On the day the scanning function was installed, passengers were observed to spontaneously scan in response to prompts from temi, with over 10 scans being carried out in the morning alone. The acquired scan data was saved in real time in a format that removed personal information, allowing it to be collated at a later date with actual security checkpoint passage data. This established a foundation for quantitatively verifying the extent to which temi's intervention encouraged faster security checkpoint passage compared to the average. One of the key results of this demonstration was that it went beyond qualitative evaluation and created a verifiable data structure.

Additionally, an automatic announcement function linked to flight departure times has been implemented. 40 minutes before departure, devices such as temi and tablets will automatically announce the deadline for passing through security. The announcement is designed to be made whether temi is patrolling or waiting in its designated location, and its effectiveness in raising awareness of time throughout the space has been verified. By combining individual notifications with a system that works on the entire environment, we aim to provide multi-layered support for time management.

Furthermore, a verification of automatic contact using telephone AI was also conducted. An automatic call was made to the target person 40 minutes before the flight's departure time, and the conversational AI provided a voice notification of the deadline for passing through security. If there were any questions, the call was automatically transferred to a call staff member, and the operational flow for switching to human response was also confirmed. This verification was conducted only among those involved in the demonstration, with the aim of verifying technical feasibility and identifying operational issues. Another feature of this demonstration is that it specifically verified the possibility of integrated control of multiple channels such as robots, signage, and telephones.

In designing the KPIs, we set the security checkpoint passage rate, passage lead time, number of guide approaches and behavior induction rate, staff experience evaluation, and user satisfaction. As the implementation progressed, it became clear that realistic adjustments were needed between the ideal indicators and the data that could be obtained, and we ultimately narrowed our focus to a quantitative evaluation platform centered on scan logs. The process of redefining KGI and KPI back and forth was not only a successful implementation, but also a methodological achievement that will contribute to ongoing verification in the future.

During the trial period from January 19th to January 30th, we counted the number of passengers who stopped in front of Temi or looked at the screen to understand the level of interest shown by passengers as it made announcements while moving. As a result, a total of 940 people were confirmed to have engaged with Temi during the trial period. This is a basic indicator of whether the audio guidance was attracting passengers' attention and leading them to access visual information. By defining and counting clear interest behaviors, rather than simply passing by, we aimed to quantitatively understand the effectiveness of Temi's cognitive guidance.

Between February 16th and 25th, we measured the number of boarding pass QR code scans. The result was 67 scans during that period. This is an indicator of the number of times temi's prompts and countdown display functions led to specific passenger actions. The acquired scan logs were saved in a structure that can be compared with subsequent security checkpoint passing data, and will serve as basic data for verifying the extent to which time awareness support contributed to actual behavioral change.

However, several issues emerged during the implementation process, including the stability of self-location estimation in a vast, busy lobby space, optimization when standard functions and extended apps run in parallel, handling complex operational data structures such as code-share flights, and a lack of a remote monitoring and notification infrastructure. It also became clear that the costs associated with establishing operational rules, fostering managerial understanding, and coordinating with other parties involved in the introduction of autonomously moving robots were not negligible factors. It was confirmed that a phased implementation process and the creation of a system that could be operated by the field itself were essential.

Overall, this demonstration demonstrated the feasibility of implementing a call-to-action model that integrates robots, AI, and IoT, toward a future vision of "personal airport guidance that prevents confusion." By combining four elements - time-aware assistance, personalized notifications, multi-channel collaboration, and behavioral data acquisition - this marks the first step toward a smart hospitality airport model that balances passenger experience with operational efficiency.

Above