In internal meetings, the workload extends beyond the discussion itself to the post-meeting organization of information. Participants need to review the meeting content, summarize the key points, identify task owners, and, when necessary, send follow-up emails or create tasks for relevant stakeholders.

This series of tasks is time-consuming, and as meetings become larger or cover more topics, the risk of missed follow-ups and inconsistent understanding increases. In particular, if task owners and due dates are not clearly documented, subsequent progress tracking can be affected. A solution was therefore needed to reduce post-meeting administrative work while improving the clarity and traceability of action items.

The objective of this PoC was to reduce the effort required for post-meeting summarization and action item organization, making it easier to turn meeting discussions into clear and actionable next steps.

The goal was not simply to shorten meeting transcripts, but to identify key discussion points, extract action items with assigned owners and due dates, and present the results in a format that is easy to share. In addition, the PoC aimed to evaluate the practical level of AI assistance in real business workflows, assuming that AI-generated outputs would be reviewed by human users before being used.

SMILE was responsible for the design and implementation of an AI Assistant PoC to support post-meeting workflows. The primary scope of work included the following:

• Prompt engineering and rule design for meeting content analysis and summarization
• Implementation of a speech-to-text pipeline through Speech-to-Text integration
• Implementation of an LLM-based meeting summarization feature
• Design of the extraction logic for action items, assignees, and due dates
• Output design for email draft generation and task system integration
• Implementation of a review workflow that enables users to easily perform final verification

A key principle of this PoC was that AI-generated outputs were not intended to be treated as final information. Instead, the solution was designed as an assistive tool, with the assumption that meeting participants would review and validate the results before use.

This PoC focused not only on automatically summarizing meeting content but also on presenting the results in a format that is immediately useful for day-to-day operations. After a meeting, what organizations truly need is not a lengthy summary, but a clear understanding of what was decided, who is responsible for each action item, and when it should be completed.

To achieve this, the output was designed around a workflow consisting of key point summarization, action item extraction, assignee and due date identification, and draft generation for sharing. As a result, users no longer need to create meeting minutes from scratch; instead, they can simply review and refine the AI-generated draft, significantly reducing post-meeting administrative effort.

In addition, meeting discussions often contain ambiguous expressions and context-dependent statements, making it important to prevent the AI from making overly autonomous judgments. Therefore, the solution was designed on the assumption that project managers or meeting participants would make the final decisions. By combining AI-generated outputs with a user-friendly review interface and output format, the PoC aimed to strike a practical balance between automation and human validation.

One of the biggest challenges was consistently extracting action items that were meaningful in real business workflows from meeting discussions. Conversations often contain abbreviated expressions and ambiguous wording, and the order of speakers alone is not always sufficient to clearly identify task owners or due dates.

To address this challenge, SMILE clearly separated the responsibilities of meeting summarization and action item extraction, while refining prompts and output rules to present task details, assignees, and due dates in a structured and easy-to-use format. The outputs were also designed at an appropriate level of granularity to support downstream processes, including email draft generation and integration with task management systems, making them readily usable in subsequent business workflows.

Furthermore, to ensure that AI-generated outputs were not treated as final decisions, the solution was designed around a human-centered review workflow. The PoC enables participants to review, edit, and validate AI-generated results before they are used, ensuring both the accuracy of the outputs and their practical applicability in real-world operations.

This PoC enabled the automatic generation of draft meeting summaries, action items, assignees, and due dates immediately after each meeting. As a result, the time required for post-meeting administrative work was significantly reduced, while also helping minimize missed follow-up tasks and improve the clarity of individual responsibilities during execution.

In this PoC, SMILE adopted a collaborative development model between the Japan and Vietnam teams. The Japan side defined the desired operational workflow and clarified the challenges in post-meeting tasks, while the Vietnam development team implemented the solution based on the agreed requirements. Acting as a bridge, the BrSE continuously aligned business intent, expected outputs, and evaluation criteria for AI-generated results, minimizing differences in interpretation and ensuring the overall quality of the PoC.

In addition, specifications and review criteria were shared between the Japan and Vietnam teams. During testing, the team evaluated not only the quality of the meeting summaries but also the clarity of action items, the representation of assignees and due dates, and the ease of reviewing the generated outputs. By maintaining short feedback and validation cycles, the team established a process that allowed the PoC to be evaluated under conditions that closely resembled real-world business operations.

Creating meeting summaries and organizing follow-up tasks are often overlooked yet time-consuming parts of everyday business operations. In this PoC, SMILE validated an approach that leverages AI to reduce this workload while structuring the outputs in a way that is practical for real-world use, assuming that users perform the final review and validation before the results are applied.

Designing workflows that transform meeting discussions into actionable next steps is a common challenge for many organizations. Rather than focusing solely on meeting minute generation, this PoC rethinks the entire post-meeting workflow, including action item management and information sharing. It serves as a practical reference for organizations looking to leverage AI to improve the efficiency of post-meeting operations.