#Insurtech #RPA #Commando
Time To Market : 4 weeks
"Sipios were truly willing to adapt to our constraints : to code in R and to technically train the Product Owner. They are capable of adapting to a totally atypical project framework with a method both standardized and rigorous."
Mathilde Michel, Head of Claims Operations
GAN provides insurance for 1.5 million customers in France.
The managers of the indemnification service of the insurance firm have the duty to recover the amount of compensation paid to their customer from the person responsible for the damage or his insurer. The company receives each week 140 new litigation files, each representing several thousands of euros and needing to be processed 10 min by a manager.
We developed a robot using R that reduces the processing time for a litigation file from 10 minutes to 30 seconds by scrapping the document's data to automatically generate the dunning mail.
litigation files to process each month
to recover for indemnification per month
hours of process to automate
We used programming languages mastered by the operational staff to develop the robot so that the GAN teams may work on it in the future. The three key points:
.svg)
In the design phase, we detected the manual, time-consuming and tedious tasks for the operational staff. The stake was to keep the decision-making fully manual in order to guarantee the absence of errors and let the operational staff have complete authority over the process.
.jpeg?width=144&height=112&name=te%CC%81le%CC%81chargement%20(26).jpeg)
I designed a pdf content extraction system based on the analysis of letter structure. It was tested on a corpus of 1000 existing letters.
The second goal was to develop a scraping robot that would work on a web platform running initially on Internet Explorer 5. The robot detaches itself from the complexity of the interface and automates actions with no added value.
We had three weeks to identify the tasks to automate in the universe of litigation file processing. Michaël and I went to Lyon in the GAN recourse service managers' offices to observe their method for managing duns and identify the tasks with no added value that we could automate. Thanks to these observations and 5 interviews, out of the 7000 files that were late, we identified 1000 that were automatable.
