Causal AI in Retail Banking

Objectives
To comply — as speedily and as efficiently as possible — with existing and evolving regulations for AI model risk, taking into account the increased emphasis on explainability, stability, fairness, bias, and trust.

Current Approach and Pain Points
Control and process framework involving manual reviews and expensive iteration process between data science team, business units, and model risk team. High reliance on SHAP and/or Lime for explaining outcomes or limited to linear models only. Ill-prepared to adhere to the latest AI regulations:  the new EU AI proposal or evolving US and Singapore regulatory guidance.

Causal AI Solution

• Causal MRM decision app — risk assess and evaluate existing models, testing features, causality, explainability, bias, fairness, sensitivity, and robustness to change
• A user interface that accelerates model review and acceptance by allowing business, DS, and model review teams to collaborate and test iterations in one session
• Accelerate model deployment and validation by 6X

Objectives
To create market-leading products that attract new customers, help retain existing customers and generate new revenue streams and additional profit

Current Approach and Pain Points
Decision-making process based on limited or variable data and heuristics; unreliable forward planning.

Causal AI Solution

• Optimized pricing — which also takes account of market conditions including competitive pressures — maximizes ROI and PnL
• Stress test launch product
• Envisage future scenarios to inform forward product planning

Objectives
To use forecasting to: i) shape and tailor products ii) maximize customer satisfaction and value iii) enhance liquidity management and iv) more accurately assess and manage client attrition costs.

Current Approach and Pain Points

Although they can provide limited projections, spreadsheet models are fragile and as such, unreliable: they cannot factor in unpredictable, non-data related developments or recommend a course of action.

Causal AI Solution

• Cause and effect time series forecasts supported with recommendations on how to optimize customer management and balances
• Tailored customer alerts help Relationship Managers optimize customer value and PnL

Objectives

To issue timely updates based on projected price moves and/or market sentiment which may have a positive or negative impact on their investors’ portfolios.

Current Approach and Pain Points

General overviews, impersonal, lacking relevancy.

Causal AI Solution

• Highly personalized and directly relevant updates based on dynamic market movement models
• Tailored updates increase PnL by improving customer engagement, building loyalty and encouraging additional product uptake

Objectives

To refine strategies and spend by understanding which channels and content are most effective in terms of customer acquisition, sales and ROI.

Current Approach and Pain Points

Inaccurate and purely correlation-based models, spreadsheets, business intelligence and no mapping of customer journey to value.

Causal AI Solution

• Assign specific values to individual channels and content, simulate future prospect and conversion scenarios and optimize marketing budget for maximum ROI

Objectives

To develop a credit risk model that maximizes PnL by promoting fair and unbiased lending, helps vulnerable customers avoid problems and is capable of responding to changing economic environments.

Current Approach and Pain Points

Simple linear models founded on limited behavioral and alternative data are unreliable in volatile market conditions. Current approaches create issues relating to bias and fairness.

Causal AI Solution

• Models that can justify and explain decisions and incorporate real-life continuous risk assessment processes increase PnL and robustness to shocks
• Bias and fairness sensitivity and counterfactuals built in
• Feed off causaLake to identify external signals that influence credit risk over time

Objectives

To avoid defaults by optimizing the in-life customer management process. As and when defaults occur, to ensure debts are managed so as to minimize losses while protecting vulnerable customers.

Current Approach and Pain Points

Escalation procedures, which are not tailored to the borrower, often involve expensive collection agencies, financial loss and dissatisfied customers. Regulatory constraints are difficult to model.

Causal AI Solution

• Creates a customer-centric escalation and a default risk and collections strategy
• Optimizes response based on the most effective channels and messaging and strikes a balance between risk and value while complying with regulations
• Reduced costs and improved PnL

Objectives

To mitigate the costs and the regulatory and reputational risks caused by online fraud, scams, money laundering and market abuse.

Current Approach and Pain Points

The rules-based approach, coupled with limited machine learning models spread across multiple point solutions and enterprise systems, make it difficult to keep up with evolving fraud attacks and modus operandi. Modifications can take up to nine months to complete and new attack vectors in social engineering scams and associated regulations, may cause acute problems.

Causal AI Solution

• Holistically models the attack vectors from domain knowledge to data
• Provides globally explainable models with simulation and intervention analysis for operational scenario modeling and strategy optimization
• A model that encompasses sub models, point solutions, raw data and domain knowledge in an over-arching, decision-making system to nullify different types of attacks
• Increases agility by 10x and reduces fraud by a further 70%

Objectives

To improve liquidity, asset and liability management by establishing the correct balance between risk, regulations and profit opportunities.

Current Approach and Pain Points

Spreadsheets and heuristic models based on risk appetite for liquidity coverage ratio and net stable funding ratio, portfolio level aggregations and simple intraday liquidity management. Cash may not be utilised as efficiently as it could be and risks may be heightened.

Causal AI Solution

• Builds liquidity, asset and liability forecasting and alerting models based on internal and external macro-economic data
• Develops finely detailed models from customer cohort to portfolio level
• Provides accurate forecasting of operational balances, price sensitivity and volatility for risk management and optimizes and extends asset and liability management
• Enhances PnL

Objectives
To ensure compliance with current and future regulatory requirements utilising accurate scenario planning, stress testing and reverse stress testing.

Current Approach and Pain Points

Linear models that draw on small data sets, offer limited and/or inflexible scenarios and involve a lengthy implementation process.

Causal AI Solution

• Accurate and explainable models that enable domain input from the start
• Identifies and models non-linear causal relationships
• Accommodates scenarios of any and all types and reverse stress testing
• Model development to implementation 30x faster

Objectives
To comprehend the relationship between budgets, risk mitigation, results, compliance and PnL and utilise funds to maximum advantage.

Current Approach and Pain Points

Spreadsheet models and heuristics underpinned with limited statistical forecasting.

Causal AI Solution

• Models that see the business as a dynamic system of moving and interacting parts and take account of the external environment
• Provides more accurate forecasts, scenario planning and optimization of budget allocation strategies
• Heightens visibility and understanding of budgets and ROI
• Substantiates intervention and strategy decisions, increases PnL and minimizing risk

Objectives
To improve business performance and efficiency by understanding how allocations, constraints and supply and demand have on resource management.

Current Approach and Pain Points
ERP systems and reporting using rules-based spreadsheet models and some statistical forecasting.

Causal AI Solution

• Models that see the business as a dynamic system of moving and interacting parts and take account of the external environment
• Provides more accurate forecasts, scenario planning and optimization of budget allocation strategies
• Heightens visibility and understanding of budgets and ROI
• Substantiates intervention and strategy decisions, increases PnL and minimizing risk