Consulting firms such as Accenture face growing pressure to deliver complex digital projects more efficiently, and at the same time Agile ways of working and composable commerce architectures increase coordination complexity and place significant pressure on BAs. In this context, Agentic AI offers potential as a more proactive and embedded form of workflow support. This thesis explores how an Agentic AI-driven project management tool might support Accenture Song BAs in increasing their work efficiency during the Design & Build phases of Agile composable commerce delivery projects.

Using the Double Diamond framework, the study applies a mixed-method approach combining literature review, unstructured and semi-structured interviews, observations, survey data, and internal documentation analysis. This research shows that BA work in this context is characterised by fragmented information, high coordination complexity, frequent replanning, and a strong dependence on system understanding and cross-team alignment. From the identified improvement opportunities, dependency management emerged as the most relevant focus area, as it was both highly desired by BAs and highly consequential for delivery progress.

To address this opportunity, two concept directions were explored and prototyped, after which Risko was selected as the primary concept. Risko is an Agentic AI-driven prototype that supports BAs in maintaining dependency and future sprint planning-status awareness by analysing project data, surfacing potential risks and opportunities, and guiding more targeted check-ins and planning adjustments. Built in n8n using mock project data and stand-ins for tools such as Jira, Confluence, and Excel, the prototype demonstrates how Agentic AI capabilities such as data retrieval, tool use, and text generation can be translated into concrete workflow support while retaining human-in-the-loop control.

The project concludes that Agentic AI has clear potential to support BAs not by replacing their work, but by augmenting it through more proactive, context-aware assistance. At the same time, the thesis shows that such support must be carefully designed around context provision, low hallucination risk, deterministic logic where appropriate, and seamless integration into existing workflows. Although Risko appears promising in terms of desirability, feasibility, and viability, further validation is still needed before it can be considered a sufficiently validated solution for Accenture.