Maersk has harnessed AI to accelerate its compliance with customs and border control with impressive results, writes Namrata Nadkarni

For as long as global trade has existed, customs compliance has run on samples. A team pulls a subset of declarations, checks them, and draws conclusions about the rest. It’s how you cope when every shipment produces a stack of commercial invoices, packing lists, certificates of origin and transport documents. 

Each document contains the final result of judgement calls on tariff classification, origin, valuation, preference eligibility and, increasingly, exposure to forced labour and sanctions further up the supply chain. Reviewing that volume manually, line by line, has simply never been possible. So businesses sample, and hope the biggest risks (and the opportunities they’re leaving on the table) aren’t sitting in the part of the pile nobody checked.

That’s the problem that a recent case study published by the World Trade Organization sets out, and it’s one I think anyone working in maritime trade will recognise immediately. The case study, based on Maersk’s own written responses to the WTO, looks at how the trade and customs consulting arm of Maersk has spent the past few years embedding AI into that exact workflow.

Embedding governance

The case study is a genuinely useful read, not because it promises AI will replace human judgement at the border, but because of how deliberately it doesn’t. Since 2022, Maersk’s consulting unit has built a portfolio of tools for different stages of the compliance chain: machine-reading tools that extract data from shipping documents to build declarations; analytics tools that scan bulk data at a scale sampling could never reach, flagging inconsistencies and unclaimed preferences. 

They have created classification tools that propose Harmonised System codes; and supply chain mapping tools that trace goods upstream for forced labour, money laundering and sanctions risk, alongside emissions calculations for the EU’s Carbon Border Adjustment Mechanism and exposure assessments under the EU Deforestation Regulation.

What struck me most, reading it, is the governance underneath all of that. Every AI-generated classification is checked by a specialist before it reaches a client. Where outputs aren’t verified line by line, they’re presented explicitly as risk-prioritisation – a pointer for where human expertise should look next, not a verdict. 

Maersk calls its rollout approach a ‘super user’ model: subject-matter experts get early access and training, then their experience cascades through the organisation as more people come on board. It’s a good template for anyone nervous about handing judgement calls to a model. In this set up, AI is doing the reading, not the deciding, which is exactly the ‘copilot, not captain’ framing that came up repeatedly in our AI at Sea roundtable.

Accelerated assessments

Financial, environmental and philosophical issues aside, the results of the AI integration are hard to argue with. In one situation cited, a customer previously needed roughly 100 person-days to map a supply chain and assess its forced labour exposure, by which point, as the case study notes, the answer was often already out of date. Today, that customer can get the equivalent analysis near instantaneously, with continuous automated mapping keeping the picture current. 

Maersk’s pilot ‘Passport for Goods’ model, which collects compliant data across the supply chain so customers and border agencies can see goods coming weeks before they arrive, has produced 60% better predictability and resilience, 40% faster border crossings and 40% lower costs in testing across multiple markets. The name caught my eye immediately – it’s hard not to think of the digital identity work happening elsewhere in shipping, including a conversation I had with Maritime Passport’s co-founders on trust as the real bottleneck to adoption. The underlying idea is the same: give the right data to the right people earlier, and friction disappears.

As Maersk puts it in the case study: “AI enables us to transform the way the border is managed. We can leverage first-hand data from upstream in the supply chain to remove any friction at the border itself.”

Culture vs technology

The information shared by Maersk is a good example of something I keep coming back to in this project and across various articles: the barrier to AI adoption in trade and shipping is rarely the technology itself. It’s data quality, governance and organisational behaviour, the exact territory my colleague Robert Platt explored in his article “Maritime’s digital bottleneck is behaviour, not code”.

Maersk’s case study makes exactly this point: AI has made large-scale analytics possible, but it hasn’t removed the dependency on sound, structured, accessible source data. The technology only works as well as the discipline behind it.

For an industry that still talks about AI in fairly abstract terms, this is a useful, concrete example of what ‘AI in trade compliance’” actually looks like in practice – and a reminder that the winners won’t be the companies with the flashiest tools, but the ones who pair them with the right human oversight.

This article is based on a case study published by the World Trade Organization, that was put together based on Maersk’s written interview responses to the WTO. All facts, figures and the quoted statement are sourced from that case study and remain the work of the WTO and Maersk.

Photo: Maersk. https://www.maersk.com/supply-chain-logistics/customs-services

Like this article? Click here to sign up for our free monthly ‘De-Mystifying Digitalisation and AI in Maritime’ newsletter and get the latest cross-sector expert insights, news, interviews, videos and case studies straight to your inbox. You can also buy a ticket to the Demystifying AI in Maritime 2026 London Summit here.