Demonstrated fraud reached €409 million in 2020, weighing heavily on insurers’ claims burden – and therefore on premiums. The current crisis should not see temptations slow down. Anti-fraud tools make the work of managers easier, but there is still room for improvement.
Insurers did not wait for new technologies to seek to detect attempts at fraud by policyholders, whose good faith is assumed by the insurance code. It was from 2015 that the intuition and flair of claims managers began to be complemented by digital tools. “From simple requests and manual reports, we have moved on to predictive algorithms, allowing us to cross-reference up to 100 pieces of data per claim”Explain Stephane Bezard, head of customer compensation at Thélem Assurances, which set up a fraud unit in 2017. A technological advance from the United States that has gradually spread to French insurers, starting with car insurance, then the HRM, two segments on which the data was the most numerous and the most qualified. “ The data is the basis. Sufficient data is needed to start using the tools and feeding the algorithms », confirms Thierry Cassagneres, Standard P&C Claims Manager & Compensation Solutions at Generali France. Today, the fight against fraud in insurance concerns all segments, both individuals and professionals. But the tools are not present in a homogeneous way at each phase of the life of the contract.
30 % : the additional share of detection thanks to digital tools in 2021
Source: Coalition Against Insurance Fraud
3,000 : the average number of insurance investigations carried out per year
Claims fraud: towards maturity
The vast majority of technological solutions deal with the detection of fraud during the disaster. “ The tools will make it possible to report alerts by cross-referencing internal data on the contract and the insured, with external data (manufacturer, mechanics, experts, social networks, etc.)”Explain Patrick Soulignac, Principal Solutions Consultant at Guidewire Software. An incapacity for work, for example, not being consistent with holiday photos posted on Facebook… “ The deployment of our solutions improves the overall system for combating fraud “, highlighted Benoit LegrosVP customer success at Shift Technology. “But beware, it is not the alert that proves the fraud! »remember Pierre-Louis BlancPresident of Alfa (Agency for the Fight against Insurance Fraud) and Director of Actuarial Management & P&C Fraud at Axa France.
The tools only make it possible to raise suspicions of fraud based on inconsistencies, missing information and recurring scenarios that must first be tested, then fed and updated regularly, in a process of continuous improvement. Then instructs the manager to sort through all the alerts in order to evacuate the “false positives” as quickly as possible, by putting them back in the flow of classic claims, in order to focus on the most relevant alerts. “ This selection step is tricky. Several issues must be taken into account: financial, commercial, brand image, fight against money laundering, etc. explains Pierre-Louis Blanc. Hence Generali’s initiative: the company has created its decision-making tool called Spider, which helps the manager to cross-check the data of the claim, its ecosystem and the third parties involved… “Even in the event that a manager has the intimate conviction that there is manifest fraud, if our technical position is too fragile, he must agree to let it go and trigger the compensation”, explains Stéphane Bezard.
In this regard, each insurer has its own strategy. At Axa, out of 100 claim files, about five are the subject of an alert in its departments and on average only one gives rise to an investigation. However, the efficiency is there. ” The day we moved from queries to algorithms, the relevance rate of alerts increased tenfold. And productivity has been divided by five, which is a good sign, because it proves that the suspicions are justified and that we must spend time on it. “, explains Stéphane Bezard. ” The insurers’ anti-fraud teams measure the relevance of the alerts generated by the solution and confirm whether the suspicion is proven or not confirms Benoît Legros.
Where does the fraud start?
False identity, mileage of the vehicle made up, inflated craftsman’s invoice…, the examples of fraud are only limited by the imagination of the fraudsters. If the insurance code does not mention the term fraud, Alfa (Agency for the fight against insurance fraud) defines it as ” an intentional act carried out in order to unduly obtain a profit from the insurance contract”.
Investigation, a real profession
Then comes the investigation stage, a profession in its own right that requires significant resources. “There are still few tools on the market for investigation, file consolidation and decision support », confirms Maxence Bizien, general manager of Alfa, which offers training in investigation methodology, an essential prerequisite before the tools. The simple fact of requesting additional documents, a second estimate or calling their craftsman can be enough to put off opportunistic fraudsters. But when it comes to finding concrete evidence, doing research on the Internet, cross-checking the data takes time.
“Our AI-based anti-fraud solution helps decision-making by explaining why the alert is generated and giving concrete elements of suspicion”, specifies Benoît Legros. Some publishers try to assert their added value against document fraud. But Stéphane Bezard remains cautious: “ Invoices are unstructured documents, very different depending on the company. It is therefore difficult to obtain satisfactory results. In addition, it has become technically very simple to modify a PDF or retouch a photo. »
Some frauds still too often go through the mesh of automatic detection and are therefore still largely compensated, even if the techniques are improving little by little. “The future lies in better automatic reading of documents and images. A better rate of transformation in this matter will make reliable what is detected with the naked eye “, hopes Thierry Cassagnères.
Anne-Claire Pichereau, associate lawyer at Neraudau
“The insurer remains responsible for data processing”
Is reconciling the fight against fraud and GDPR a problem?
The insurer always remains responsible for the processing of the data it has collected from the insured. The publishers of anti-fraud tools that use them, with the agreement of the insurer, are therefore only subcontractors.
In addition, since 2018, the GDPR requires insurers to inform policyholders that their data is likely to be used in the fight against fraud and money laundering. The processing of personal data therefore presents a “legitimate interest”, which constitutes its legal basis.
What is the legal framework for the burden of proof?
In the event of fraud, the insurers and their lawyers use the sanctions provided for in the contract, in particular in order to pronounce the nullity of the contract (in the subscription phase) or the forfeiture of the guarantee (in the compensation phase). The supporting documents are generally not initially communicated to the insured. However, the burden of proof rests with the insurer. It is up to him to demonstrate the insured’s bad faith and the existence of manifest fraud. The file compiled by the managers must therefore be solid, in the event that the insured party initiates litigation.
Improve detection upon subscription
If the suspicion is there and the financial stakes are high, it is always possible to commission an insurance investigator certified by Afnor as part of the process implemented with Alfa or to organize a shadowing. But all of this comes at a cost. In France, less than 3,000 insurance investigations are thus carried out each year on 12.7 million property and casualty claims in 2021. To decide on the advisability of further investigations, insurers have recourse to their lawyers or lawyers.
More mature on compensation, the efforts of insurers are now moving upstream. ” The fight against fraud must not neglect detection from the moment of subscription. We still have progress to make, learning lessons from recurring claims and proven fraud confirms Stéphane Bezard. ” For example, we have found that RIBs from neo-banks more often lead to fraud says Pierre-Louis Blanc. Tools for intelligent reading and for searching for alterations or inconsistencies can help them: falsified identity document, deletion of a responsible claim, modified care sheet, etc.
Three phases with more or less mature tools
- Compensation : the tools are mature and numerous (Shift Technology, Friss, Kube, Polonious, etc.). They send increasingly relevant alerts to managers thanks to machine learning.
- The subscription : document fraud detection tools (iT soft, omni:us, Kofax, etc.) that verify the digital integrity of a document or photo are beginning to yield encouraging results.
- The investigation : too few tools make it possible to consolidate the files to transform a doubt into proof. private investigators are only assigned to high-stakes cases.
A necessary global supervision
But the sophistication and multiplicity of tools lead to another problem. Whether offered by large groups or insurtechs, the tools are often specialized in one aspect of fraud (detection / investigation) and in one stage of the process (underwriting / compensation). Some are trying to broaden their spectrum, but the attempts are not yet conclusive. “When the tools arrived, we believed in the magic solution that would do everything, summarizes Maxence Bizien . But most only deal with detection, which is only part of the problem. And however intelligent it is, an AI can only do one thing at a time. Each use has its own tool. »
This multiplicity can harm the productivity of managers, who need to be trained. ” Most of the solutions on the market are not always well integrated into the information systems (IS), which can lead to a loss of time and performance”, confirms Maxence Bizien. Axa is thus limited to one or two tools, in addition to the main IS. Thélem Assurances bet on the test & learn in order to test the effectiveness of the tool and its good handling by fraud managers, who are often former claims managers. ” The ideal would be to have a global supervision tool, such as exists in cyber (Siem), security or nuclear, which would make it possible to aggregate all the detection and investigation tools so that they talk to each other, and thus be able to decide better “, believes Maxence Bizien. While waiting for future technological advances, Stéphane Bezard wants to be pragmatic: “Alerts are never more relevant than when they are manual, that is, when they come from a manager, an agent or an expert. »