Research & publications

Research on strategic reasoning, heuristics and decision support.

The Institute’s research programme explores how structured analysis, heuristic knowledge and AI-assisted interpretation can support strategic choice without removing human judgment from the process.

Recommending Actionable Strategies: A Semantic Approach to Integrating Analytical Frameworks with Decision Heuristics

Renato Ghisellini · Remo Pareschi · Marco Pedroni · Giovanni Battista Raggi

Ghisellini, R.; Pareschi, R.; Pedroni, M.; Raggi, G.B. Information 2025, 16, 192. Published 1 March 2025. DOI: 10.3390/info16030192.

The article presents a semantic approach for connecting structured analytical frameworks with decision heuristics. It uses the 6C framework as a principal test case, vector-space representations and semantic similarity for framework–heuristic matching, and constrained LLM use for explanation and reporting.

Publisher / DOIPDF

From Extraction to Synthesis: Entangled Heuristics for Agent-Augmented Strategic Reasoning

Renato Ghisellini · Remo Pareschi · Marco Pedroni · Giovanni Battista Raggi

arXiv:2507.13768v1 [cs.AI], 18 July 2025.

This preprint extends heuristic selection toward compositional synthesis. It studies how multiple, potentially conflicting heuristics can interact through semantic interdependence and be synthesized into coherent strategic narratives, with a Meta vs. FTC case study and preliminary semantic evaluation.

arXiv recordPDF

Research agenda

Questions that remain open.

Strategic-choice validity

How closely do model-generated rankings and cross-readings align with expert strategic judgment across different domains?

Robustness and sensitivity

How stable are results when evidence, contextual estimates or descriptor structures change within plausible ranges?

Heuristic composition

When does synthesizing several strategic logics create insight that is genuinely different from simply listing highly ranked recommendations?

Human-AI division of labour

Which stages benefit from AI assistance, and which should remain explicitly algorithmic or human-controlled for traceability and trust?