From pilots to institutional decisions

They identified a four-stage model for integrating generative AI into strategic decision-making, with enablers such as leadership and secure platforms and barriers such as strategic ambiguity and the risk of hallucinations. The model includes admission gates that demand quality, traceability, explainability and accountability, and shows that adoption moves from pilots to institutionalization once governance and human-AI roles are established.

Amid the avalanche of promises around generative AI, strategy directors keep receiving proposals for pilots that never reach the company's real agenda. The pressure to show agility and analytical capacity collides with the need for consistent, traceable decisions, and the gap between experimenting and institutionalizing becomes an operational risk.

To understand how that gap is closed, the authors carried out a single case study in a global multi-brand conglomerate, combining 27 semi-structured interviews with executives, internal documents and industry reports. The structured inductive analysis produced a process model that captures both the drivers (active leadership, secure platforms, dedicated investment and staff training) and the obstacles, which include strategic ambiguity, low awareness of hallucination risks and a lack of maturity in prompt engineering.

The resulting model describes four sequential stages: awareness and exploration, experimentation and pilots, formal adoption and integration, and institutionalization and transformation. Each phase is guarded by admission gates that demand quality, traceability, explainability and accountability, and it redefines the human-AI relationship through "human-in-the-loop" schemes that redistribute responsibilities. The move from pilot to routine decision happens when clear governance and specific roles are established that legitimize the entry of AI outputs into deliberative processes.

Leaders should turn that framework into concrete actions: define admission criteria before any deployment, create governance committees that oversee the quality and ethics of the results, invest in infrastructure that guarantees data security and train teams to master prompt formulation. Only by aligning the technology with decision structures and accountability does generative AI go from being an experiment to a reliable strategic engine.

Conclusions

• Generative AI should be seen as support within decision systems, not as an autonomous oracle.
• Scaling its use requires clear governance that includes criteria for quality, provenance and explainability.
• Active leadership and investment in secure platforms and training are critical to move from pilots to institutional adoption.
• The risks of hallucinations and the lack of data readiness must be mitigated before integrating the results into formal deliberations.