Business World

Big Data for nudging

Nudging is the careful construction of decision-making environments aimed at helping decision makers making wiser choices, possibly at their advantage. The first applications of nudges regarded mainly public policies and programs aimed at a general public, with no particular distinctions between potential users. Tax compliance, energy saving and even increasing men’s aim in public toilets (interestingly achieved drawing a fly in urinals), just to cite a few of the issues tackled with nudges, were addressed with unique solutions thought to maximize (or at least increase) the average response of the public. As of today, social media and new technologies make it possible to take communication of firms and organizations with consumers at a personal level.

Turns out, the same nudge does not have the same level of effectiveness in every situation and on every person. Context matters, both for the nudgers and the nudgees[1]. This is why, while nudges have been evolving since they were first studied and applied, the recent outstanding innovations in fields like AI, Deep Learning and so on, powered by the increasing availability of enhanced computational power, data and Big Data have attracted the attention of behavioural scientists and choice architects.

One of the first suggestions of personalized nudging comes from Cass Sunstein, co-author of the book “Nudge”, in a paper of 2012 in which he recognized that “The existence of heterogeneity argues against impersonal default rules” [2]and suggested instead that in those situations in which this occurs and active choosing is highly burdensome for the individuals, personalized default options are provided by the firms or issuing organizations themselves.

The combination of nudging techniques and the use of big data and machine learning leads to so called Hyper Nudges, highly personalized nudges that take into consideration individual needs, responsiveness to previous nudges, social context and every possible interesting metrics that can potentially influence the responsiveness to a certain nudge in order to choose those manipulations of the choice architecture that will have the highest effect and bring the highest utility to the subject.

Personalization can mainly occur in two ways: within nudges (measuring the effect of the nudge, for example tailoring the default option in an opt-out program) or across nudges (applying different nudges to different individuals).

Pros, Cons and some challenges: What about privacy?

As we have already seen, tailoring nudges to individuals leads to more effective policies and applications to programs that benefit their users, while listening to the necessities and goals of every consumer. The ultimate goal is not to increase application and compliance to generally beneficial options at any cost, but to listen to everyone to measure the effect of that specific measure and prevent that the few ones for whom such option would be if not useless even hurting (for example including someone who already saves too much in an opt-out retirement plan) will be subject to the same policies.

Traditional nudging has always been subject to criticism from those who fail to see the balance of the so called “Libertarian Paternalism” (as Cass Sunstein and Richard Thaler called it in “Nudge”) at work:  nudges are a way to guide individuals toward best choices for them (paternalism) but in a gentle way, leaving untouched the right to choose, even the right to make bad decisions (libertarian). Common criticisms to the theory of nudges are usually accusations of being too libertarian or too paternalistic.

The use of Big Data, on the other hand, has always attracted criticisms regarding the privacy of the data subjects. The collection of personal data, but also of more “innocuous” data such as Cookies and anonymous metrics, has always been considered alarming, and countless are the legal cases linked to this issue.

It is therefore of no surprise, that the combination of two such controversial techniques is also often seen with a negative eye. The challenges are approximately always the same: granting the access to Big Data and guaranteeing transparency in the use of such data, task made even more complex by the fact that there is not a unique nudge to make transparent anymore, but rather the fact that everyone is subject to different changes in the choice architecture. Finally, another critical issue regards a series of ethical and moral questions linked to personalized nudging.

Libertarian Paternalism is still applied, therefore the right to opt out from personalized nudging is always available, but the question now becomes: what if personalized nudging becomes too costly for the user to turn down? In other words: what if hyper nudges actually increase so much the quality of services using it that the option to opt out becomes substantially superfluous? Is such a scenario still to be considered as libertarian?

Laws and regulations will probably have to adapt to the increasing attention in personalized nudging, in order to limit the personalization to avoid ethical issues while at the same time allowing and fostering innovation and research on the subject.

Personalized Nudging and Language Learning

I am pretty confident that the majority of the readers of this article know what Duolingo is. The platform allows learning of a great number of languages through continuous exercises. To be effective, Duolingo needs that its users access the platform at least once a day, completing at least one of the sets of exercises provided. Whoever tried to learn a language on their own knows that being constant is hard. Duolingo knows it as well, at the point that skipping one day of exercise will disappoint Duo, the green owl, as an incentive for the users to stop procrastinating.

To increase the number of consecutive days that its users spend doing exercises (the so called “streak”), Duolingo sends notifications every day, to remind the users of their daily dose of learning. Reminders are one of the most popular forms of nudge. They reduce the cognitive load coming from having to remember things that one needs to do, with benefits both for users and providers of the services reminded. But Duolingo takes this a step further, sending different notifications to different users according to their responsiveness, language studied, recent notifications sent and so on.

Users are grouped in clusters of individuals with similar characteristics, and every new notification is tested on small groups of representatives from each cluster. A bandit algorithm is used to update the score of each notification (given by the percentage of users responding to the notification in a specific time frame), based on the effectiveness of previous choices. In this way, the messages sent to the users are carefully chosen among the ones that maximize the probability of them taking the lecture.

This simple example of nudging personalization allowed the company to increase both the existing users’ use of the platform and the retention of new ones, with benefit for all the language learners, who managed to study and learn more. The data used was data already retained by the platform and the students are not forced to complete any lesson, making it consistent with the concept of libertarian paternalism. Moreover, transparency is guaranteed by the fact that users can easily have access to Duolingo’s blog page also from the App, page in which such mechanisms are described in a very straightforward way.

If you wish to learn more about the concept of Liberatarian Paternalism, check out the links in the following page.

(scroll down to access the next page)



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s