For an ethical approach towards data science projects in ‘pay how you live’ insurance.
Data is the new oil
If you ever look for data science-related articles online, there’s a 99% chance that you’ll come across dramatic, supposedly inspirational statements such as “data is the new oil”. Is this comparison premonitory? While oil was considered as the best energy source years ago, we are currently witnessing its negative impact on our environment, and rather see it as a curse now. Everyone is talking about the power of data science right now, however, people become more and more aware of the stakes of data management. Will data science know the same destiny as oil? As a business that heavily relies on data, the stakes of ethical data management are crucial for insurers.
What’s the place of ethics in the insurance debate?
Several current practices are raising ethical questions, especially regarding data science projects in life & health insurance. In the US as well as the UK, several insurers (including John Hancock Financial, one of the leaders in life insurance in the United States) have completely changed their business model: from 2019 onwards, they only offer life & health insurance policies based on their policyholders’ apps and connected watches, such as Fitbit and Apple Watch. These watches can be used to monitor everything from what you eat to your heart rate, blood pressure, blood-sugar levels, and sleep patterns.
This information is then used by the insurer to evaluate the policyholders’ lifestyles and to try to encourage them to have healthier behaviors. Insurance companies now offer a variety of incentives to policyholders who meet several very precise and demanding daily fitness goals: premium reductions, healthcare credit, store vouchers…
This might look like great news for policyholders because if they abide by the rules and live a healthy lifestyle, they’ll pay less and have discounts in their favorite stores! However, let’s see how this might become a problem ethically speaking.
On the one hand, customers want products that are custom-made, as close as possible to their needs and overall faster to get. Insurers are now, to a certain extent, able to offer them this type of product, but in return, they will ask for more and more information in order to meet these demands.
A kind of bargaining is underway: how much privacy are policy holders willing to sacrifice in order to access customable products? And how far should insurers go in their search for data and in its management, especially regarding sensitive information like health?
Several kinds of bias can happen with this business model. Whether it is for personal or systemic reasons, some people will hardly be able to meet the fitness goals set by the insurer, raising a discrimination issue. For instance, people with low income may not be able to afford healthy, organic and nutritious food and would probably struggle to get their recommended 10.000 steps a day if they live in a disreputable neighborhood. Additionally, if they have children but cannot afford childcare, how will they free up the time needed to exercise frequently enough to hit the goals?
Policyholders may be seduced at first by the possibility of saving on their insurance premiums, but what will happen when they do less sport? They won’t be able to go back and ask that their premium is no longer based on the data coming from their connected devices. Overall, this kind of pricing raises the ethical question of the freedom of policyholders: should insurance companies dictate the number of McDonalds trips you take and to what extent you go to the gym?
Also, what would happen to people who don’t want to have their lifestyle monitored by their insurance company? Will they have to pay more in the end? Will they even have the choice?
With this level of pricing personalization, what about the solidarity and mutualism principle that’s at the very core of the insurance business model?
Another pitfall of data science in insurance is the dilution of liability, with the risk that, when faced with an ethical problem, the only answer would be “it is the algorithm that calculated it, I’m not responsible for it”.
Ethics 101: how can insurers answer these considerations?
As Duncan Minty (Ethics Consultant in insurance) declared, the major challenge for insurance companies will be for them to actually take a step back and really question themselves in terms of (un)ethical practices by avoiding at all costs to think it’s okay since “everyone else is doing it”.
Ethics should be a central topic in insurance companies and should therefore be subjected to a high level of scrutiny within the structure. Just like for any other risk insurers can be exposed to, someone in the company should be in charge of monitoring data science projects and of conducting ethical impact assessment in order to ensure an ethical data management.
Insurers should implement a strict approach to:
- the data itself (including its quality, origin, and governance),
- algorithms (by testing historical data to detect possible ethical biases),
- the way they manage algorithms (by determining who manages, monitors them and who is responsible in the event of a problem).
Some principles should guide insurers in their data science projects:
- Transparency of the data
- Traceability of the data
- Absence of bias
- Consent from customers
- Respect for privacy
- Self-questioning from the companies regarding this topic
- Discipline in the processes
- Listening to feedback from customers
- Proportionality, by asking only for the information you really need
Insurance companies and regulators definitely have a role to play in ensuring ethical data science projects. Insurers must play a preventive role by explaining the advantages and disadvantages of connected technologies (such as fitbit) and set an example of transparency in the sector.
As for regulators, who have already taken measures regarding data management (ever heard of GDPR?), they should continue to monitor and anticipate the developments of the market to spot possible unethical practices. Regulators should consider whether it is necessary to harmonize the management of ethical issues at national or European level, or whether insurance companies should be autonomous in the way they deal with these issues.
In a nutshell, insurance companies should never ask themselves “can we?” without asking “should we?”. Without doing so, customers will lose trust in the sector and favor alternative actors who do value ethical data management in their company.
Written by Léonie Correard
Analyst at Asquare Partners