Artificial intelligence and deep learning may soon deliver 24/7 advisors that double as insurance fraud detectors. First, insurers must learn to use data.
Stéphane Tremblay, team leader at the National Research Council’s (NRC) Data Analytics Centre, specializes in automated learning. The Analytics Centre helps businesses in all sectors face the challenge of big data. The NRC invests $1 billion in research and development each year.
The Analytics Centre stands out because its staff of about 80 includes 30 experts in automated learning. “Our mandate is very close to private businesses. They pay for our services. We have done a few projects with insurance companies,” Tremblay said at a recent group insurance conference hosted by Segic.
Asked about the industry’s progress in artificial intelligence compared with other sectors, Tremblay replies: “Insurers have invested more than average in research. The implementation of solutions is slower than in other sectors.”
Could the typically static culture of the industry be causing resistance to change, which it perceives as a risk? He would not confirm this possibility.
Tremblay did mention that insurers are very active in experimentation and risk assessment. They may use these strengths to detect fraud and gather customer information more effectively, which will help them reduce their costs. “Insurers will have to automate tasks much more if they really want to adopt artificial intelligence,” he cautions.
If insurers manage to overcome these challenges, they may be able to deliver advisors who work 24/7, he continues. “[These insurers] must gather both internal data and external data that come from social networks or connected devices,” he says.
They must also process unstructured data. To explain this concept, Tremblay shared the story of a business that wanted to analyze its absenteeism rate. “By taking aerial photos of a parking lot, we could see that absenteeism spikes on Mondays and Fridays, and just before statutory holidays. We observed the same phenomenon by putting a sensor at the entrance. We also noted a variation in temperature within the building. These are not very intrusive analyses. They have no impact on privacy,” Tremblay explains.
Similarly, for employees who have identity cards with a built-in RFID (radio-frequency identification) chip or a Fitbit watch, their movements can be tracked without raising any issues concerning privacy protection, he adds.
Fraud: a global challenge
Armed with sufficient data, the 24/7 advisor can help insurers tackle fraud and privacy issues, Tremblay says. “Fraud has become a global challenge for insurers, which make it a top priority.”
There are many examples of applications and studies of fraud in other sectors, Tremblay adds. Regarding decision-makers’ resistance, he says that although it may be difficult to convince top management to invest in artificial intelligence to enhance the efficiency of everyday operations, it is much easier to convince them of AI’s ability to deter fraud. “It’s a win-win solution,” he says, referring to the return on the investment.
To attract more interest, the project should aim to assess the risk of fraud for all members, on a monthly basis. This involves gathering internal, historical and annotated data. “Annotated data give examples of fraud. The history behind this data can also model cases of fraud. A fraud is never unique, although it is not always perpetrated in the same way across various examples,” Tremblay explains.
Data sharing between insurers adds to the feasibility of the project, as does access to social network data, public cameras and other open data, he explains. It makes modelling easier. You just need to take the time to clean and process the data. “We will need to convince users that it takes three months to develop an idea that takes a minute to express. After modelling and deployment, you also need to improve and automate the model,” he underlines.
Insurers must also position the 24/7 broker as serving all members of a group plan, not just some of them. “We often observe this splitting in fraud, when the insurer inspects a sample. Doing it for everyone makes it easier to detect and reduce risks and fraud. Similarly, the model can inform members of health risks and promote a healthy lifestyle for everyone, depending on their particular case,” he explains.
The best approach is to merge internal data with transparent data collection via smart devices, he adds. “If a plan promoter can persuade members to wear a Fitbit watch, for example, it can determine the risk factors. Individual risk can be assessed, and the organization can reduce the costs per employee,” he says.
Insurers or plan promoters that plunge into artificial intelligence should expect to invest a large sum over the long term. “Go to the prototype approach step-by-step. Implement a culture of automation,” he recommends.
This is easier said than done, such as employee resistance. “Many people think that artificial intelligence will eliminate jobs, but in fact it lets experts realize their full potential rather than get bogged down in repetitive tasks,” he explains.
He invites businesses to adopt a data strategy to increase their “analytic maturity.” A mature company will have internal data structured in silos, along with unstructured external data. It will then add hybrid data, whcih is internal data integrated with external data. Its analytical capacities will evolve from descriptive to prescriptive, by incorporating diagnoses and predictions. The system can then make decisions in real time.