Use data to predict vehicle breakdowns, reduce their frequency and enhance the effectiveness of preventive actions. actions.
To control the cost of breakdowns and improve the customer experience, insurers are deploying preventive measures for their policyholders.
The challenge was to go beyond a generalist approach, using data as a strategic lever to precisely identify :
- the most exposed policyholders,
- the most at-risk vehicles,
- the contexts most conducive to breakdowns.
In a data-rich environment (claims history, vehicle characteristics, geographical data, weather, etc.), the aim was to structure, cross-reference and exploit this information using data science approaches in order to transform a volume of data into operational recommendations.
The aim of the mission was to select the really relevant data and design specialized predictive models capable of detecting the signals that herald breakdowns.
Support in place
In-depth analysis of breakdown files along 6 axes: weather, vehicle age, make, geography, vehicle characteristics and insured profile.
This exploration phase enabled us to identify the first significant correlations, bring out weak signals and structure the key variables intended to feed the predictive models.
Implementation and training of predictive models to anticipate different types of failure with a high level of reliability.
An analysis of the importance criteria enabled us to identify the determining factors for each category of failure.
For example, the pneumatic class is more frequently predicted when several conditions are met:
- German vehicle brand,
- electric vehicle,
- mileage between 0 and 50K,
- less than 15 years' seniority.
These results help explain predictions, boost confidence in models and make AI directly usable by business teams.
Using the predictive scores generated by the models, define 3 targeting scenarios for each breakdown category.
This approach makes it possible to arbitrate between :
- very precise targeting of the most at-risk profiles,
- or a broader strategy maximizing policyholder coverage.
Technology thus becomes a strategic steering tool, making it possible to establish a segmentation that facilitates the design of personalized prevention campaigns, and to align economic performance, operational efficiency and customer experience.
The results
And tomorrow?
The results pave the way for larger-scale operational deployment: activation of automated campaigns, dynamic targeting of policyholders and integration of predictive scores into business tools.
Eventually, these models can be enhanced with new data sources and integrated into a continuous improvement process, so that prevention can evolve towards a predictive, personalized model, controlled in near-real time.
Data and artificial intelligence thus become levers of differentiation and performance. .
