Small and medium enterprises (SMEs) are starting to use Multi-Agent Generative Systems, combining autonomous agents with perception, decision, and action capacities and large language models (LLMs), but their integration into strategic processes remains limited and faces significant challenges. Intense competition, market uncertainty, cognitive biases, difficulties in long-term planning, resource constraints, and a limited data-driven mindset make it difficult for SMEs to innovate and remain resilient, while skepticism and sensationalism around Generative AI further complicate adoption. To address these challenges, this paper proposes an adaptive, process-centric, and hypothesis-driven Evaluative AI framework for accessible innovation and resilience in resource-limited entrepreneurial contexts. The framework introduces a modular, multi-agent architecture where specialized evaluative agents find contextualized evidence for or against business hypotheses under continuous refinement, using open-source LLMs grounded in internal company data and trustworthy public sources, including Companies House APIs, to test patterns and profitability of similar firms. Agents operate autonomously under a central orchestrator, enabling plug-and-play integration, shared memory, and contextualization through business configuration and vector storage. Iterations are weighted by business-configured variables, such as team development, risk, or budget, while a reasoning agent ensures consistency across scenarios, allowing the decision maker to adjust accordingly. The framework also supports systematic business impact assessment, including robustness, reliability, coverage, and usefulness metrics, ensuring outputs are interpretable, verifiable, and actionable. This design lets SMEs flexibly manage their agent team while adapting to evolving strategic demands. The framework’s effectiveness will be evaluated through a real-world industrial case study in Cliqpod, a digital marketing SME.