Rethinking anticipatory action in conflict situations

What if the very assumptions that underpin most anticipatory action models are the reason they struggle in the face of conflict and violence?

Anticipatory action has gained momentum across the humanitarian sector. But its application to conflict and violence remains significantly less developed than its application to climate-related hazards. Much of this gap stems from the expectation that conflict can be predicted with the same relative certainty as climatic events, and that pre-agreed triggers and anticipatory action plans can define when and how to act.

While fixed-risk models, which are structured around fixed triggers and pre-agreed financing, are effective for more predictable hazards, they are less able to anticipate the volatile, non-linear dynamics and complex political and social conditions that characterise violence and conflict. How do you take anticipatory action in conflict situations in the face of such uncertainty? New research suggests that acting early before a violent crisis peaks is not only feasible, it shows real promise for preventing and mitigating the humanitarian impacts of conflict and violence.

Based on the first comprehensive review of the Start Fund’s alerts for conflict-induced displacement and election-related tensions, this blog explores how flexible, expert-led decision making and qualitative risk analysis are enabling effective anticipatory action in contexts where fixed-risk models are either inappropriate or unavailable.

Why triggers can fall short in conflict settings

Conflict-driven crises stem from complex and volatile socio-political and economic dynamics. This makes them difficult to predict using traditional, data-driven models. Unlike natural hazards, which have regular patterns and forecastable data, conflict lacks consistent indicators, making it hard to establish predefined thresholds or action plans. Insisting on fixed triggers in such settings can mean missing critical windows to act.

Forecasting conflict requires a fundamentally different approach from natural hazards. The Start Fund’s approach to anticipatory action is uniquely suited to conflict contexts precisely because it avoids trigger-based financing. Instead, it relies on dynamic, context-specific risk analysis and expert consensus to inform funding decisions.

Unlike data-driven models which use pre-agreed triggers, action plans and prepositioned funds, the Start Fund operates flexibly, releasing funds based on emerging needs. The Start Fund’s ‘no/low-regrets’ approach also enables agencies to act on early warnings even if crises don’t fully materialise, with the option to return unspent funds. This reduces the pressure caused by the need to wait for a crisis to escalate before taking action.

This adaptability is especially valuable for anticipating violence and conflict, particularly in fragile and conflict-affected settings where the risk landscape can shift rapidly. The Start Fund’s dynamic risk analysis uses diverse, hyperlocal data – both qualitative and quantitative –allowing for a more nuanced understanding of risk as it evolves. By drawing on local expertise and focusing on humanitarian impact rather than precise markers of conflict onset, the Start Fund’s approach offers the flexibility to act in fluid environments. Its expert-driven, consensus-based model enables faster, more context-sensitive decision making, typically within 72 hours.

But while this human-centred process allows for greater adaptability, it also introduces challenges, such as information gaps and potential biases, especially when forecast data is limited or uneven.

In conflict settings, timely and relevant data is often sparse, politically sensitive or rapidly outdated. As a result, Start Network members frequently draw on qualitative data and analysis tools, such as trend analysis, historical comparison and expert insight, to inform their alerts. While these methods may lack statistical depth, they allow agencies to triangulate risk information using local knowledge and on-the-ground observations.

Decentralised, expert-led decision making

One of the Start Fund’s strengths is its decentralised allocation process. Decisions are made within 72 hours by a committee of Start Network members who review each alert based on its credibility, relevance and alignment with the fund’s niche. This naturalistic, consensus-based process allows for rapid action while maintaining a strong emphasis on peer accountability and local legitimacy.

Importantly, this approach embraces, and is deliberately built to enable, decision making within uncertainty. Instead of aiming for predictive certainty, it focuses on responding to the risk of humanitarian impact. It accepts that forecasts will never be perfect and instead asks: do we know enough to act now, even if the crisis hasn’t fully unfolded?

The Start Fund’s model does not offer a silver bullet. But it can be appropriate for use in conflict contexts because it acknowledges and works within the constraints of those environments. In settings where data is poor, politics are complex and risks evolve rapidly, the Start Fund enables anticipatory action where others may not.

Moving forward: what can the sector learn?

As interest in scaling up anticipatory action continues to grow, the sector must grapple with how to extend it into and for conflict-affected contexts. This requires rethinking the way we understand anticipatory action and embracing models that are flexible, dynamic and hyperlocal.

Start Network is building further capacity in this area with a three-year project on anticipatory action in conflict settings funded by the Netherlands Ministry of Foreign Affairs. This project seeks to strengthen and scale up the Start Network’s approach to conflict anticipation and build on learning to better prevent and mitigate the humanitarian impacts of violence.

The Start Fund’s approach may not provide guarantees, but it enables anticipatory action in places where fixed-risk models are unavailable or simply cannot function. That matters. In many conflict-affected settings, the choice is not between perfect data and flawed data; it’s between acting early based on imperfect but credible analysis or acting too late.