Improving the lives of children, communities and society remains a core priority for our public services and policymakers. Yet, many Australians worry that their children’s futures will not be significantly better than their own. Frustration grows when some government services are not responsive to what communities are asking for or experiencing.

School completion rates – one of the strongest predictors of future socio-economic and health outcomes – are declining, with nearly one in five students in Australia not completing Years 10 to 12 in a way that sets them up for success. While this statistic is widely reported, it tells us little about who these students are, why they struggle or what can be done to better support them.

For improving school completion rates, many solutions already exist – varied education services, stronger local leadership, regulatory reforms and innovative technology use – but sustained progress remains elusive.

Similarly, childhood obesity, mental health and housing have long been a wicked problems, each with multiple interconnected factors. For instance, addressing childhood obesity likely requires more than improving diets or activity levels, but also strengthening parental wellbeing, specifically maternal education.

The complexity of multiple, intersecting domains often undermines policy responses or drives a focus on immediate problems, rather than upstream driving causes. So how can we better inform policy decisions, prioritise program trade-offs in tight budgets and implement interventions effectively?

Why does Bayesian Adaptive Learning matter for policymakers?

Policy cycles are not linear; they are adaptive, fast-moving and involve multiple decision-makers. Traditional evaluation methods are not keeping up.

Listening to lived experience is critical for informing policy priorities and adapting service delivery. However, existing systems often are not designed to listen at the scale or speed needed.

Evaluations such as randomised controlled trials (RCTs) can be too slow, too costly and miss the real questions policymakers and communities urgently need answered. A new paper argues that it is time to think differently, showing how Bayesian Adaptive Learning (BAL) can help us make faster, smarter decisions.

Rather than “evaluate at the end and hope for the best”, Bayesian Adaptive Trials (BATs) help policymakers learn as they go: adjusting, refining and improving interventions in real time.

BAL shifts evaluation from being a once-off audit to a continuous, embedded part of how policies are prioritised, designed and delivered.

The Bayesian Adaptive Learning cycle

The BAL cycle is simple but powerful:

1. Identify current beliefs about an outcome (e.g., school completion).
2. Test assumptions through targeted interventions.
3. Gather new data.
4. Update beliefs and refine strategies.
5. Test again.

Figure 1: Bayesian Adaptive Learning cycle

Each cycle helps policymakers zero in on what matters most – and what works best – for different communities and contexts.

The BAL cycle in practice

In education, Bayesian learning can adapt tutoring programs while they are underway, helping schools identify in real time which approaches best support different groups of students. No more waiting years for final evaluations.

BATs also help policymakers rethink long-held assumptions. For example, for a long time, it was assumed that reducing absenteeism was the key to improving school completion.

However, Bayesian analysis of longitudinal youth survey data (LSAC) shows a deeper story: a sense of belonging at ages 12 to 13 is a stronger driver of academic success and lower absenteeism than attendance interventions alone.

Belonging improves academic performance, general health and behavioural outcomes – all of which reduce absenteeism and boost school completion. Yet evidence about what works to foster belonging remains inconsistent.

Bayesian methods can help policymakers navigate this complexity by accounting for local variation: something traditional evaluation methods often miss.

What is new about this approach?

Bayesian reasoning has existed for centuries, but advances in computational power and digital platforms now make it a practical tool for dynamic policy environments. BAL does not replace traditional methods like RCTs – it complements them, particularly when:

• You need to make decisions before full certainty is possible.
• You want to adapt program design and delivery dynamically to adjust to different community needs and experiences.
• You need fast insights for limited budgets and urgent challenges.

Key benefits for policy and program design

• Faster, flexible learning: Programs adapt in real time as new evidence emerges.
• Cost-efficient decision making: Focus data collection on what matters most, saving resources.
• Real-world relevance: BAL is built for messy, fast-changing environments such as education, housing and community development.

Challenges policymakers should keep in mind

• Choosing the right tool for the right question: BAL is not the best fit for every policy problem. Policymakers need to be clear about what decisions they are trying to inform. As our colleague Alex Fischer discusses in a recent article, there are emerging methods to help ask the questions that matter most.
• Cultural change: Governments and agencies often seek certainty before acting. BAL embraces uncertainty, encouraging continuous improvement rather than perfect prediction.
• Building internal capability: Successful use of Bayesian methods requires investment in skills, systems and culture across policy departments.

Why it matters

At a time when public trust is fragile and budgets are tight, governments need to be smarter about how they learn, adapt and deliver real change.

Bayesian Adaptive Learning offers a path forward: an opportunity to embed learning loops into everyday policy work, ensuring that decisions are better, faster and more responsive to the people they aim to serve.

If we want public policy to truly change lives – especially for communities facing persistent disadvantage – we need to get better at learning while doing.

This article draws on the research paper, “Bayesian adaptive trials for social policy”,  which provides applied examples and statistical methods.

This article was co-authored by:

Sally Cripps is Professor of Mathematics and Statistics and co-director at the Human Technology Institute, University of Technology Sydney.

Ben Gales is the Executive Director of the Office of Social Impact, Queensland Government and former Chief Impact Officer at the Paul Ramsay Foundation.

Anna Lopatnikova is an Associate Professor the University of Technology Sydney and head of cross-functional programs at Human Technology Institute.

Alex Fischer is a Fellow at the Human Technology Institute, University of Technology Sydney.

Gilad Francis is the THRIVE program director at the Human Technology Institute, University of Technology Sydney.

Hadi Afshar is the lead research scientist for the THRIVE Program and Associate Professor at the University of Technology Sydney.

Roman Marchant is the Head of Research for THRIVE and Associate Professor at Human Technology Institute, University of Technology Sydney.

Image credit: agsandrew from Getty Images