Nohena Insights · Engineering
How AI document extraction improves customs declarations
AI document extraction improves customs declarations by reducing manual data entry errors and creating a verifiable link between source documents and the final declaration data, increasing accuracy and reducing compliance risk.
Nohena · 12 October 2026 · 9 min read
How document extraction improves Nigerian customs declarations
AI document extraction improves Nigerian customs declarations by systematically reducing manual data entry errors and creating a verifiable link between source documents and the final declaration data. This process provides a structured, auditable foundation for the Single Goods Declaration (SGD), which helps licensed agents prepare more accurate and defensible lodgements.
For decades, the work of preparing a customs declaration has been defined by manual transcription. An agent or their data entry clerk would sit with a stack of papers, a commercial invoice, a packing list, a bill of lading, and a Pre-Arrival Assessment Report (PAAR), and manually type the relevant details into a declaration form. This workflow is not only slow but also a significant source of compliance risk. A single misplaced decimal point, a transposed digit in an HS code, or an incorrect currency code can lead to customs queries, delays, and penalties. The rise of robust AI document extraction for customs provides a precise, scalable alternative to this legacy process. You can learn more about related compliance topics in our insights section .
How AI document extraction for customs works
AI document extraction for customs uses optical character recognition and natural language processing to identify, read, and structure key data points from trade documents like invoices, packing lists, and bills of lading. It transforms unstructured information, such as PDF files and images, into structured data that can be used to populate a declaration.
The technology operates through a multi-stage process:
Document Ingestion: The system accepts a variety of file formats, including PDF, JPG, and PNG, which represent the standard outputs of commercial and logistics operations.
Text Recognition: Using Optical Character Recognition (OCR), the system converts the pixels of an image into machine-readable text characters. This is the foundational step of reading the document.
Contextual Understanding: This is where the 'AI' becomes critical. The system uses natural language processing (NLP) models trained on millions of trade documents to understand the context of the text. It learns to identify not just that a string of characters is a date, but that it is the bill of lading date , or that a number represents the total invoice value in a specific currency.
Data Structuring: Finally, the system maps the identified and validated data points to their corresponding fields in a customs declaration schema, such as the one required for Nigeria's Single Goods Declaration (SGD).
Key documents for extraction
A robust extraction process is built to understand the specific documents essential for Nigerian import clearance. The data from these sources must be consistent and accurate for a smooth declaration process.
Commercial Invoice: This is the primary document for valuation. Key extracted fields include the seller, buyer, invoice number and date, Incoterms, currency, item descriptions, quantities, unit prices, and total value.
Packing List: This document corroborates the physical details of the shipment. Extraction focuses on quantities, net and gross weights, and package details to ensure they align with the commercial invoice and freight documents.
Bill of Lading (B/L) or Air Waybill (AWB): This is the contract of carriage. The system extracts the consignee, shipper, port of loading, port of discharge, vessel or flight details, container numbers, and the description of goods.
Form M: As the mandatory pre-import document, the Form M establishes a baseline for the transaction. Its number is a critical data point for linking the shipment to its initial authorisation.
Pre-Arrival Assessment Report (PAAR): Issued by the Nigeria Customs Service (NCS), the PAAR is a critical control document. Extraction captures the NCS-approved HS code, valuation assessment, and preliminary duty calculation, which are essential for the final declaration.
Regulatory Certificates: For regulated goods, certificates like the SONCAP for general goods or NAFDAC permits for food and drugs are required. The system can extract certificate numbers, issue dates, and expiry dates to confirm compliance.
The impact of extraction on declaration accuracy
Automated extraction directly improves declaration accuracy by eliminating transposition errors, ensuring consistency across documents, and flagging discrepancies before lodgement. This shifts the agent's work from tedious data entry to high-value review and validation.
Reducing manual data entry errors
Manual data entry is inherently susceptible to human error. These small mistakes can have significant consequences.
Value Transposition: Typing an invoice value of USD 15,200.00 as USD 12,500.00 or USD 152,000.00 will lead to an incorrect customs value and duty assessment, triggering queries and potential penalties.
HS Code Errors: The Nigerian tariff uses a 10-digit HS code structure based on the ECOWAS Common External Tariff (CET). A one-digit error can change a product's classification entirely, shifting the duty rate from 5% to 20%, for example. Automated extraction from the PAAR ensures the correct, NCS-provided code is used.
Currency and Unit Mismatches: Incorrectly keying in the currency (e.g., EUR instead of USD) or the unit of measure (e.g., cartons instead of pieces) creates fundamental inconsistencies in the declaration.
Ensuring cross-document consistency
A customs declaration is a synthesis of multiple documents. The NCS risk management system, and platforms like the B'Odogwu portal, are designed to cross-reference data. Discrepancies are a primary trigger for interventions.
By extracting data from all source documents into a single, structured view, a system can automatically run consistency checks. This is a core function of decision support.
For example, it can flag if:
The FOB value on the commercial invoice does not match the value declared on the Form M.
The gross weight on the packing list differs significantly from the weight on the bill of lading.
The goods description on the invoice is inconsistent with the description on the bill of lading or the PAAR.
Identifying these conflicts before lodging the SGD allows the agent to seek clarification from the importer and resolve the issues, preparing a more coherent and defensible declaration.
Verifying against regulatory requirements
Accurate data is the foundation for applying complex customs rules correctly.
Valuation: The invoice value is the starting point for customs value. Extraction ensures this is captured precisely. The system can then correctly apply additions mandated by the WTO Valuation Agreement's Article 8, such as freight costs and insurance. If an insurance certificate is not provided, NCS procedures mandate the addition of 1.5% of the Cost and Freight (CFR) value. Automation ensures these calculations are applied consistently.
Tariff Classification: By extracting the 10-digit HS code from the PAAR, the system ensures the declaration uses the classification already assessed by customs, minimizing the risk of a classification dispute post-lodgement.
Preferential Origin: For goods claiming preferential treatment under schemes like the AfCFTA or the ECOWAS Trade Liberalisation Scheme (ETLS), accurate data is essential. Extraction can pull the country of origin, which is the first step in assessing eligibility for reduced tariffs, such as the 0.5% ETLS levy. This helps in preparing the documentation needed to prove compliance with rules of origin, like the AfCFTA's ~40% regional value content threshold.
Manual data entry vs automated extraction
Automated extraction offers superior speed, accuracy, and auditability compared to manual data entry, which is prone to human error and is significantly slower at scale. The choice between these methods is a choice between a traditional, high-risk workflow and a modern, risk-managed one.
Integrating extraction into the clearance workflow
Integrating document extraction into the clearance workflow involves using it as a preparatory step to create a draft declaration that a licensed agent then validates, refines, and lodges. This model preserves the agent's essential role while augmenting their capabilities.
A modernised workflow looks like this:
Document Collation: The importer or agent gathers all necessary digital documents (invoice, PAAR, B/L, etc.).
Automated Processing & Verification: The documents are uploaded into a decision support system. The platform uses AI to extract and structure the data, automatically populating a draft SGD. It simultaneously flags any identified discrepancies between documents or against known rules.
Expert Validation by Agent: A licensed customs agent reviews the extracted data and the flagged issues. This is the critical human-in-the-loop step. The agent applies their expertise to resolve ambiguities, confirm the tariff classification and valuation treatment, and make final judgements.
Preparation for Lodgement: Once validated by the agent, the system prepares a complete, lodgement-ready declaration and document pack. Nohena prepares ; the licensed agent lodges.
Lodgement by Agent: The licensed agent, as the legally responsible party, formally lodges the verified declaration through the official customs portal, such as NICIS II or its successor.
This workflow respects the lodgement boundary. The technology prepares and recommends; the human expert decides and acts.
FAQ
Is AI document extraction 100% accurate?
No system is 100% accurate, but high-quality AI extraction models achieve very high accuracy rates, often exceeding human performance for repetitive tasks. The key is a workflow that includes human validation by a licensed professional to review the extracted data and handle complex edge cases or poor-quality documents before lodgement.
Can this technology handle handwritten documents?
The ability to read handwritten text, known as Intelligent Character Recognition (ICR), varies. While modern systems have improved, accuracy is generally lower than for typed text and depends heavily on the clarity of the handwriting. For critical customs declarations, typed documents are strongly preferred to ensure data integrity.
Does automated extraction replace the need for a customs agent?
No, it enhances the agent's capabilities rather than replacing them. Extraction automates the low-value, repetitive task of data entry, freeing up the agent to focus on high-value work: strategic advice, classification expertise, valuation assessment, risk management, and handling exceptions. The agent remains responsible for the final validation and lodgement of the declaration.
How does document extraction help with customs audits?
It creates a clear, digital audit trail. Each piece of data in the declaration can be electronically traced back to its specific location on a source document. This makes it much faster and easier to respond to information requests from the Nigeria Customs Service (NCS) and demonstrate due diligence during post-clearance audits.