Non-governmental organisations (NGOs) increasingly operate in environments where regulatory, financial, administrative, and reputational pressures shift fast and unpredictably. The sector’s standard risk tools (registers, likelihood-impact matrices, key risk indicators) might help to analyse existing risks, but they do not expose the causal chain from an early signal to its operational consequence, and they do not stay updated without maintenance that under-resourced NGO staff simply cannot provide. We tackle both problems. First, we adapt the bow-tie methodology from safety-critical industries to the NGO context, anchored on a single top event: the loss of operational viability in a country, with five threat categories and explicit barriers. Second, we instantiate an LLM-assisted maintenance system that classifies incoming signals, updates a versioned, evidence-grounded knowledge base, and flags barriers for human verification, while leaving consequential decisions to people. We illustrate the framework with a worked example from public reports (CIVICUS, ICNL), showing structured, transparent decision support as a concrete form of positive impact for NLP applied to civic resilience.

Hi, I’m Gisela, GLS’s assistant. How can I help you today? This is how the Spanish deployment of the GLS Chatbot greets its customers. Behind the greeting is a governed multi-agent system for parcel-logistics customer service, built on self-hosted open-source small language models with RAG. An orchestrator separates routing, execution, and generation; probabilistic components are bracketed by deterministic safeguards, customer-impacting actions require explicit user confirmation, and out-of-scope, abusive, or manipulative inputs are refused. In production since November 2025, the system has handled over 1.7 million conversations across Spain and Germany. In Spain, the average realised cost during May 2026 was approximately 0.07€ per conversation, with sub-3-second latency. We describe the architecture, evaluate the Spanish deployment on task completion, routing conformance, and RAG answer quality, and report lessons from production.