The Problem

Every shipment generates commercial invoices, packing lists, certificates of origin, transport documents and customs declarations. Behind each document sits a set of compliance determinations: tariff classification, origin, valuation, preference eligibility and, increasingly, exposure to forced labour, sanctions, and environmental regulations.

As it is difficult to review this volume of data manually, the standard approach has been to audit samples. A sampling approach involves periodic review of a subset of declarations, from which broader conclusions are drawn. It is slow, resource-intensive and blind to whatever falls outside the sample, so the greatest risks and opportunities can be missed entirely. Undetected misclassification or incorrect origin exposes companies to penalties, retroactive duty assessments, and reputational damage where exposure of forced labour or sanctions goes unnoticed in upstream supply chains. Importers also fail to claim preferences available under trade agreements because no one has the capacity to check eligibility across every line. 

The Solution

Maersk is a global shipping line and logistics integrator spanning maritime, air and inland transport, warehousing, terminals, customs brokerage and supply chain management. Its trade and customs consulting unit advises customers on trade strategy and compliance. Since 2022, the unit has embedded AI into a portfolio of tools and services with customs being one of the first use cases, and it expects AI to be core to their operations in the future.

The portfolio addresses different stages of the compliance workflow. Machine-reading tools extract data from shipping and commercial documents to create customs declarations. Analytical tools scan bulk declaration data at a scale that sampling could not reach, flagging inconsistencies, potential misclassification, incorrect origin and unclaimed preferences. Classification tools generate proposed Harmonized System (HS) codes. Supply chain mapping tools trace goods upstream to identify forced labour, money laundering and sanctions risk; related tools support the calculation of emissions for reporting under the European Union's (EU's) Carbon Border Adjustment Mechanism and the assessment of exposure to the EU Deforestation Regulation (EUDR). Products covered by the EUDR include coffee, cocoa, palm oil, soy, timber, or products derived from them, and importers need to be able to prove that any products it imports to the EU are deforestation-free. The company is also developing tools that allow customers to interrogate their own customs data using natural language.

Human oversight is integrated into these processes. AI outputs follow one of two paths: they are either reviewed and approved by trade experts before delivery to clients (every AI-generated classification, for instance, is verified by a specialist), or they are presented explicitly as risk-prioritization rather than definitive answers, so that client expertise can be directed to the areas identified.

Tools are acquired through a mixed model of specialist third-party providers, in-house development and generic productivity tools, with all processing managed within Maersk's own data security environment. Deployment follows a "super user" model in which some dedicated users, particularly experts in the subject matter, have early access, training and advanced capacity to use the AI tools. Their knowledge and experience then cascade across the organization when more users are onboarded.

Overview

Challenges

Maersk manages these constraints through a governance framework combining security controls, quality assurance of input data, and a human-in-the-loop review policy under which outputs are checked before they reach clients.

The impact

Glossary