In this conversation with Kai Bellman, a postdoctoral researcher at Delft University of Technology (TU Delft), he shares practical steps and personal reflections drawn from applying Behavioral System Mapping to a case involving adaptation to climate change.

Most of the problems the Bescy Behavioral Systems Group discusses—social, organizational, environmental—aren't simple. They're dynamic, interconnected, and shared across many actors. The group has featured interesting speakers who discussed the convergences and tensions within the Behavioral Systems field and the complexity of those "wicked problems." So when we sat down with Kai Bellmann, a postdoctoral researcher in the Multi-Actor Systems Department at TU Delft, the throughline of the conversation was the one that animates his research: how do you connect the micro level of individual behavior with the macro level of the system it lives in?

You can watch the full conversation with Kai here. 

Kai's path into the field is telling. He came up through business psychology, then political psychology and international relations, then governance and policy sciences and kept running into the same gap. Governance researchers would name a structural problem (say, insufficient finance for climate action) without quite acknowledging that the structure is made of the individuals inside it. His contribution to the topic as a behavioral scientist was that if we want system change, we have to understand structures through the lens of individual behavior and behavior change. The goal for all of us as applied behavioral scientists is to uncover levers for change in a system—and then work out where we can intervene most effectively.

The five steps of behavioral system mapping

The Behavioral system mapping draws on causal-loop diagramming, the UCL group's work on behavioral system mapping, and efforts at the European Commission and Monash University. What Kai’s work does is take stock and lay out a clear workflow. It all starts with a genuine behavioral problem big enough to warrant the effort and then:

1. System exploration and boundary setting: Start with an actor scan, not a target behavior. This is the key departure from how behavioral scientists usually work: instead of jumping straight to “citizens who fly on holiday,” you first ask who is actually involved in the problem—which surfaces a far wider set of behaviors worth addressing.

2. Data collection and behavioral analysis: Focus on behaviors—and on behavioral processes, sequences of behaviors across multiple actors. A single target behavior doesn't give you a map. Instead of “what influences your decision to fly?” you ask how the decision is reached: how people inform themselves, talk to family and colleagues, engage with the media.

3. Construction of the map: The exciting (and hardest) part. Build it iteratively and collaboratively, grounded in the causal relationships in your data, experimenting with the right level of granularity. A behaviors table can anchor it; participatory workshops are another route.

4. Validation and refinement: Some consider this the most important step. Take the map back to participants and outsiders to check whether it captures their lived reality, i.e., a co-design safeguard against the researcher's own assumptions.

5. Analysis and interpretation: This is where a map earns its keep. It's a twofold exercise: you don't just visualize your behavioral analysis, you then analyze the map itself—hunting for feedback loops, central behaviors and actors, and levers that could produce ripple effects. The output: behavioral dynamics, their determinants, candidate levers, and testable hypotheses.

One broad framework is essential for the analysis stage, something like COM-B or the Theoretical Domains Framework, because narrow, predefined models can't capture the range of determinants a real system throws up. (Monash's social-ecological COM-B, which adds system levels, is one recent effort in this direction.)

A tangible example: the reliance on precedents

To show what a map actually adds, Kai walked us through his research on Dutch climate-adaptation policymakers, which surfaced nine behavioral dynamics. His favorite is the reliance on precedents.

Local policymakers believe they need proven, cost-effective examples to convince local politicians and citizens to invest in adaptation. So they go looking for role models, often the two or three big municipalities with the budget to run pilots. By doing this, the results are that a disproportionate share of national adaptation ends up shaped by what happens in a handful of cities. And those cities are themselves driven by very specific citizens complaining about local conditions. Map it out and you can trace the whole chain, from a resident's complaint in one city to solutions adopted nationwide.

Where do you draw the boundary?

The obvious risk, once you start stepping back actor by actor, is ending up trying to “map the whole world.” Kai was candid that there's no perfect answer, but offered the practitioner's toolkit:

•  Instrumental: Include actors formally involved (via policy documents) or those who materially contribute to the problem.

•  Normative: Include actors affected by the problem, or those you believe should be involved.

•  Pragmatic: The availability of data is another important determinant; sometimes you have to map the actors you can get data on. The map is always a snapshot of part of a system.

Anchoring the scope to one actor's decision context (as he did with policymakers) makes demarcation far easier than starting from a broad societal issue with no focal actor. For those who want to go deeper, he pointed to Leon Hermans' work “Actor and Strategy Models” .

From map to intervention

A map isn't the end. From his map, Kai ran a Delphi panel with behavioral science and policy experts to rank behaviors by two criteria: which are impactful in the system, and which are actually changeable (financial constraints, for instance, rarely shift via behavioral means). That prioritization led to a concrete tool — the Climate Adaptation Reflection Navigator, a metacognitive checklist that prompts policymakers to reflect on the behaviors the map flagged as important, such as whether they've connected with other departments or looked for existing precedents.

Notably, the intervention landed as a boost rather than a nudge—building capability rather than steering choice. It performed well in experimental evaluation at the individual level and was refined through two review workshops with policymakers. Whether it shifts the system is the next, harder question, testable only once it's in practice.

The practitioner's notes

•      Tools: Kai favors Miro for its collaborative, sticky-note-friendly canvas. Vensim (system dynamics) and Kumu are alternatives; he often starts by sketching on paper or in PowerPoint to keep things loose and iterative.

•      Communicating the map: Never open by showing the full, arrow-covered map — people switch off. Start with a finding or a lever, explain it in your own words, then reveal only the relevant slice of the map. Keep sessions interactive and findings-led.

•      Timeframe: A participatory workshop can yield a draft map in a few hours (with prep). An interview-based map is a matter of months. His own analysis ran to roughly 3,200 codes, which distilled to a few hundred observable behaviors. AI can help with several steps, though he hasn't yet leaned on the emerging AI-assisted mapping methods himself.

•      Equity: The approach carries a built-in equity tilt, simply because it moves attention off the usual end-user and onto businesses, policymakers, and differentiated population groups. But it doesn't automatically deliver a social-justice lens; that still requires consulting other disciplines.

Where the method is heading

Asked about the frontier, Kai named two directions he's genuinely excited about. First, a temporal component—building comparable maps across time (a meeting a week later, a country a year on) to watch dynamics evolve and detect what interventions actually move. Second, system archetypes—recurring behavioral dynamics like “reliance on precedents” that show up across very different challenges, paired with intervention tools for addressing them. Quantitative links to agent-based models are on the horizon too, tying this conversation neatly back to our earlier ABM sessions (watch it here).

About us 

Behavioral Systems Group  is a group for individuals interested in the intersection of behavioral science and systems analysis. Why are we interested in this field?

Behavioral scientists provide a genuine understanding of what drives behavior, while also helping us make sense of qualitative data. Crucially, they offer a view of human beings that transcends the rational model of “homo economicus”.


Meanwhile, systems thinking takes us beyond individual-level determinants, focusing instead on the context that shapes decisions—how other actors influence an individual, and how these interactions reinforce specific dynamics. This allows us to design interventions tailored to how people actually make decisions, rather than relying on a standardized, "one-size-fits-all" approach to influence their behavior.


The goal for all of us as applied behavioural scientists is to uncover levers for change in a system—and then work out where we can intervene most effectively.

•  Bellmann, K. P., de Vries, G., & Scholten, L. (2025). Understanding slow progress on urban climate adaptation: An empirical analysis of behavioural dynamics in Dutch policymaking. Journal of Environmental Psychology, 105, 102666.  This paper is the empirical study behind the “reliance on precedents” example and the nine behavioral dynamics.

•  Bellmann, K. P. A Stepwise Guide to Behavioural System Mapping (working paper, SSRN). Here you have the five-step methodology at the heart of the talk.

•  Bellmann, K. P. (2025). Nudging alone won't produce lasting behavioural change. LSE Business Review. Check the short, accessible entry point to the argument.