Paving the way to Dynamic Customer Interaction for Digital Payment Platform

A Fintech company with billions of transactions records aims to begin their AI journey and to translate the wealth of data that they have to actionable strategies. The project was apportioned into 3 distinct phases: i) Discovery, ii) Solutions Creation, iii) Handover

Following our AIDesign Thinking Methodology, ThinkCol communicated with different departments to identify the best use cases for the company. We brainstormed more than 200 ideas with the clients on potential AI ideas and thereafter pinpoint with the client on two specific use case that were crucial towards the business based on feasibility and business value. Such direction was approved by management with the support of business users. A presentation to management was also done to align the vision.

ThinkCol then constructed meaningful customer segmentation strategies. User attributes were engineered from billions of transactional data and unsupervised learning clustering models were being created. Azure Databricks and Spark was used as billions of transactions were analysed in a distributed way.

As there were some gaps in the data, data enrichment was performed through analysing CRM data and creating new features, combining previous data with scraped third-party data as well as imputation. With this database of inferred data and clusters, propensity models were developed through various techniques and algorithms to achieve a 40%+ increase in action compared to control during A/B testing.

As we were able to prove the predictive power of our models, model explainability was then performed using Shapley values (SHAP) to understand from the machine learning models what drives each different customer to perform certain actions. Such attributes were then integrated into their CRM system to automate the correct ads to be given to each individual customer. Hence based on the model prediction of what is more likely to drive person A’s action, a specific ad for person A will be automatically sent to him via the CRM system. The entire process was deployed with Azure DevOps and Azure ML to ensure a seamless process.

ThinkCol through co-building with the client on the two use cases accomplished the following objectives:

• To perform propensity modelling on each individual customer through analysing billions of transaction data
• To improve data quality and integrate transactional data with external data, and
• To integrate and automate CRM with our AI models.

After project implementation, empowerment training was conducted to communicate technical details, ensuring that the firm’s internal staff was capable of using ThinkCol’s models efficiently without explicit guidance. ThinkCol enabled the company to become more resilient and self-sustaining, and to maintain its appeal amidst the plethora of available alternatives in the FinTech sector.