AI Assistant Cutting Meeting Organization Time by 50%

Boldly is a high-end, subscription-based staffing provider with a team based in North America and Europe. The company provides highly skilled remote employees to represent businesses and execute tasks responsibly and effectively.

Boldly’s executive assistants faced a significant challenge in organizing meetings for senior management, primarily due to the need to constantly switch contexts while managing multiple meeting schedules. This process was all-consuming and labor-intensive. Assistants were required to meticulously review lengthy email correspondences in reverse order to keep up with the evolving status of each meeting. This situation highlighted the urgent need to streamline the meeting organization process, aiming to reduce administrative overhead and enhance productivity.

As the need for an automated solution became apparent, another crucial challenge emerged: data privacy and security. The email threads handled by the assistants contained sensitive information that required strict confidentiality and secure handling to prevent unauthorized access.

Accessing extensive and relevant data sets is essential for building a robust AI system. However, in this case, it was complicated by limited access to real-world data.

To save time spent on meeting organization, our team leveraged Large Language Models (LLMs) to create a sophisticated AI-based virtual assistant. This tool is capable of sifting through email content and providing immediate overviews of the current status of a meeting by condensing lengthy, multi-threaded conversations into concise updates.

Key Features:

• Smart Information Extraction: The model identifies and extracts critical details from emails, such as proposed meeting dates, locations, attendee responses, and even food preferences of participants.
• Data Aggregation and Processing: After extraction, the tool aggregates and processes the data, presenting it in a clear and concise format that is easy to understand.
• Real-Time Status Updates: Assistants receive up-to-date information on the current status of each meeting, which aids in efficient planning and organization.
• Email Thread Summaries: The solution generates bulleted histories of email threads, ensuring assistants have a comprehensive understanding of each meeting’s progression and facilitating streamlined organization.

To overcome the challenge of limited access to real-world data, DLabs.AI developed synthetic datasets that closely mimic real-world examples. This approach provided a rich and diverse set of data crucial for training our AI models effectively.

To ensure the confidentiality and security of sensitive information, all data is anonymized before processing. Furthermore, we use a secure cloud infrastructure to meet rigorous data privacy requirements.

Our solution achieved a substantial ~50% reduction in the time assistants spent on scheduling meetings compared to traditional, manual methods. This efficiency gain was confirmed through tests conducted by the client alongside real-world users.

Interestingly, the initial assumptions were only 20-30% time savings, which made our solution outperform the client’s expectations.

Within less than a working day, we set up the production infrastructure, leveraging the configuration from the development phase. Notably, the entire infrastructure is capable of scaling down to zero, thereby incurring only minimal maintenance costs.

The AI tool effectively extracts various types of information, including proposed meeting dates, food preferences of participants, meeting locations, and detects acceptances of meeting dates. This robust capability ensures that all pertinent details are captured and utilized for meeting organization.

Aggregated and processed data are clearly displayed to assistants, showing the current status of the meeting organization. Assistants also benefit from an overview of the email thread’s history, which is summarized in bullet points, facilitating a better grasp of the meeting’s progression.

We successfully mirrored real-world scenarios using a synthetic dataset, capturing most common cases. This innovative approach has allowed us to train our AI models effectively, despite the challenges of limited access to real-world data.