Bringing Analytics to Life

Reading time: 5 mins

Basem Morris reports on the intersection of predictive analytics and behavioural science to reveal some fascinating insights.

Human life is astonishingly complex. A lot of data being collected these days is attempting to capture an element of human behaviour, and most models being built are trying to predict some aspect of human behaviour. But what if we were able to, not only predict, but actually influence human behaviour? This all becomes possible at the point where predictive analytics and behavioural science intersect.

Behavioural Nudges

Nudge theory was made famous by American economist Richard Thaler in his 2008 best seller titled ‘Nudge’. When behaviour change is the objective, “Nudge theory argues that positive reinforcement and indirect suggestions can be more effective than direct instruction, legislation and enforcement.”

“This is me, planting an idea in your mind. I say, ‘don’t think about elephants’. What are you thinking about?”

Here are some examples of behavioural nudges:

  • People are generally averse to change. So what happens if we set the optimal choice as the default option? E.g. Set double-sided printing as the default setting on the office printer to save paper.
  • Since most of us have a tendency to conform to social norms, if I tell you that everyone else is doing it, will that increase your likelihood of doing it too?
  • We dislike finishing in last place. So what happens if you compare my behaviour to that of my peers and show me that I rank below the average in a particular area (e.g. my fitness)? Will that influence my behaviour?
  • If you walk into a supermarket to buy baby formula and you see a sign saying, ‘Maximum 6 tins per customer’, would you be nudged to buy 2 instead of 1? This is an effective technique known as ‘anchoring’.

One of the first applications of nudge theory came out of the UK, where the government ordinarily sends out reminder letters to people who are late in paying their taxes. The UK government’s behavioural insights team took a randomly selected group of non-payers and added an extra sentence to their reminder letter.

“9 out of 10 people in <your town> pay their tax on time.”

By simply adding this one line for the randomly selected group, tax compliance rates increased by 23% and saved the government £190m. Take a minute to think about the implications of this. By merely changing your words and tweaking the way you interact with your customers, you could make your organisation millions of dollars. In fact, many governments around the world have been quick to adopt behavioural nudges. The US and UK governments are leading the way and closer to home, the NSW government has a number of behavioural insights units, working on issues like domestic violence and child support.

The Intersection of Predictive Analytics & Behavioural Science

Predictive analytics alone can be a powerful tool; and an understanding of behavioural science is also powerful, even if adopted in isolation. But when predictive analytics and behavioural nudges work together, they accomplish more than they could by themselves.

Predictive analytics helps predict an outcome, while behavioural science helps influence the outcome.

Predictive analytics can produce some valuable insights, but it is often unclear what to do about these insights.

This is where predictive modelling is enhanced by the addition of behavioural science, as it allows you to operationalise your insights, convert insight into action, and bring your analytics to life.

Analytics professionals often focus on HOW to build an accurate model, whereas the main focus should always be on WHY the model is being built – to solve a real-life problem. Constant recognition of your ultimate goal is vital. A model alone is insufficient. Coupling analytics with behavioural science forces you to keep the ultimate goal of your analytics in mind – a real-life impact.

Equally, predictive analytics enhances the application of behavioural science, as it allows you to analytically identify segments of the population to nudge, as well as statistically measure the impact of nudges.

Therefore, behavioural science and an understanding of human behaviour are essential tools for any analytics professional, and vice versa.

Nudging New Mexico – A Real-Life Example

In the US, more than 1/8 of all unemployment benefits are paid to people who are not eligible for benefits due to dishonest claims. The government of the state of New Mexico wanted to tackle this problem and Deloitte was engaged to help. Can claimants be “nudged” to be more honest?

In New Mexico, the process of claiming unemployment benefits is all done online. Initially there is an online application process where claimants report that they have lost their job. From that point on, claimants log on weekly to report the progress of their job search. If certain criteria are met, the unemployment benefits are automatically deposited into the claimant’s bank account. Generally, people find it easier to be dishonest to a computer screen than to another person, face to face.

The first step was to identify the key areas in the application process where inaccurate information is usually given. These were:

  • The reason for unemployment: During the initial application process, applicants are asked for the reason they lost their job. Only people who report that they were laid off through no fault of their own are eligible for benefits.
  • Reporting weekly earnings: Any weekly income that you report will be deducted from your unemployment benefits.
  • Work search requirements: Applicants are asked to report on their weekly work search activity. To be eligible for benefits, you need to report that you have performed at least two pieces of work search activity (e.g. apply for a job, attend an interview, etc.)

Now that those areas are identified, what do we do about it? Traditionally, the government would throw resources at the problem to investigate would-be fraudsters. But what if we used predictive analytics to identify potential fraudsters and then tried “nudging” them, in real time, to do the right thing.

Here’s how it worked in practice:

I lost my job and I log on to the website to complete the initial application process. I am asked whether I was fired or laid off. The truth is I was fired, but I think it was extremely unfair! So I choose, ‘laid off’. At this point, the predictive models in the background flag me as an applicant with a relatively high probability of fraud. Immediately, a pop up appears on my screen showing a letter addressed to my previous boss asking him to verify that I was laid off, rather than terminated. This proved to be quite an effective nudge.

I’ve been unemployed for a few weeks now and I log on every week to report my weekly activity. This week, when asked about my weekly earnings, I think to myself, ‘Well, I did do a few shifts at McDonalds but I earnt nothing, just pocket money. Plus I’m sure everyone else is working part time and not reporting it. Once I get a full time job, I’ll report it.’ I choose ‘No income’. All of a sudden, a pop up appears on my screen, which I have not seen before on any of the previous weeks, saying, “99 out of 100 people in <your county> report their earnings accurately. If you worked last week, please ensure you report these earnings”. This proved to be one of the most persuasive nudges.

For the weekly work search requirements, claimants were asked to commit to a detailed work search plan for the coming week – which channel will they use to search for jobs? How many phone calls will they make? How many jobs will they apply for? It turns out that if people commit to a detailed plan, they are more likely to follow through with it.

As a result of these and other nudges, claimants were half as likely to commit fraud, twice more likely to report new earnings accurately and 20% more likely to find work in the next few months. And this project saved the government of New Mexico millions.

Imagine what the application of analytics, coupled with behavioural science, can do for your organisation.

CPD Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.

About the author

Basem Morris

Basem is a data analytics actuary at Deloitte who brings together a unique blend of technical expertise and a commercial acumen. With extensive experience both locally and overseas, he focuses on enabling businesses to realise the power of data through advanced analytics.

Comment on the article (Be kind)

Likes:0
Comments:0
Print

No Comments

Also this month