Social media intelligence and profiling in the insurance industry…

…it’s not only the price you pay that will be affected

This piece was written by PI research officer Dr. Tom Fisher.

The financial services industry is eager to gather more and more data about our lives. Apart from mining the data they have historically collected such as credit history, they are looking to use our social media profiles to reach into our friendships and social interactions. They are using these data in new and unexpected ways, including personality profiling to determine the risk of lending to you, and thus the price you will pay.

Firstcarquote, a product from the car insurance company Admiral Insurance, is a leading example of such unexpected data exploitation. In a semi-aborted product launch in November 2016, the company proposed to use young drivers’ Facebook posts to determine their personality and use this information to offer discounts on car insurance. Not limited to causing damage to its own reputation, Admiral’s move raises serious questions about the future of the insurance industry.

While Admiral may have been forced into backing away from the use of young people’s Facebook posts to price insurance, the company is still working with companies that are involved in in the business of personality profiling and using social media intelligence to make predictions and decisions about people. Social media intelligence is collecting and analysing data from social media networks, like Facebook; this can include information contained in profiles, posts, images, and metadata like location. Are we ready for the consequence of social media-based risk scoring?

Admiral’s firstcarquote

In November 2016, Admiral Insurance announced a new product, firstcarquote, which used Facebook profile posts to determine young drivers’ personality and risk profile. Media coverage was amplified when Facebook announced the product would be blocked for breaking the terms of their Platform Policy.

The exercise may have been a media stunt for “a fairly standard new app for first time drivers” allowing Admiral to claim on firstcarquote.com it was “Probably the most talked about car insurance product ever …”. After the failure of the first lauch attempt, Firstcarquote then relaunched with a mandatory “login via Facebook” and personality quiz replacing the analysis of Facebook posts.

Personality profiling and insurance

This interest in social media intelligence was not just a one-off flirtation by Admiral. Data from social media is beginning to play a growing role in the financial sector, and Admiral continues to market personality profiling as a legitimate way to offer young drivers discounts on their car insurance.

In relation to firstcarquote, the app still has a mandatory Facebook sign in, but this grants access to “account information”, rather than a person’s posts. Account information is an individual’s name, email address, gender, birth date, current city, and their profile picture. An app that asks for this information is not required to go through Facebook’s app review process.

More interesting and less covered than the app is Admiral’s collaboration with VisualDNA. VisualDNA is a personality profiling company, with a focus on credit scoring for lenders. According to a MasterCard report, VisualDNA’s work in the credit sector measures attributes like “openness”, “neuroticism”, and “emotional stability” to assist with credit scoring, particularly for those “thin-file” customers who had a limited credit history.

Both parties saw this as beneficial partnership. VisualDNA, who having “already proven the benefits of applying our patented and unique psychometric tests to credit assessment” were “delighted to have the opportunity to work with Admiral to demonstrate the power of this approach to other forms of risk assessment such as motor [insurance].” Equally, Admiral believed “The more we know about our customers’ likely behaviour the better placed we are to price risk on an individual basis, giving customers an even more accurate price.”

The depth of the relationship is perhaps seen in the involvement of computer scientist Dr. Yossi Borenstein as “principal data scientist on firstcarquote”, who had previously worked with VisualDNA, where he was head of risk analytics. He has also published on the subject of VisualDNA’s personality tests.

Further, Admiral use VisualDNA’s cookie on their website, although it is unclear as to whether this is for profiling for advertising or risk purposes.

Admiral’s initial press release made a broad claim for the validity of personality testing for young driver insurance: “It’s scientifically proven that some personalities are more likely to have an accident than others.” They later weakened this to “a growing body of scientific evidence that…personality traits can be linked to driving risk”. Finally, as a post on their Facebook page puts it, “So, does your personality affect the way you drive? The short answer is: probably.” But, whether there this is a link between risk and personality for young drivers or not, the use of personality testing, and new data sources, has broader consequences for the insurance sector.

Social media data as intelligence

The use of social media intelligence is a developing field in the financial sector. Facebook’s Platform Policy may have prevented access to the data about people’s posts on this occasion, but the use of social media data to make predictions and decisions about people in the insurance sector is developing.

In this instance, the firstcarquote case attracted a great deal of publicity and Facebook deemed that the initial version of firstcarquote violated its platform policy. It purportedly measured personality by analysing, for example sentence length and use of exclamation marks; words like “always” or “never” (as opposed to “maybe”) could suggest overconfidence; and arranging to make appointments at a specific time, rather than a generic “this evening”, was allegedly a measure of how organized an individual is. Facebook’s platform policy governs how third-party apps can make use of Facebook. According to Facebook, Admiral violated term 3.15, which was introduced into the policy in May 2016:

“Don’t use data obtained from Facebook to make decisions about eligibility, including whether to approve or reject an application or how much interest to charge on a loan.”

Despite this, data from social media still play a role in the financial services industry. This is particularly the case for groups of people who have not had access to financial data before — requiring these sectors of society to be required to reveal more about themselves than others. For example, the use of social media intelligence for loans is aimed at those who do not have an existing credit history — typically, the poorest and most vulnerable. Similarly, with firstcarquote, it was a group without previous access to affordable insurance who were required to reveal more about themselves. It begins to seemthat some groups in society are more able to keep their privacy than others. Is this fair?

One typical example of the use of social media data is the work of Big Data Scoring (BDS), a credit scoring company who took on Admiral as a client in March 2016 announcing at that time, BDS “tap[s] into the broadest source of information from across the Internet, using all publicly available information”.

One of their products, BDS’ Digital Footprint DATA gathers information from a variety of sources that are considered ‘publicly available’, including some social media, blogs, and information through Google searches. It checks information like the device a person is using, and detailed information about their location. We note that the description of the cookie on the Admiral site includes,

“information about the user’s web browser and computer, which pages are accessed on the loan application site, how long the user spends on each page, how the user interacts with the pages and what actions they take during their visit.”

Detecting how a website is used can be a tool for fraud detection, and detecting bots. But cookies enable the linking of data to draw conclusions on every user of the loan site: the way in which you fill in an online form is becoming as important as the information we put into the form.

BDS’ original product from 2013–4, offered a social media score to loan providers to supplement or replace their existing decision-making processes. After giving access to a broad range of Facebook data — including profile data, status updates, likes, and locations — the algorithm made a decision on the likelihood that an individual would repay a loan. As the CEO and co-founder, Erki Kert, stated,

“All in all, that is c.a. 5,000–10,000 lines of data for each client. …. We often know more about borrowers than their family or relatives do.”

He also downplayed the serious privacy implications of such a statement:

“From applicants point of view, allowing access to all this Facebook data costs just 2 mouse clicks”.

The social consequences of industry shaping human behaviour

Young driver insurance doesn’t just affect the wallets of young drivers: it has an impact on society more broadly. The price of insurance isn’t only a measure of risk: it also potentially affects behaviour. And influencing the behaviour of young drivers is important. Despite making up less than 2% of license holders, young drivers are involved in almost 10% of all fatal and serious crashes.

More traditional pricing has the potential to have a positive impact on the safety of our roads: for instance, by pricing more powerful (and risky) cars higher, it encourages young drivers to choose more appropriate vehicles. While the price of insurance isn’t the only factor affecting the behaviour of young drivers, it is as the very least using another tool to encourage them to use the roads more safely, benefiting us all.

But, think about how basing price on social media posts would change behaviour. Would encouraging young people to use full sentences on Facebook do anything to make roads safer? Where is the social benefit of that?

In fact, we can see a social harm in the using social media in this way in financial services. It can change the way we communicate, which affects us all: it removes the honesty from our communications if we think that it might affect our financial standing. Is a friend posting a message because of a genuine desire to share their lives with their friends and followers, or are they giving a particular impression because they’re looking to apply for a mortgage in 6 months time?

The elements of social control in this can be more insidious. For instance, China is developing a social credit system; this is reportedly a method to influence behaviour online by developing a “social credit score”. Similarly, a lender in India is rejecting loan applications from people who post political messages on Twitter, as the lender fears that it is more difficult to get money back from the politically active if they don’t repay.

The future of the insurance sector

As the CEO of the financial services tech firm ZestFinance, and former Chief Information Officer of Google said in 2013, “We feel like all data is credit data, we just don’t know how to use it yet.”. The same could be said of the insurance industry, but this has consequences for the insurance industry itself.

A 2016 UK Financial Conduct Authority publication on the role of “big data” in the consumer financial market, identified as a risk the increased segmentation of the market, with the consequence that those customers deemed riskier are priced out the ability to get affordable insurance.

The risk can be taken to its conclusion: with more and more data collected, with a strong claim of the predictive power of algorithms, the size of the segments in the market are reduced, eventually to one. In other words, an insurance company would be able to quantify the precise risk of an individual. At some point along this process, insurance as we know it faces an existential threat: insurers are no longer working on the basis of collectivised risk that is at the heart of insurance. Thus the change to insurance, and how it works is a major one — that affects all of us in society.

Conclusion

Despite the measures taken by Facebook, it’s clear that both personality profiling and social media intelligence continue to have a continuing relevance in the insurance sector. This presents a challenge that goes beyond the price that individuals pay for their insurance, and it affects us all. We have to ask the question, is this fair? And, ultimately, there is a danger to the insurance industry itself.

Using social media intelligence in the financial services industry risks affecting how we present ourselves on social media. The selves that we present on social media become the selves that we would present at a meeting with our bank manager: feeling slightly uncomfortable in our Sunday Best, avoiding at all costs controversy, politics, and protest. That is, at the very least, a less interesting world to live in.