A one-size-fits-all approach to nudging can be ineffective, or even backfire. We know we need to recognise heterogeneity across and within groups of people to deliver successful behaviour change, but how to do this? Eyal Peer and Stuart Mills present a new framework to personalise nudging.
Why Personalisation Matters
Nudging – subtly guiding people toward better choices – has long been celebrated in behavioural public policy. But traditional “one-size-fits-all” nudges can fall short. Because people are different, they can work well for some, and poorly for others. A nudge which emphasises the pro-social benefits of vaccination might convince many people. But for some, nudges which tackle their fear and uncertainty might be more persuasive. Personalised nudging is the idea that interventions can be tailored to individuals or groups, based on relevant differences, to boost effectiveness and reduce inequality.
With digital environments and data analysis tools like AI becoming more pervasive, the conditions for personalised nudging are ripe. Yet, research into the area remains fragmented and inconsistent. Our new article proposes the first comprehensive framework for personalised nudging. We seek to disentangle a mess of terminology and perspectives by asking how personalisation happens. Doing so offers researchers and practitioners a common conceptual foundation from which to build, as interest into personalised nudging intensifies.
Five Features of Personalised Nudging
There are broadly five features of any intervention which matter for personalisation. Two of these could be called technical requirements. Firstly, data is needed. To personalise based on some relevant difference between people or groups, data about those differences must be available. Without data, a choice architect would not be personalising – they would be guessing. Secondly, the choice environment must be sufficiently malleable. Malleability is a new idea we bring to behavioural public policy. A choice environment is malleable if it can be changed. There is no point advocating for a social norm nudge if only a default option can be used. Personalisation is all about change, and if the environment is insufficiently malleable, the potential for personalisation becomes limited.
Three of these features could be called the behavioural requirements. Firstly, there is the content of the intervention. What does the intervention say? How much (or how little) does it say? Secondly, there is the design of the intervention. How does the intervention say it? Does it emphasize gains or losses, benefits to oneself or benefits to the community, and so on? Thirdly, there is the mechanism of the intervention. For instance, does the intervention lever the present bias, or the status quo bias? When we talk about personalisation, we are inevitably changing at least the content, design, or mechanism of the intervention.
The PeN Framework
We use these five features to systematically build up a taxonomy of personalised nudges – what we call the PeN Framework.
| Personalisation Level | What Changes? | Example |
| Named Nudge | (Simple) Content | Text messages which use recipients’ names to boost fine payments |
| Individualized Nudge | Content | Home Energy Reports showing personalized consumption feedback |
| Tailored Nudging | Design | Anti-smoking warnings customized by gender, emphasizing different health consequences |
| Targeted Nudging | Mechanism | Password nudges based on different decision-making styles |
| Adaptive Nudging | Everything | Medical feedback messages which change in content, design, and mechanism based on past behaviour |
Each level focuses on changing at least one behavioural requirement (content, design, or mechanism), and entails different amounts of data. For instance, a named nudge only requires a person’s name, and perhaps their pronouns. Target nudging, though, will often require behavioural and psychological data, such as a person’s personality, risk preferences, and so on. Most interestingly, levels like adaptive nudging will require what we call adherence data – data about a person’s past behaviours. Each level also creates new demands on malleability. For instance, as above, targeted nudging can only be used if each of the different nudge mechanisms available can actually be implemented.
Next Steps
The PeN Framework highlights some interesting questions about personalised nudging. For instance, consider targeted nudging. If one accepts that some people will be more persuaded by a present bias nudge than a status quo nudge (and so on), this also implies that – for some people – not nudging might be the best behavioural ‘intervention.’ This is a relatively unexplored area of behavioural public policy, but a natural conclusion of our framework.
The framework should also prompt questions about data and privacy in behavioural public policy. Are the benefits of a more persuasive intervention greater than some of the costs which might be involved in acquiring the data to personalise that intervention? This question highlights a potential misunderstanding of our framework which we wish to presage. The PeN Framework should not imply that any given level of personalisation is better than another. There is simply not enough evidence to determine that, for instance, targeted nudging is always (or even often) better than individualised nudging. The PeN Framework is better understood as a map of personalisation – given technical and behavioural constraints, what type of personalisation might be feasible to policymakers and other practitioners?
Finally, by presenting personalised nudging in this taxonomical way, we hope that the PeN Framework unifies multifaceted terminology in the field and galvanises important methodological and ethical research. For instance, there is growing discussion of whether personalisation should be bottom-up (individual-level data being used to design interventions) or top-down (population-level data being stratified to determine relevant differences)? In addition, there are important questions being raised about the distributional effects of behavioural interventions, and the potential role personalisation might play. These are important discussions which, we hope, our PeN Framework can support as personalised nudging becomes a more prominent feature of behavioural public policy, and behavioural science.
Read more here.
Eyal Peer is an Associate Professor at the Federmann School for Public Policy and Governance at the Hebrew University of Jerusalem, Israel. Peer is a social psychologist specializing in research on judgment and decision-making and a co-founder of For a Change (www.forachange.org.il) which focuses on applications for behavioral public policy that promote healthy and sustainable behaviors.
Stuart Mills is an Assistant Professor of Economics at the University of Leeds, UK. Mills holds a PhD in Economics and specialises in research on AI, Behavioural Economics, Dark Patterns and Digital Economy.