporate announcement documents, financial reports, news, community articles, and stock data. AI will mine this to optimise the risk profiles of investment portfolios.
AI can automate financial theories and compute traditional asset and portfolio risk allocation mechanisms, such as Markowitz Theory. But the usefulness of this dissipates in the ‘noisier’ financial markets of today where it’s often harder to take just one, two, or three factors into account to drive a predictive model.
Multiple factors can move volatile markets these days, so accurately predicting expected returns and advising customers about optimal capital allocation, needs traditional theories to be refined and new algorithms to be devised. Taishin has the human and technological capabilities to do this. Its activities in this area will be supported by using more AI models in the future, such as the Multi-armed Bandit algorithm. It will also use AI to further enhance customer experience (CX) at the bank.
Taishin Bank International also won in the • Outstanding Private Bank: North Asia category at the 30th annual Private Banker International (PBI): Global Wealth Summit & Awards 2020 and achieved Highly Commended (HC) status in the Outstanding Global Private Bank: Global category.