Our client is a leading provider of Individual & Group Retirement and Life Insurance plans in the US. The client had lost sizeable Assets Under Management (AUM) from receding contributions and closure of 403(b) plans by its customers.
There existed an opportunity to enable Financial Advisors and Customer Care Agents retain customers through predictive insights and proactive retention strategies. Specifically, the client wanted to:
— Identify at-risk customers to enable Financial Advisors engage with customers ahead of time
— Empower Customer Care Agents with reasons for attrition and talking points that help retain customers that are already in the process of attrition
- Identifying relevant data and evaluating it for utility – analyzed over 15 data sources and aligned with business teams for effective use in the analysis
- Inconsistent data with limited information and history, partial data dictionary, and limited business context for most data elements
- For Advisor persona, we built a suite of predictive (machine learning) models to identify
- Customers that will likely make withdrawals
- Customers that will likely stop contributing
- Likely path of withdrawal – third party or internal movement, full or partial withdrawal
- For Customer Care Agent persona, we built an AI and NLP-based solution to identify customers who have a high probability to be retained. The solution provides talking points that will increase the chances of retention. The intent is to change their mind to withdraw a submitted account closure request.
- Built over 50 hypotheses related to behaviors that need to be predicted and addressed. Analyzed over 15 data sources capturing plan/contract details, customer & agent demographics, customer-agent interactions, web log details, and customer satisfaction surveys.
- Evaluated multiple modeling techniques including Logistic Regression, Gradient Boosted Models (GBM), Random Forest (RF) and XGBoost models to determine key predictive variables.
- Analysed customer call transcripts using NLP and text mining to determine key phrases and themes that could guide agents with valuable talking points.
- Used LIME (Local Interpretable Model-agnostic Explanations) to explain individual predictions of the machine learning models and developed a dashboard to inform advisors and agents with insights and strategies.
- Models predicted at-risk customers that could be retained with over 85% accuracy
- Identified USD 1 Bn in assets under management (AUM) that could be secured through data-driven retention strategies