Doxis Blog  IDP & AI

AI for customs brokers: Why you need a human in the loop

| Casper Luchies

Two men discussing customs declaration with a tablet, showcasing human involvement in customs processes.

Customs brokers spend a large part of their day retyping the same shipment data. One load can need several declarations, each drawn from the same commercial invoice and packing list. Every line typed by hand is a chance for an error that stays hidden until Revenue queries the declaration.

That workload keeps growing. According to Revenue (2026), it processed over 60 million import declarations in 2025, against 1 million in 2020.

AIS V2 is adding to that load right now. Revenue's AIS V2 guidance states that AIS V1 can no longer be used to process import declarations from 27 January 2027. The new version gives each declaration a new structure with renamed data elements, so brokers are relearning where every value goes.

More declarations in an unfamiliar format mean more manual keying, and that is where AI for customs brokers comes in. AI cannot replace customs brokers. What AI can do is free up their time to increase document handling speed and accuracy.

In this article, we'll show you exactly why customs brokers will not be replaced by AI, and why keeping a human in the loop gets you the best of both.

Key Takeaways

  • AI reads commercial invoices and packing lists and fills in the declaration data; the broker keeps classification, valuation and origin decisions
  • Most declaration errors start with re-keying the same shipment data by hand, once for every declaration a load needs
  • The strongest setup combines both: AI captures the data and the broker checks each field next to the original document before anything goes further
  • Your team sets the rules that decide which documents go to review, such as a missing origin or a value above a set amount
  • Brokers spend their time on the shipments that need judgment, which is the part of the job customers pay for

What is AI for customs brokers?

AI for customs brokers is software that reads customs documents and captures the data a declaration needs. It recognizes values such as commodity codes, country of origin, weights and quantities on commercial invoices and packing lists. The broker then starts from pre-filled fields instead of typing every value from a scanned PDF.

Why AI won't replace customs brokers

AI will not replace customs brokers because the decisions that carry legal weight sit with the brokers. AI is good at reading documents and capturing data quickly. Deciding what that data means for a declaration, and answering for it, still takes a person.

What a customs broker decides on a declaration

Suppliers often print a code that is too general, outdated or simply wrong for the goods in the box. Spotting that takes knowledge of the products, the tariff and the rules of origin, which a broker builds up over years of handling shipments.

Why liability can't move to software

The responsibility is written into the declaration itself. Revenue requires every representative to state on the import declaration whether they act as a direct or indirect representative. Under indirect representation, the broker shares liability for the customs debt with the importer, as HMRC sets out for GB traffic.

Revenue holds a person or a company to account for what is declared, and software cannot sign off on a customs debt.

Human in the loop: Combining AI with broker expertise

Human in the loop is a way of working where AI captures data from customs documents and a person reviews it before it is used. The broker checks the captured fields against the original invoice, corrects them where needed and approves them. Only then does the data move on to the declaration.

This setup gives each side the work it does best. AI handles the volume of reading and typing, and the broker keeps the decisions that need knowledge of the goods and the rules.

Automate Work. Accelerate Business.

Bring together AI, ECM, and workflow automation in one powerful enterprise platform.

How AI and the broker split the work

Working together starts with a clear split: the software reads, the broker decides.

Re-keying is where errors start

Most errors start well before any classification question: they start at the keyboard. Take one truck from Holyhead to Dublin Port. That single load needs an export declaration on the GB side, an import declaration in AIS and an entry summary declaration for safety and security.

Each of those declarations draws on the same commercial invoice and packing list. A broker in a team of four types the weights and commodity codes several times over, often from a scanned PDF with 40 line items. One transposed digit in a commodity code can change the duty rate.

Multiply that by 50 loads a day and the team spends most of its hours copying. The judgment work the broker is trained for gets squeezed into whatever time is left.

Let AI capture the data

When an invoice arrives by email, AI captures the commodity code and country of origin exactly as printed on the document. It does the same for every line item on the invoice. The software takes over the typing, so your team starts every declaration from pre-filled data.

The broker still makes the call

That captured code is a starting point for the broker. If the supplier printed a six-digit code, or the wrong one, the broker still makes the classification call.

Before any captured value is used, the broker checks it against its source. The broker opens a document and sees the captured fields on one side and the original invoice on the other, so every value can be checked without switching windows.

Each value comes with a confidence score, which shows the broker where to look twice. Where a field should only hold certain values, such as an Incoterm, the broker picks the right one from a list. Once approved, the data moves on to your customs software, where the declaration is filed.

You decide how much human review is needed

Not every customs team wants to review documents in the same way. That's why Doxis gives you the flexibility to decide how Human in the Loop fits into your operations.

Some brokers prefer to review every extracted data element before it moves into their customs software. Others want to focus only on exceptions, allowing correctly matched information to move through the process without unnecessary manual checks.

For example, Doxis can be configured to identify discrepancies between:

  • Different documents: A commercial invoice and packing list showing conflicting quantities or weights.
  • Master data: An extracted value that doesn't match the information stored in your existing systems.
  • Business rules: Missing information, unexpected values or conditions that require additional checks.

Depending on how the workflow is configured, these discrepancies can be flagged for human review, allowing brokers to focus their attention where it's actually needed.

The idea isn't to remove the broker from the process, but to give them control over when and where their expertise is required.

What your brokers do with the time they get back

Once the keying is gone, brokers spend their day on the documents that need them. They review flagged shipments and resolve classification questions, and they have time to answer the importer who wants to know why a load is held at Dublin Port.

That is also the work customers pay a broker for. A team of four that no longer types 150 declarations a day can take on more loads with the same people.

Give your brokers pre-filled declarations with Doxis

Your brokers lose hours retyping the same invoice data into every declaration a load needs, and every retyped line is a chance for an error Revenue will find. With Doxis, your brokers start from pre-filled data and check it next to the original. Every classification decision stays with them.

Doxis AI.dp runs on Doxis Intelligent Content Automation, which brings document management and workflow together on one platform. Every customs document and every review decision is stored in one place, so you can answer a Revenue audit from the archive without searching inboxes.

  • Commodity codes and origin captured from invoices and packing lists
  • Every field checked against the original document on one screen
  • Rules your team sets to send risky shipments to review
  • Warnings on flagged fields so brokers know what to check
  • Approved data passed on to your customs software
  • An audit-ready archive of documents and review decisions

Doxis is a Leader in the Gartner® Magic Quadrant™ for Document Management 2026. Request a free demo and see how review works on your own invoices.

Automate Work. Accelerate Business.

Bring together AI, ECM, and workflow automation in one powerful enterprise platform.

FAQs on AI with human in the loop

Will AI replace customs brokers? 

No. AI takes over reading and keying data from customs documents, while classification and the responsibility for what is declared stay with the broker.

Is AI accurate enough for customs declarations? 

Accuracy depends on the documents and the setup, which is why a broker reviews the captured data before it is used. The review step catches errors before they reach Revenue.

Who is responsible for the declared duty when software fills in the data? 

The declarant and, depending on the type of representation, the customs agent. Software that captures data does not change who Revenue holds liable.

Can AI classify goods and assign commodity codes? 

AI can read the commodity code printed on an invoice. Deciding whether that code is correct for the goods remains a classification decision for the broker.

Does human in the loop slow declarations down? 

Reviewing pre-filled fields takes less time than typing them from scratch. Rules let your team send only the documents that need attention to review.

Which documents can AI read for customs? 

Commercial invoices and packing lists are the most common. AI can also read documents such as certificates of origin and transit documents.

Casper Luchies

Casper Luchies is an Account Executive at Doxis with a passion for helping organisations get rid of manual, paper-heavy processes. He works closely with finance, procurement and IT teams to understand where invoices, orders and other business documents slow them down, and translates those challenges into practical solutions with Doxis’s Intelligent Content Automation (ICA) platform, natively embedded in SAP. With a solid understanding of the SAP ecosystem, purchase-to-pay (P2P), order-to-cash (O2C) and e-invoicing, Casper focuses on building long-term partnerships and making sure every customer sees real, measurable value from automation.

Subscribe to Doxis News

You might also be interested in

How can we help you?

+49 (0) 30 498582-0
What is the sum of 3 and 7?

Your message has reached us!

We appreciate your interest and will get back to you shortly.

Contact us

Table of contents