AVA Proativa: Sales Assistant with Proactive Intelligence, Continuous Assessment and Secure Orchestration for CRM

• Grant number: 25/29032-4
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: September 01, 2026
• End date: May 31, 2027
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Julio Fernando Pato Paulillo
• Grantee: Julio Fernando Pato Paulillo
• Company: AGENDOR SERVICOS DE INTERNET LTDA
• CNAE: Desenvolvimento e licenciamento de programas de computador não-customizáveis
• Principal investigators: Tulio Gabriel Monte Azul
• Associated researchers: Caio César Brandini da Silva ; Guilherme Vilar Balduino

Abstract

Despite the widespread dissemination of Customer Relationship Management (CRM) systems, studies and empirical evidence indicate that a significant portion of small and medium-sized enterprises (SMEs) in Brazil have low operational adoption of these tools, resulting in incomplete data, low commercial predictability, and loss of efficiency. This problem is particularly critical in consultative B2B sales, where most interactions between salespeople and customers occur through instant messaging applications, especially WhatsApp, outside the formal CRM environment. As a consequence, strategic information remains scattered in informal conversations, hindering data-driven management.In this context, this applied research project proposes to investigate the scientific and technological feasibility of using generative artificial intelligence-based assistants to interpret real sales conversations in informal natural language and to proactively and securely support the structured recording of commercial information in CRM systems. The research starts from the recognition of relevant uncertainties not yet resolved by the literature and the state of the art, such as: (i) the ability of large language models (LLMs) to understand commercial intentions in informal Portuguese with sufficient accuracy; (ii) the limits and effects of contextual proactivity in human-AI interactions in the consultative sales domain; and (iii) the possibility of executing actions in critical corporate systems in an auditable, predictable, and regulatory compliance manner.To address these uncertainties, the project proposes the experimental development of the Proactive AVA, a sales assistant integrated with the Agendor CRM and WhatsApp, conceived as a research artifact. The methodological approach combines experimental software engineering, LLM-based AI techniques, retrieval-augmented generation (RAG) strategies, and the construction of deterministic validation mechanisms (guardrails), which separate the probabilistic decision of the model from the effective execution of actions in the CRM. Additionally, a continuous evaluation framework (AI Evals), oriented towards the sales domain, will be developed, capable of systematically measuring the system's performance through metrics such as task success rate, action validity, hallucination rate, and latency.The central objective of Phase 1 is to reduce the scientific and technological uncertainties associated with the application of LLMs in B2B consultative sales, delivering as a result a functional proof of concept (PoC), experimentally validated in a controlled environment, accompanied by quantitative evidence on the technical feasibility of the proposed approach. It is expected to demonstrate potential gains in CRM data completeness, reduction of the salesperson's operational effort, and increased system adoption, establishing a solid foundation for subsequent validation stages in a real-world environment and technological scaling.From an impact perspective, the research has the potential to contribute both to the advancement of knowledge about AI-based proactive agents in sensitive corporate contexts and to the effective digitization of commercial processes in Brazilian SMEs, aligning technological innovation, methodological rigor, and economic relevance. (AU)