APC 技術ブログ


株式会社 エーピーコミュニケーションズの技術ブログです。

Enhancing Audit Efficiency at Hapag-Lloyd Through the Use of Generative AI


Session speaker is Michael Schermer, a Product Specialist from Databricks. He delved into the transformative applications of generative AI in enhancing the audit processes at Hapag-Lloyd. This session is also joined by Uri Danijel, Corporate Audit Director at Hapag-Lloyd, and Anja Sinikova, a Senior Specialist at Databricks. We are actively involved in deploying an advanced AI tool named 'Implementer,' aimed at automating routine tasks and facilitating rapid data retrieval and analysis to reduce operational costs and improve overall efficacy.

Based on the information above, the session content was revised as outlined in the pre-created article titled "Enhancing Audit Efficiency at Hapag-Lloyd with Generative AI" and the abstract provided in "Session Overview." The structure followed the enumerated list in "Section Theme List," and the sections targeted for revision were specified under "Target Section Themes." The transcription used for the article creation was provided under "Target Section Bodies," and changes were made according to the "Target Section Articles."

The key concept enhancing audit efficiency at Hapag-Lloyd: Standardization and Assessment of Models

In the realms of generative AI solutions implemented at Hapag-Lloyd, standardizing and evaluating models are fundamental. This section discusses methodologies adopted to ensure optimal outcomes for improving audit efficiency.

Clarifying Objectives and Conducting Initial Tests

Initially identifying clear objectives is crucial. At Hapag-Lloyd, specific goals based on audit challenges are set to ensure the project progresses effectively and focused. This includes formulating hypotheses and objectives that guide the development process. Following this, initial tests to validate these hypotheses are conducted, providing a roadmap for further model development.

Importance and Methods of Model Evaluation

A robust evaluation framework is essential to assess the effectiveness of the model. This not only involves using standard quantitative metrics to measure performance but also employing custom datasets to reflect the unique needs of corporate auditing. These datasets help understand how the model performs under specific conditions unique to Hapag-Lloyd’s requirements.

Applying Comparative Analysis and Evaluation Insights

Insights from model evaluation are crucial. They provide information for decision-making on model optimization, aiding in refining AI tools until they meet the set benchmarks. Furthermore, by comparing different models, the team can select the most efficient model suited for Hapag-Lloyd’s operational environment. Decisions are backed by data from target evaluations, ensuring the selected model performs optimally in actual audit scenarios.

This structured approach not only enhances the reliability of AI solutions but also customizes them to address specific challenges faced during audits at Hapag-Lloyd. Through careful testing, evaluation, and comparison of models, the project moves closer to significantly improving audit efficiency.

Section: Integration and Implementation

At Hapag-Lloyd, the focus is on enhancing audit efficiency through the exploitation of cutting-edge technology on the Databricks data intelligence platform, particularly focusing on enhanced information retrieval and scalable model delivery. The methodologies used to review and assess generated audit findings within the audit process are highly strategic.

Detailed Approach:

  1. Review of Generated Findings: Auditors thoroughly scrutinize each generated finding, discarding any that are superfluous. This action is crucial to ensure the relevance and accuracy of audit results.

  2. Utilization of Datasets: The assessment of these findings is conducted using specific datasets, employing methods such as multiple coding to enhance the accuracy of evaluation.

  3. Adoption of Various Evaluation Techniques: A dual-approach is utilized to expedite the evaluation process, encompassing both small-scale functionalities for initial evaluation and extensive judgment capabilities for detailed assessments. Auditors evaluate the accuracy of the findings' presentation and conduct test calls based on these judgments to verify results.

  4. Ensuring Accuracy in HTML Sections: In HTML documents, rigorous measures are taken to ensure all information is correctly captured and reflected, guaranteeing accurate reporting of test results.

These integrative actions have significantly improved audit efficiency at Hapag-Lloyd. AI and data-driven strategies not only ensure the accuracy of audit results but also shorten the time required for completion. These efforts, leveraging AI and data intelligence through concrete evaluation and application of datasets, hold significant potential to transform the future of audits at Hapag-Lloyd.

Section: User Interaction and Feedback

At Hapag-Lloyd, the user interaction and feedback process are crucial in refining the audit functions. Initially, small models handle inquiries, often requiring manual adjustments based on user input. This feedback aids in creating more effective prompts submitted to the base model, establishing a unified and comprehensive approach.

A two-tiered reasoning table is strategically used, ensuring accurate reflection of input data within the system. This two-layer inquiry mechanism significantly enhances the transparency and functionality of the entire user interface structure. Within the user interface, users can directly enter inquiries into newly integrated tabs, enabling real-time visibility of results.

To further enhance the user experience, several results are displayed for reference. Moreover, users possess the unique ability to navigate through documents presented in multiple languages via a designated path known as the 'Cool Path.' This interactive approach fosters a dynamic feedback loop, streamlining the audit process and propelling Hapag-Lloyd toward more advanced and efficient audit methods.

Direct user interactions facilitate the system's continuous evolution, reducing technical barriers and creating an intuitive, user-friendly platform. By incorporating proactive user feedback, Hapag-Lloyd constantly improves the system's capabilities, aiming to maximize audit efficiency and precision across operations.

Section Themes: Metrics, Monitoring, Data Integration

To enhance audit efficiency, Hapag-Lloyd has adopted the Databricks data intelligence platform, integrating advanced technologies such as Retrieval-Augmented Generation for scalable model delivery. This section explores the effectiveness of these technologies in the audit process and focuses on the metrics used to assess their impact.

Particularly, the session emphasized metrics revolving around the ease of information access. This significantly reduces the time logistics staff invests in data retrieval, simplifying the process and directly contributing to efficiency improvements. Ease of access not only speeds up the process but also makes the system intuitively useful for various user personas. Plans to add more indexed personalization were discussed, promising customized data interaction experiences in future implementations.

Data security, while a critical aspect of any data integration and monitoring setup, was not extensively covered in the current measures. However, discussions were held about ongoing plans to enhance the data protection framework and effectively prevent breaches, forecasting improvements that would offer more robust security measures in the future.


In conclusion, the integration of Generative AI into Hapag-Lloyd's audit process is clearly reshaping their audit framework. With continuous improvements in monitoring and data management anticipated, the future looks promising for effectively using AI to streamline operations and maintain data integrity, creating a more resilient and efficient audit system.

The continuous evolution of AI and data integration strategies at Hapag-Lloyd is essential to advancing audit efficiency. Effectively leveraging these technologies not only simplifies the audit process but also ensures high levels of data security and customization to meet the diverse needs of various users. Moving forward, deeper integration and more sophisticated monitoring are expected, enhancing these systems and paving the way for proficient, cutting-edge audit mechanisms.

About the special site during DAIS

This year, we have prepared a special site to report on the session contents and the situation from the DAIS site! We plan to update the blog every day during DAIS, so please take a look.