For a Singapore bank’s event, we built a Copilot Agent for automated testing of prompts.
Overview
A leading bank based in Singapore wanted to run an innovation event where people could try prompt writing and see how well their prompts performed in real time using a live Copilot setup.
The bank needed a smooth, quick, and reliable implementation. Our team stepped in to design a complete Copilot implementation using Microsoft 365 tools and Copilot Agent Flow. They needed a simple implementation that combined automation, quick scoring, and centralized reporting to support their internal teams and event visitors.
Client Background
The client was one of Asia’s leading financial institutions, known for its focus on digital and customer-first services. For its innovation event, the bank joined hands with us to show how prompt quality can impact real AI results. They wanted an experience that would guide attendees, give them real results instantly, and help the bank collect clean insights without any manual effort.
Challenges
- Slow and manual scoring: Event teams struggled to review prompts quickly as each assessment took time and impacted the quality of engagement during live sessions.
- Scattered event data: Prompt submissions, user details, and accuracy results were stored across multiple locations. Hence, it was difficult for teams to analyze patterns, compare inputs, or create a complete view after the event.
- Uneven participant experience: Attendees wanted instant scoring, but manual processes could not support it.
- Limited visibility for event staff: Teams had no clear way to monitor submission volume or prompt quality. Without a live dashboard, they could not see trends or understand how users interacted with the activity.
Solutions
- Automated Copilot Agent Flow: The entire scoring process was powered by a structured Copilot Agent Flow, hence, the manual scoring was removed and gave consistent results for every participant.
- QR-based submission process: Implemented a QR code that captured basic details and the user’s prompt, then stored everything in Excel automatically which reduced turn over time and provided event teams accurate data without any manual handling.
- Instant scoring and email delivery: New entry and triggers were automated after form submission through Power Automate. Each participant received a personalized score with improvement suggestions within moments.
- Centralized SharePoint and Excel logging: The data was stored in a connected Excel file and SharePoint list. It also created a clean audit trail for future use.
Tech Stack
- Microsoft 365 Copilot Agent Flow
- Power Automate
- SharePoint Online
- Excel Online
- Power BI
- Outlook Connector
Business Values
- Speed and efficiency: Processing time dropped from around 5 minutes to 10–15 seconds, and manual effort was reduced by over 90%.
- Scalability: The system can manage more than 1000 submissions with around a 99% success rate, running each flow in seconds while supporting multiple users at the same time.
- AI Analysis Quality: Prompts averaged a score of 82, with strong performance in intent and goal setting.
- Reusable Setup: The same model can support future events or internal teams with minimal changes.
Future Scope
This setup gives the bank a strong base to build richer AI-driven experiences. The project opened doors for broader AI services adoption across the bank. The next steps include adding real-time Power BI dashboards so teams can see live activity and trends during events. With these additions, the bank plans to support more events, internal programs, and customer interactions, making AI feedback a natural part of daily operations across the bank.