EU Customs Reform and ICS2: Software Prep for 2026

A Rotterdam-based 3PL filed Entry Summary Declarations through its old customs broker until 2023. By the end of 2024 the operation moved to direct house-level filings under ICS2. In Q1 2026 the team is filing roughly 18,000 ENS records per week, with reject codes appearing on approximately 6% of them. Each reject is a truck stopped at the port, a container held in inspection, or a driver waiting two hours for a re-submission. The 2025 leniency from customs authorities is gone. The 2028 customs reform is on the horizon. The software stack has to handle both.

This post is for freight forwarders, 3PL operators, and customs broker teams scoping their 2026 to 2028 customs software. It assumes the basics of ICS2 (Import Control System 2) and goes deeper into the operational realities of running ENS filings at production scale, the NCTS Phase 5 cutover lessons, and the EU Customs Reform that will reshape data flows starting in 2028.

The 2026 ICS2 baseline: what every ENS filing must contain

ICS2 Release 3 went live on 1 March 2024 for maritime and air carriers, and expanded on 1 September 2024 to road and rail. The deployment window for full house-level filings closed on 1 April 2025. As of 2026, pre-arrival ENS declarations are fully mandatory for all modes, all consignments, all importers.

The ENS dataset under ICS2 R3 is materially richer than ICS1. Required data points include:

• 6-digit Harmonized System code (not just chapter level). Generic “general merchandise” descriptions are rejected.
• Accurate consignor and consignee, with full address and contact data. Forwarder names in the consignee field are rejected.
• Goods description clear enough to identify the actual product. “Electronics” or “textiles” alone is rejected.
• House-level data for consolidated consignments, not just master-level.
• Container, package, and seal numbers with explicit volumetric data.

The data quality bar is higher than most legacy TMS extracts can meet without intervention. The 6% reject rate the Rotterdam operation sees is typical; some forwarders report 8 to 12% before remediation. The economic impact is direct: every reject is a re-submission, a customs broker hour, and often a vehicle or container dwell cost.

The integration model that works at scale has three layers:

1. A data quality gate that runs ENS pre-validation before submission. Same rules as the customs authority validator, surfaced upstream so rejects happen against your own copy first.
2. A normalization service that takes shipment data from operational systems (TMS, WMS, eAWB feeds, sea waybill feeds) and produces a canonical ENS-ready record.
3. A submission service that talks to the national customs authority gateway, manages retries, and surfaces structured errors back to operations.

The gate is the part that pays back fastest. A pre-validation that catches 70% of would-be rejects upstream cuts re-submission load by an order of magnitude and removes the operations team from the customs error queue.

NCTS Phase 5 cutover lessons

The New Computerised Transit System Phase 5 went live across EU and Common Transit Convention countries by mid-2024, replacing Phase 4. NCTS handles transit declarations: goods moving under customs control between origin and destination customs offices.

Phase 5 changed the technical interface (new message structures, broader data sets) and added new functionality (House Consignment within Transit, separate declarations for safety-and-security data). The cutover surfaced patterns worth documenting because the EU Customs Reform will replay them at larger scale in 2028:

• Test environments lag production. Several national NCTS test environments were unavailable or unstable for weeks after production cutover. Building integrations against incomplete test data is a known cost.
• Reject message formats vary by member state. The XML schema is harmonized; the human-readable explanation of why a declaration failed is not. Operations teams need a translation table per national authority.
• High-volume operations need parallel filing capability. Single-stream submission queues do not scale past 5,000 to 10,000 declarations per day per gateway. Production-grade integrations use connection pools and per-gateway rate limiters.
• Late corrections are expensive. Phase 5 allowed amendments under a stricter window. Logistics platforms now retain the full pre-submission state to support corrections without rebuilding the declaration from scratch.

By 2026, most forwarders have stabilized on Phase 5. The lessons survive the cutover and apply to whatever comes next.

EU Customs Data Hub: what to architect for now even though go-live is 2028

The EU Customs Reform package (Commission proposal COM/2023/258) reached political agreement in late 2025. The reform creates a single EU Customs Authority and a single EU Customs Data Hub. Roll-out is phased from 2028, with full mandatory adoption staged into the early 2030s.

The Data Hub model is significantly different from the current per-member-state customs IT landscape:

• A single submission point for customs data, replacing 27 national portals.
• A unified data model that absorbs both customs declarations and safety-and-security data.
• A new economic operator status: Trust and Check trader, an evolution of AEO that allows direct data sharing rather than declaration-by-declaration filing.
• Removal of the EUR 150 de minimis for low-value consignments, hitting eCommerce parcel logistics especially hard.

For 2026 architecture, the Data Hub is too far away to integrate against directly (the specification is in draft and the API surfaces are not stable). But it should inform two design choices today:

1. Data should live in a canonical internal model, not in the format of the national gateway you happen to use. When the Data Hub goes live, the team that translates from a clean canonical model to a new API spends weeks; the team that has to refactor a customs-gateway-shaped data model spends quarters.
2. Trust and Check readiness is the new AEO ambition. Trust and Check requires real-time data sharing with the customs authority, full traceability of goods movement, and demonstrated control quality. Logistics operators that target Trust and Check status by 2028 to 2030 need to start building the data foundation now.

Eliminating the EUR 150 threshold: parcel-volume software implications

The EU imported approximately 4.6 billion low-value parcels in 2024, more than 90% from China. Under current rules, parcels under EUR 150 are exempt from customs duty (though VAT applies under IOSS or special arrangements). The Customs Reform removes the EUR 150 threshold entirely.

The operational implication is enormous. Every parcel will need a full customs declaration. For dominant cross-border eCommerce flows (AliExpress, Temu, Shein, plus regular international DTC retail), this multiplies the declaration volume by orders of magnitude.

The software response that is taking shape:

• AI-assisted HS code classification at the parcel level, replacing the current marketplace-supplied (often inaccurate) HS data. Classification accuracy in the 88 to 94% range is achievable with a fine-tuned model trained on parcel descriptions, brand, and product imagery.
• Bulk filing APIs at the platform level rather than per-parcel. The Data Hub specification is expected to permit aggregated submissions for high-volume operators.
• Pre-clearance flows where the parcel data is filed before the parcel physically arrives at the EU border, with the customs decision propagating back to the originating platform.
• Declarant-of-record arrangements with marketplaces taking on declarant responsibility for non-EU sellers.

Forwarders, postal operators, and integrators (DHL, FedEx, UPS, GLS, DPD, plus the national postal services) are all investing in parcel-volume customs platforms now. Mid-size 3PLs that handle cross-border eCommerce flows for their customers face a build-or-partner decision in 2026 to 2027.

Build vs buy: CargoWise customs, AEB ASSIST4, or custom on top of TMS

The customs software market for European logistics has consolidated around three patterns:

• TMS-native customs modules. CargoWise (WiseTech Global) integrates customs deeply into the forwarder TMS. Transporeon offers customs functionality through partnerships. Descartes provides a broad customs and trade compliance suite with strong North American and growing EU footprint.
• Specialist customs platforms. AEB ASSIST4 (Stuttgart-based, strong in DACH region). Dakosy (Hamburg, large maritime focus). Customs4trade CAS (Belgium, broad EU coverage). MIC Customs Solutions (Austria, multinational shippers). These platforms are typically deeper on EU customs functionality than TMS-native modules.
• Custom microservices on top of an existing TMS. Realistic at large forwarders (1 million plus annual declarations) with in-house engineering. Not realistic for mid-market.

The decision usually comes down to declaration volume and the diversity of customs procedures. A forwarder handling primarily standard import flows in two member states gets adequate coverage from a TMS-native module. One handling transit, IOSS, customs warehousing, and inward processing across six member states almost always ends up with a specialist platform.

Pricing in 2026 is competitive: EUR 0.30 to EUR 2.50 per declaration depending on complexity and volume, with implementation costs of EUR 50,000 to EUR 350,000 for mid-size operators.

Data quality automation: HS code classification with ML

HS code accuracy is the single most important data quality lever in customs operations. A wrong HS code at the 6-digit level triggers a reject; at the 8 or 10-digit level it triggers an audit, a duty correction, or a penalty.

The 2024 to 2026 evolution of HS classification automation:

• Rule-based classification (text matching against HS descriptions) reaches 60 to 75% accuracy on typical commercial descriptions. Useful for high-confidence cases but leaves a long tail.
• Embedding-based retrieval against the full HS code text catalog reaches 80 to 88% top-1 accuracy on real commercial descriptions, with top-5 accuracy in the mid 90s.
• LLM-based classification with retrieval-augmented prompting and structured output reaches 92 to 96% top-1 on well-described goods, with structured uncertainty flagging that lets human reviewers focus on the actual borderline cases.

Production-grade classification automation in 2026 typically combines all three: rule-based for high-confidence matches, embedding retrieval for narrowing, and LLM with retrieval for the long tail. Accuracy claims should always be measured per goods category; classification is much harder for electronics and chemicals than for furniture or apparel.

A measurable target: a forwarder that introduces ML-assisted classification typically reduces ENS reject rates from 6 to 8% down to 1 to 3% within six months, with reject reduction concentrated in the goods-description and HS-code fields.

Trust and Check trader readiness: audit trail, real-time data sharing

Trust and Check is the new economic operator status under the Customs Reform. It builds on AEO (Authorized Economic Operator) but extends in two directions:

• Continuous, structured data sharing with the customs authority rather than declaration-by-declaration filing.
• Self-assessment of duty and import VAT under defined parameters, audited rather than pre-cleared.

The economic case is strong: Trust and Check traders get streamlined clearance, lower controls, and potentially deferred duty payment. The infrastructure cost is real.

For a forwarder or shipper targeting Trust and Check status by 2030, the foundations to build in 2026 and 2027:

• A canonical, complete record of every shipment, end-to-end, that can be queried by the customs authority on demand.
• Structured supply chain control evidence (supplier vetting, origin documentation, valuation evidence) accessible electronically.
• Operational monitoring that surfaces compliance anomalies before they propagate to a declaration.
• An audit posture that can produce three years of clean records on 48 hours’ notice.

These are the same foundations that make ICS2 operations cleaner today. The work is double-counted in a good way.

A 2026 punch list for European customs software

Based on what is and is not working at production scale across European forwarders:

1. ICS2 reject rate below 3%. If your current rate is above that, the gate-and-normalize pattern pays back in two quarters.
2. NCTS Phase 5 fully stable with documented error translation per member state.
3. HS classification at 92% or better for goods description sourced from customer feeds.
4. Customs data in a canonical internal model, separate from any specific gateway adapter.
5. AEO status confirmed, with a documented path toward Trust and Check by 2028 to 2030.
6. Volumetric capability to handle 5x current daily declaration volume without queue collapse, as preparation for the EUR 150 threshold removal.

The customs software stack is no longer a side concern. It is a primary determinant of border throughput, customer experience, and audit cost. Forwarders that treat it that way are gaining share in 2026.