This post is also available in: Français (French)
Emin Ayik, claims director of Anadolu Sigorta, one of the biggest turkish insurers, presented a rich return of experience during Insurance Claims Management Conference in Prague in January 2020. Two main elements caught our attention and which we will present to you here.
The implementation of RPA solutions by Anadolu Sigorta
Anadolu Sigorta has deployed RPA (Robotic Process Automation) technologies to automate part of the insurer’s processing. This does not apply to all, or all of the time. More precisely, the RPA is beneficial for processes that are not complex and with a large workforce. In these cases, this allows transactions to be processed more quickly and without errors, as long as it is possible to apply a clear and strict procedure. So the gains were significant for Anadolu Sigorta, as you can see below.
Watson and damage recognition
First of all, you must have an order of magnitude of the volumes treated by Anadolu Sigorta. These are 11,000 new auto contracts per day and 1,200 claims, 63% of which relate to partial damage. With the help of the famous IBM Watson artificial intelligence engine (which I tested here for its chatbot part, [in French only]), the company implemented an automation solution for the detection and evaluation of claims, based on photos taken by the insured. Several steps were necessary:
- Data and photo collection: angles of view, distance, light or resolution are all variations that it is not always easy to correct. The company therefore used an input photo processing application
- Engine training: We had to teach artificial intelligence to recognize the elements of vehicles, as well as the severity of the damage, all on the basis of reference images for which the answer was known. This is supervised learning.
- Then, Anadolu Sigorta set up classifiers with several phases:
- Classification of parts: damaged part, undamaged part, element other than a vehicle part
- For damaged parts: classification between parts to be replaced and parts to be repaired
- Indication of severity of damage
The results achieved are beyond 80% success, which generates a significant improvement in operations. This solution is reminiscent of what Tractable offers, and which you can find on the following video.
To go further
Find the full Amin presentation support below:
|0 downloads||2020||adm_jinnbee||25-02-2020 8:54|
Let’s discuss your difficulties in this area to set up solutions adapted to your volumes and your needs!