Insurance-Machine Learning and AI


Insurance use cases

Customer touch points have exploded in the last few years, especially due to increase in digital interactions

  • Fragmented and noisy data
  • Higher Volume, Velocity and Variety of data
  • Datasets could be present on third party systems
  • Complex and fuzzy matching may be required to combine data
Insurance use cases

Big Data tools and analysis have become essential to handle these new requirements

The benefits of getting a comprehensive view of customers are many,

  • Better profiling - A more nuanced understanding of customers’ needs, preferences and state of life
  • Reduced customer churn - Early identification of dissatisfaction, quicker complaint resolution
  • Empowering employees - Enables employees to better convert cross-sell and upsell opportunities

Effective Marketing through Big Data and Marketing Analytics

Media consumption has become extremely fragmented with customers spending more time on online and mobile platforms

Big Data and Marketing Analytics

Consumer behavior is now influenced at various moments and touchpoints and the traditional marketing funnel has become less applicable.

Big Data can help make sense of this complexity in various ways,

More granular segmentation

More granular segmentation

  • Better measurement
  • Accurate targeting
Full data modeling

Full data modeling

  • Sampling biases avoided
  • Robust model estimates
Machine Learning

Machine Learning

  • Discovering non linear relationships
  • Exploring more variable combinations
Cross product relationships

Cross product relationships

  • Identifying synergies
  • Upsell / cross-sell oppoortunities
Joint Optimization

Joint Optimization

  • Marketing Effectiveness
  • Marketing Efficiency
Attribution modeling

Attribution modeling

  • Bottom up approaches to determine credit
  • Better understanding of customer journeys

Text Mining of Claim reports

Problem: Analyze text in claim reports to understand claim causes

  • New predictive variables can be derived from text, which can be used together with traditional variables for better risk assessment

Data Cleaning

Keywords from raw data

  • Removing stop words (‘of’, ‘is’, ‘the’, etc..)
  • Stemming verbs (damaged, damaging damag)


Identifying patterns

  • Keyword should occur at least 6 times in the document
  • Hclust clustering


  • Most common keywords (causes)
  • Co-existing keywords
  • Predictive power of keyword occurrence in risk assessment