AI use case

AI for customs classification and declaration preparation

AI that reads what is being shipped (the commercial invoice, the product data and sometimes a photo), proposes the tariff classification code with its reasoning and a confidence score, drafts the customs declaration with value, origin and parties, and sends only uncertain or high risk entries to a licensed customs specialist before filing.

By Len Debets · Last verified 27 September 2026 · 3 public deployments

90%
Reported automation rate
United Parcel Service, organization claim.
90%
Reported handling time reduction
ZLS Zoll und Logistikservice GmbH, vendor claim.
USD 400,000 to USD 4.2 million
Indicative value per year
A customs broker or cross border retailer filing 200,000 declarations a year. Worked example, see how it is calculated.

What problem does it solve?

Every item that crosses a border needs a tariff classification code. The first six digits come from the World Customs Organization's Harmonized System; countries then add their own digits, and one wrong digit can change the duty rate, trigger a licence requirement or hold the shipment. The rules for choosing a code are precise, national tariff nomenclatures change every year and the Harmonized System itself every five years, but the input is usually a vague commercial description ("parts", "gift", "cotton top") typed by a shipper who is not a customs expert.

Customs brokers, carriers and cross border retailers have handled this with specialist teams that look up codes, retype invoice data into the declaration system and chase shippers for missing facts. That model breaks when volumes jump. After the United States ended its de minimis exemption in 2025, low value parcels that had entered with minimal data suddenly needed a formal, dutiable clearance, and UPS reported a tenfold increase in daily customs entries. Errors are costly in both directions: an underpaid duty can be reassessed years later with penalties, and an incomplete declaration holds the parcel at the border while the customer waits.

How does it work?

  1. Collect the facts. The system reads the commercial invoice, packing list and product master data, or the shipper's own description at booking. Where the shipper is a consumer, some carriers ask for a photo and propose the description from it, as DHL Express does.
  2. Propose a code. A model trained on past declarations and binding rulings, combined with retrieval over the tariff nomenclature and explanatory notes, proposes the most likely code for the destination country, with the reasoning and a confidence score.
  3. Validate the entry. Rules check the code against the declared material and use, the value against the invoice, the country of origin, licence and sanctions requirements, and any additional duties that apply to that code and origin.
  4. Draft the declaration. The system fills the declaration or entry in the format of the customs system, with duties and taxes calculated.
  5. Route by confidence. High confidence, low risk entries go to filing; low confidence codes, high value goods, controlled items and missing data go to a licensed specialist, who can also ask the shipper for more information.
  6. Learn from corrections. Specialist corrections and customs queries flow back into the reference data, so the same product is classified the same way next time.
Audience
Back office
Autonomy
Supervised agent
Adoption
Early adopters
Channels
API and system to system, Internal tools, Email

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

Value benchmarks for AI for customs classification and declaration preparation
KPIMedianReported rangeData pointsClaimed by
Automation rateToo few to pool
90%
11 organization
Handling time reductionToo few to pool
90%
11 vendor
Cost reductionToo few to pool
Not pooled: up to 70%
0plus 1 up to1 vendor

Value drivers: Lower cost to serve, Speed and cycle time, Compliance quality, Customer experience.

Indicative value

A customs broker or cross border retailer filing 200,000 declarations a year

USD 400,000 to USD 4.2 million

Specialist time released per year

How this is calculated

Formula: declarations * minutesPerDeclaration / 60 * timeSavedShare * costPerHour. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Declarations or entries per year declarations, declarations per year200,000200,000The reference organization.
Specialist minutes per declaration today minutesPerDeclaration, minutes per declaration1030Editorial assumption covering classification lookups and data entry for a mixed parcel and freight workload. Swiss Post puts manual classification at 15 to 45 minutes per item; repeat items with a known code take far less, hence the lower range. Replace with your own time study.
Share of specialist time the AI removes timeSavedShare, fraction of time0.30.6Conservative against the benchmarks on this page (Digicust reports a 90% reduction in processing time per clearance at ZLS; UPS says its brokerage technology cleared 90% of dutiable parcels without manual intervention in September 2025), because the specialist still reviews the uncertain entries.
Fully loaded cost of a customs specialist costPerHour, USD per hour4070Editorial assumption, replace with your own cost.

What it leaves out: Values only the specialist time released. It leaves out the cost of the platform and the integration, the value of fewer border holds and faster delivery, and the avoided cost of duty reassessments and penalties, which can be larger but is hard to predict.

Who already uses it?

3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

DHL Express

Germany · Logistics and transportation · 2026

ProductionGrade B

In May 2026 DHL Express launched an AI feature in its international booking flow: the shipper photographs the item with a smartphone or other connected device, a server side computer vision model classifies it and proposes a structured, customs compliant item description within seconds, and the shipper reviews, edits or overrides it before submitting. DHL Express says it is live in eight markets (Canada, Germany, Hong Kong, the Netherlands, Singapore, South Africa, Spain and the United Arab Emirates) with a wider rollout planned through 2026. The announcement does not say who built the model, and no outcome figures were disclosed.

No outcome disclosed.

United Parcel Service

United States · Logistics and transportation · 2025

ScaledGrade B

UPS, one of the largest customs brokers, told investors on its third quarter 2025 earnings call that it enhanced its customs brokerage with agentic AI to cope with a tenfold rise in daily customs entries after the United States ended the de minimis exemption. Chief executive Carol Tomé said that in March 2025 about 21% of the 13,000 packages a day that needed a dutiable clearance were cleared without manual intervention, against 90% of 112,000 packages a day in September 2025. UPS also says it uses AI and machine learning to check Harmonized System classification codes, tariff rules and policies on imports.

  • Automation rate: 90%, September 2025, US packages needing a dutiable clearance (112,000 a day)
    "In September 2025, the carrier cleared 90% of 112,000 daily packages with no manual intervention."
    Claimed by: organization

ZLS Zoll und Logistikservice GmbH

Germany · Logistics and transportation · 2025

ProductionGrade C

ZLS, a customs clearance and warehousing firm in Neuhaus am Inn, Germany, founded in 2018 with a focus on trade with Turkey, uses Digicust's AI platform to take over the manual, error prone data entry for its customs clearances. An earlier German version of the case study names the product as Dexter IDP and says it fills customs declarations from different data sources and assigns tariff numbers from the goods description. Digicust reports that the processing time per extensive clearance fell from three to four hours to 10 to 15 minutes, with costs down by up to 70% and fewer entry errors.

  • Handling time reduction: 90%, per extensive customs clearance, from 3 to 4 hours to 10 to 15 minutes of processing
    "How ZLS Logistik Service achieved a 90% reduction in processing time and 70% cost reduction through Digicust's AI-powered automation solution"
    Claimed by: vendor
  • Cost reduction: up to 70%
    "Up to 70% reduction in costs through automation, allowing ZLS to improve their competitive position in the market."
    Claimed by: vendor

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Historical declarations with the final, accepted classification codes
  • Product master data with materials, use and composition
  • Current tariff nomenclature, explanatory notes and binding rulings for each destination
  • Lists of controlled, licensed and sanctioned goods and additional duty measures

Systems to integrate

  • Customs filing system or national single window (for example ICS2 in the EU or ACE in the US)
  • Order, booking or transport management system that supplies shipment data
  • Product information or ERP system with the item master
  • Duty and tax calculation engine
  • Case tool for specialist review and shipper queries

Complexity: Medium

Proposing a code from a good description is well understood. The work is in the data (clean product master data, past declarations with the final code), the integration with the customs filing system and the broker's workflow, and in drawing the line between what may be filed automatically and what a licensed person must review.

  1. 1

    Measure the baseline

    Take a sample of recent entries and record the time per entry, the share with a later correction or customs query, and where the delays sit (classification, missing data, retyping). This is the benchmark the AI has to beat.

  2. 2

    Build the reference set

    Assemble past declarations with the final accepted code, the binding rulings you hold and the product master. Remove codes that customs later corrected, or the model learns the mistakes.

  3. 3

    Run in shadow mode

    Let the AI propose codes and draft entries next to the specialists for several weeks. Compare at the level that matters (six digits and the full national code) per product family, and set the confidence threshold per family, not one number for everything.

  4. 4

    Automate the confident, low risk share

    File automatically only where the confidence is above the threshold and the goods are not controlled, high value or subject to additional duties. Everything else goes to a specialist with the AI's reasoning attached.

  5. 5

    Close the loop with shippers

    Ask for missing facts at the moment of booking, in plain language, instead of after the parcel is held. A photo or a guided question is cheaper than a hold.

  6. 6

    Monitor and recalibrate

    Track corrections, customs queries and post clearance audits per code and retrain before each annual nomenclature change.

Guardrails

  • Automatic filing only above a confidence threshold set per product family, and never for controlled, sanctioned or licensed goods
  • Every proposed code carries its reasoning and the rule or ruling it relies on, kept with the entry for audit
  • The licensed broker or declarant remains responsible and signs off the rules and thresholds
  • Declared value and origin are checked against the invoice and never changed by the model
  • Personal data of consignees is limited to what the declaration requires and masked in logs

KPIs to instrument

  • Share of entries filed without manual intervention
  • Classification accuracy at six digits and at the full national code, on a weekly checked sample
  • Customs holds, queries and post clearance corrections per thousand entries
  • Specialist minutes per entry and time from arrival to release
  • Duty reassessments and penalties over time

Human in the loop

Licensed customs specialists review every low confidence code, every controlled or high value item and every entry customs queries. They own the classification rules, approve threshold changes and sample automatically filed entries every week, because a classification error can repeat across thousands of shipments before an audit finds it.

Common failure modes

Confidently wrong at scale
A systematic error on one product family repeats on every shipment until an audit finds it. Sample automatically filed entries and watch corrections per code.
Garbage descriptions in, garbage codes out
The model cannot classify "gift" or "parts" reliably. Push back to the shipper for facts instead of guessing, and treat vague descriptions as low confidence.
Stale tariff data
Nomenclature changes and new additional duties make yesterday's correct code wrong. Update the reference data on the effective date and rerun affected products.
Automation that hides accountability
The declarant is liable whatever the tool did. Keep the reasoning, the data used and the approver with every entry.

What are the risks and rules?

EU AI Act

Depends on design

The classification and declaration work in the back office is minimal risk: classifying goods and preparing customs declarations is not listed in Annex III, which covers border control only where AI assesses natural persons (point 7). If the design adds a shipper facing assistant that asks for missing information in chat or by email, that assistant is limited risk and carries the transparency duty of Article 50 (people must know they are dealing with AI). The obligations that matter most come from customs law: the declarant stays responsible for the accuracy of the declaration whatever tool prepared it.

Rules that apply

Guidance

Controls to put in place

  • Documented classification rules and confidence thresholds per product family, approved by a licensed specialist
  • Audit trail per entry with the proposed code, reasoning, data used and approver
  • Weekly sampling of automatically filed entries and monitoring of customs queries and corrections
  • Change control on reference data around each nomenclature update
  • Controlled goods and sanctions checks that the AI cannot bypass

Frequently asked questions

Can AI classify goods for customs without a human?
For standard goods with good descriptions, often yes, within thresholds a licensed specialist sets. UPS says its brokerage technology, which it enhanced with agentic AI, cleared 90% of US parcels that needed a dutiable clearance without manual intervention in September 2025 (about 21% in March 2025). That figure covers the whole entry, not classification alone, the other 10% needed people, and the declarant stays legally responsible for every code.
How much time does AI save a customs broker?
It depends on how much of the work is retyping and looking up codes. Digicust reports that processing time per extensive clearance at the German customs agent ZLS fell from three to four hours to 10 to 15 minutes. Plan more conservatively for mixed or complex goods.
Does it help consumers and small shippers too?
Yes, at the moment of booking. DHL Express lets shippers photograph an item and proposes a customs compliant description that they can edit, live in eight markets since May 2026, to avoid holds caused by vague descriptions.

How to cite this page

Blits.ai AI Use Case Library, "AI for customs classification and declaration preparation", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/customs-classification-and-declaration. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 27 September 2026: First published

Related use cases

Cross industryGovernment and public sector

AI document intelligence for unstructured forms and documents

AI that takes documents in any format, such as scanned forms, PDFs, photos, emails and handwritten notes, splits and classifies them, extracts the required fields with a confidence score, validates them against business rules and source systems, and sends only the uncertain cases to a person before the data enters the downstream process.

Deployments
5 public, best grade B
Reported accuracy
at least 90%
Ancine, vendor claim
Logistics and transportation

AI agent for parcel tracking and delivery exceptions

An AI agent that answers "where is my parcel" and resolves delivery exceptions for parcel carriers and postal operators, such as missed deliveries, redelivery or a change of address or pickup point, delays, customs holds and lost or damaged parcel claims, on chat, messaging and phone, and hands disputes and claims above set limits to a human with the tracking history attached.

Deployments
5 public, best grade B
Reported contact deflection
50%
Evri, organization claim
Cross industryBanking

AI for continuous controls testing and control self assessment

AI that moves control testing from periodic samples to continuous, full population assurance: it collects evidence from source systems, maps each artefact to the control it supports, tests every transaction or record against the control's rule, flags exceptions for a human to judge and prepares the risk and control self assessment from incident and loss data for the business to review.

Deployments
3 public, best grade B
Autonomy
Supervised agent
Banking

AI examination of trade documents under letters of credit and collections

AI that reads the full document presentation under a letter of credit or collection (bill of lading, commercial invoice, packing list, certificates), extracts and cross checks the data, tests it against the instructions and the ICC rules (for letters of credit, the credit terms, UCP 600 and ISBP), and lists discrepancies by severity with the rule cited, so qualified examiners focus on the genuine exceptions.

Deployments
3 public, best grade B
Autonomy
Supervised agent