How Nudges Reduce Insurance Fraud: Small Changes Make a Difference

Dec 1, 2018

Signing image
Signing image
Signing image

Evie Monnington-Taylor is a Senior Advisor at the Behavioural Insights Team (BIT), working on international programs for the ‘nudge unit’. This blog is based on her talk at FRAUDtalks Conference 2017, where she advocated the effect of small changes to stimulate honest behavior. She will explain how nudges reduce insurance fraud. We all cheat. At least a little bit. Let me give three simple examples from my own life..

  • First I bought some breakfast, paid with a five pound note and received change for ten pounds. Did I mention the mistake, or did I just walk away with the money?

  • After that I took a taxi and when asking for the receipt I got one without date or amount on it. When claiming my expenses, did I write down what I paid or did I inflate the number a little bit?

  • In the evening, I received a call from my mother: she recently bought a car for my younger brother and she needs to insure it. I told her about the advice I got from a friend to make herself the primary driver, making the insurance premium much cheaper than having the insurance policy in the name of my inexperienced brother. She would not really be lying as she would use the car during the weekends. What do you think my mother did?

Why do we cheat?

I am not going to tell you what I did in those three situations, but research shows that if given the opportunity to cheat most of us do cheat, a little bit. A TV show in the US did a test whereby they gave diners extra change with their bill. It was $10 extra and just over 50% of the diners took the extra money and did not say a thing.

There has been evidence of cheating in the lab too. Academics Nina Mazar and Dan Ariely ran an experiment whereby they asked the participants to solve a number of problems and earn money per correct answer. What they were testing was how people reported their scores of this test. In this experiment they randomly divided the participants into two groups. The first group had to take their answer sheet to the experimenter who would mark the correct answers and pay out accordingly. But the second group was asked to go up to the experimenter and tell him their number of correct answers. In other words, they were given the opportunity to cheat for financial gain. Now let’s see what happened. The first group had an average of 3.4 correct answers. But those that had the opportunity to self-report had on average 6.4 correct answers. So when given the opportunity to cheat, they did cheat.

Preventing insurance fraud

And it is this kind of cheating with low level effort, which is hard to detect and difficult to punish that is costing insurers a lot of money. In 2013 the estimated costs for insurers and financial firms in the UK was 1.8 billion pounds per year. In general fraud is thought to add 50 pounds to every insurance premium in the UK. It is also impacting our public services as doctors and policemen are wasting time processing reports and dealing with people that are not actually sick. Imagine you could just prevent this kind of fraud instead of fighting it.

Examples from other industries to learn from

I will show you three ways in which behavioral informed changes have prevented low level cheating in three different areas. All of them are small changes, but the results from experiments show they have a surprisingly big effect. These changes could also be applied to the insurance industry.

Sign at the bottom - or the top

Example number three is a study looking at a policy review form used by a US car insurance company on which drivers had to declare how many miles they had driven in the previous year. The idea being that the more miles you declare, the more your car is used and therefore your premium has to be higher. And at the end of this form there is an honesty statement, saying “I promise that everything I have written in the above is true and correct”. And so, what they did was simple: they took this honesty statement from the bottom of the form and they moved it to the top.

They divided 13,000 drivers who were receiving this policy review form from the insurer into two groups and randomly allocated them to have either the honesty statement at the bottom or the top of the form. The findings? Those who sign after, at the end of the form, declare on average 30,095 km. But those who sign before declare on average 42,000 km. That difference was worth $48 per insurance premium for the insurance company, over $500 million for the whole group. Quite a lot of money. And again, it is a small change with a disproportional effect.

Tax in Guatemala

Guatemala has one of the lowest tax revenues as a percentage of GDP in the world. It is 12%, whereas the Latin American average is 26% and the world average is 32%. We wanted to see how we could improve this by making two small changes to the reminder letter sent out to those Guatemalans who were late in paying their tax. The first one we tried was: “According to our records 64.5% of Guatemalans declared their income tax for the year 2013 on time. You belong to the minority of Guatemalans who have yet to declare their tax.” And then we tried a different line in a different letter: “Previously we have considered your failure to declare as an oversight. However, if you don’t declare now we consider it an active choice.”

Then we ran an experiment and randomly divided the 43,000 negligent tax payers into three groups. The first group got the original letter, the second group got the first message with what we call social norms and the third group got the active choice message. What happened? In the control group, the average amount of tax paid per letter sent was $10. In the social norms group, we more than doubled that to $23 and with the active choice group we nearly tripled the amount to $29. Social norms are really effective in changing behavior because we all look to others to inform our own behavior. If no one else is paying their tax why should I bother to pay my tax? But if you tell people that the majority of people are paying their tax it can change their behavior, because we want to act like everybody else. The active choice is effective, because when people are cheating or lying it is easier not to correct misleading information. You are not actively stealing if you do not mention receiving too much change in a shop. What we did was to make not paying tax an active choice, thus removing the excuse of ‘forgetting’ or ‘not knowing’. It resulted in $800K extra revenue for the Guatemalan government in one year, which is not very much, but the amounts are quite small as you can see. We did something similar in the UK and it brought in 210 million pounds in one year. As the examples show, small changes can make a surprisingly big difference - but proper testing on effectiveness is key.

Speeding in the West Midlands

The next example is about speeding in the West Midlands, UK. People that get caught speeding will most likely not stop speeding, but just focus on not getting caught again. That means quite a lot of them are going to reoffend and get caught again and again. We wanted to see if behavioral insights could change that and tried to emphasize the negative effect that speeding can have. We took the Notice of Intended Prosecution (NIP) which you fill out in the UK to declare “I was the driver” and then pay the fine and accept the points on your license. We simplified this letter because it was very  unpleasant and complicated. We also added an extra page. Here we tried to emphasize the fatal consequences of speeding by using the symbolic image showing flowers by the side of the road and we added the line: “Over the last 5 years 779 children were killed or seriously injured on the road in the West Midlands alone.” We also tried to explain that a lot of thought goes into setting speed limits, taking into account the amount of accidents in the area because we don’t want history to repeat itself. We then ran an experiment with 11,000 drivers in the West Midlands and randomly divided them into two groups. The first group received the original letter, the NIP. The second group received the simplified letter, plus our additional page with the photo and the text. Looking at the number of drivers who were caught speeding again within six months, the results spoke for themselves. In the control group 3.1% of people reoffended within 6 months and in our treatment group only 2.45% did. This might seem like a small difference, but it is a 20% decrease as a result of sending an extra page in a letter that is sent out anyway. Again, this is a small change that really makes a disproportional difference to behavior. Imagine what could happen if you explain to your customer the consequences of cheating on their insurance or inflating their claim or misrepresenting their circumstances? Could that have an effect on your customer’s behavior as well?

One more thing: test, always test!

Imagine a world where you can use small changes like the ones in the examples above to prevent insurance fraud. They make a disproportionate difference. No more technology, budgets and teams for finding and fighting this kind of fraud, but preventing it at the source. Seriously consider incorporating small changes informed by behavioral science in your arsenal of tools to combat insurance fraud, because it is the way to prevent low level cheating. One more thing: test it, because it concerns very, very small changes and people might not believe they can make such a difference. And remember: every context is different. What works in the US might not work in the UK, what works on one form might not work on another. Consider small changes, they can have a big effect.