White Papers AI Autonomy Driving a Practice-Oriented Logistics Paradigm Shift

Definition of Key Concepts

Digital Logistics Transition and Concept of Operation Phases

1^st Wave (Digital Logistics)

It refers to the first wave of digital transformation based on a platform. Instead of scattered Excel files and emails, it processes quotation, documentation, and tracking tasks on an integrated platform, greatly enhancing visibility through data centralization and integration of business processes. However, there remains a limitation in that people are still the ones entering information into the system and handling exceptions.

2^nd Wave (Intelligent Logistics, AI-Powered Logistics)

Large Language Models (LLMs) and AI agents directly understand the context and process the work of operators. Beyond simple monitoring, the core is for AI to read and datafy unstructured data (emails, messengers, etc.) to support practical “hyper-automation”.

Datafication

It is the process of converting unstructured data, such as special requests or exceptional cost agreements exchanged via email that were not captured in the system, into structured data through AI analysis. This allows AI to automatically reflect information in the system that previously required manual verification by operators.

Hidden Risk

Risks not evident from the rates on the tariff table, relying on information based on the experience of seasoned operators, such as the tendency for transshipment delays by certain shipping lines or seasonal weather issues. The AI agent replaces this with actual data analysis, such as ETA compliance rates and delay histories by route.

AI Agent-based Workflow Automation Concept

AI Agent

An intelligent system that directly understands the context of workflow of operators and assists or substitutes decision-making. It is applied across all key stages of international transportation, including quotation analysis, booking automation, visibility enhancement, risk sensing, and cost settlement.

Booking Agent

Operates 24/7 and reads shipping documents (B/L, invoice, packing list, etc.) received by email using natural language processing to identify which purchase order (PO) they correspond to and automatically registers them in the system. If there is a discrepancy between the documents, it cross-checks with previous master data and notifies the relevant staff.

Predictive ETA/ETD

It is the estimated arrival and departure date, which is not based solely on past schedules, but is re-predicted in real time by learning hundreds of thousands of track data through machine learning. If tracking information is missing, the AI completes an unmanned control loop by automatically filling in the blanks by crawling the carriers’ website or contacting partner companies directly.

Item-level Business Impact Analysis

It is a method that analyzes and reports in detail even "which PO's which part is delayed" by sensing global news in real-time, filtering risks affecting the company and combining those with documents secured at the booking stage.

Knowledge Graph / Context Layer

Designed to supplement the “Why” behind decision-making that the system cannot record. By learning the exception handling methods and decision history (Decision Memory) of operators scattered across messengers, meeting minutes, and emails, AI enables context-based rational judgment at a human level.

Intelligent Logistics Innovation (2^nd Wave), Exploring Through Key Questions

• Q1.

  Digital transformation (1^st Wave) has sufficiently addressed logistics tasks?

  Platform-based digital logistics have greatly enhanced visibility through data centralization and task integration. However, unstructured data such as special requests or agreements on exceptional costs exchanged via email still do not get captured in the system, and the responsibility of verifying and handling these tasks remains with humans.

• Q2.

  What is different about AI agents compared to existing digital platforms?

  The existing platform was limited to accumulating and displaying structured data. The AI agent goes a step further from here, directly reading and datafying unstructured data, understanding the practical context of work, and assisting in decision-making to support “hyperautomation”.

• Q3.

  What value does AI provide in the quoting and sourcing process?

  AI replaces ”hidden risks” such as tendency of delays by carrier or seasonal risks, which experienced staff used to consider intuitively, with real data analysis. It helps operators choose the optimal routing to meet their objectives by comprehensively considering ETA compliance rates, additional costs, and carbon emissions.

• Q4.

  How are booking and document processing tasks automated?

  The AI-based booking agent operates 24/7, instantly registering booking emails received even at night or on weekends into the system by understanding the context through natural language processing. If there is a discrepancy in the document information, it cross-checks with past data and requests verification from the staff, so the role of manager shifts from simple typing to a role focused on approving exceptional cases.

• Q5.

  How does AI respond when cargo location information is missing?

  When the AI agent detects that essential information is missing from the system, it automatically fills in the blanks by crawling carriers’ websites or sending emails/messages directly to partners to confirm responses. In addition, it calculates machine learning-based predictive ETA/ETD to provide more accurate arrival and departure schedules.

• Q6.

  How does AI respond differently to global risks compared to traditional methods?

  Previously, when issues such as canal blockades or port strikes occurred, we had to manually filter through Excel to check the impact on our cargo. AI senses 60,000 pieces of global news daily to refine only the risks with actual impact, and combines this with booking-stage document information to analyze and inform us at the item level which order (PO) and which part will be delayed.

• Q7.

  Can AI replace 100% of all logistics operations?

  It is difficult in the short term. This is because the existing system only records the results (What) but cannot capture the reason for the decision (Why). Until knowledge graphs and context layers learn this contextual information, operators must take on the role of decision-makers to control exceptional situations and plan alternative routes.