21 December 2017 07:00
The artificial intelligence and deep learning could be of benefit to the participants of an advisor to accompany them 24 hours on 24, in addition to detecting insurance fraud. To achieve this vision, insurers will have to learn how to use the data.
Team leader of the data analysis centre at the national research Council of Canada (NRC), Stéphane Tremblay specializes among other things in machine learning. The analysis centre has been helping companies from all sectors face the challenge of massive data. The NRC is investing annually a billion dollars in research and development.
Particular entity, the centre for analysis account for approximately 80 people, including 30 experts in machine learning. «Our mandate is very close to the private companies. It is they who pay us for our services. We have also done a few projects with insurance companies «, he revealed during the Symposium in group insurance of the firm Segic.
Mr. Tremblay asked about the position of the industry compared to other sectors, in the field of artificial intelligence. «The insurers have made more investments than the average in the research. The implementation of the solutions is slower than in other sectors. «
Is it the culture of a stable industry which leads to a resistance to change, which it perceives as a risk ? The researcher cannot say with any certainty.
He adds, however, that insurers are very active in the experimentation and the assessment of risk. They can use these forces to better detect fraud and to better collect the information of customers. It will help them to reduce their costs, says Mr. Tremblay. «Insurers will have to achieve greater automation of tasks, if they really want to move on to artificial intelligence «, he warns however.
If insurers manage to overcome these issues, they may come to the advisor who works 24 hours on 24, has-t-he launched. The insurer who takes the challenge will have to collect more of its internal data external data coming from social networks or the connected devices «, sets out the researcher.
He will also have to deal with unstructured data. To explain this concept, Stéphane Tremblay has shared an experience with a company that wanted to analyze the rate of absenteeism. «Taking aerial photos of a parking lot, we could see a significant increase in the rate of absenteeism on Mondays and Fridays, and at the approach of the holidays. We were able to observe the same phenomenon by putting a sensor at the entrance door. We could also observe a variation of the temperature inside of the building. These are not analyses to be very intrusive. They have no impact on the private life, » describes Mr. Tremblay.
Similarly, employees with an identity card with an RFID chip (radio-frequency identification), or a watch, the Fitbit could be tracked in their movements, pose particular issues in respect of the protection of privacy, according to the expert.
Fraud : a global challenge
With all the necessary data, the consultant 24/24 will help insurers to deal with the issues of fraud and privacy, believes Mr. Tremblay. «Fraud has become a global issue for insurers, which make it a high priority,» he says.
There are several examples of applications and studies on fraud in other sectors, » he said. To determine the resistance of decision-makers, Mr. Tremblay believes that it may be hard to convince senior management to invest in the artificial intelligence to increase the efficiency of current operations, it will be much easier to arrive at the subject of the fraud. «This is a win-win solution «, he said, referring to the return on the investment.
To better convince, the project should give themselves as a goal to assess on a monthly basis the risk of fraud to all members. All of this will be achieved by collecting data internal, historic and annotated. «Data annotated to provide examples of fraud. All the history that is behind these data is used to model the case of fraud. A fraud never happens in a unique way, even if it will not always be perpetrated in the same manner through multiple examples, » explains Mr. Tremblay.
The sharing of data between insurers will add to the feasibility of the project, as well as access to social media data, cameras, public and other open data, said the head of the center for data analysis. He adds that the modelling will be done so easily. This will take more time is the cleaning and transformation of data. «It is necessary to convince them that it will take three months to develop an idea that takes a minute to express. After the modeling and the deployment also come the improvement and automation of the model. «
Adherents followed the trace
Then, the broker 24/24 must be positioned by the insurer as someone who accompanies all the members of a group plan, not only a fraction of them, said Mr. Tremblay. «We often observe this splitting in terms of fraud, where the insurer inspects a sample. Do it for all, they help to detect and reduce risk and fraud, » he said. Similarly, the model will inform the members of health risks and promoting healthy life habits from each, according to its peculiarities. «
For best success, it suggests to merge the internal data with a data collection transparent made from smart devices. «If the plan sponsor is able to convince the members to wear a watch, the Fitbit, for example, this would help to determine the risk factors. He can evaluate the risk individually and thus to reduce the costs by employees. «
The insurer or the plan sponsor who launches the adventure should expect to invest for the long term in a significant way. «Go to the approach of the prototype, in small steps. Implement a culture of automation. «
This is easier said than done, admits the researcher. «When I wanted to automate the process of Statistics Canada, several of which were still made by hand, I faced a lot of resistance. Many believe that the artificial intelligence is used to eliminate jobs, when in reality, it permits experts to realize their full potential rather than get bored with repetitive tasks, » said Mr. Tremblay.
It invites businesses to adopt a data strategy to increase their «maturity» analytical «. The company mature has internal data to be structured in silos, but also external data and unstructured data. It will append then data hybrid, either of the internal data integrated with external data. Its analytical capabilities previously descriptive to become prescriptive, passing by the diagnostics and predictions. The system will be able to make decisions in real-time and automated..