The four analytical approaches to customs risk management

The World Customs Organization Risk Management Guidelines define risk management (RM) as a “systematic application of management procedures and practices providing Customs with the necessary information to address movements or consignments which present a risk”. Thus, RM is an expansive and continuous process that involves identifying the risks and threats, analysing, quantifying and classifying them utilising rigorous methods and applying appropriate countermeasures taking into account the balances and priorities of revenues collection – security – trade facilitation.

Today, most of the Customs Management Systems (CMS) have been upgraded and enhanced with the introduction of risk assessment modules, applying mainly selective and random rules. During the last decade, various risk assessment expert systems have been implemented mostly by private firms specialized in the area of customs operations and using various targeting methodologies. The main ones to be considered are as follows:

A. Machine Learning Approach

This approach uses quantitative techniques to compute a risk score on trade transactions based on the historical trade transactions and findings from customs controls (Classification, Valuation, Scanning, Inspection …). Each characteristic of Customs Declaration (Criteria) is scored and weighted using trained and validated Machine Learning (ML) models. Any new transaction will be given an overall score by combining the weighted scores of its characteristics. The computed score is thus compared against predefined acceptability thresholds (customs capacity/trade facilitation target) in order to define the appropriate action to be performed. This approach relies on an establishment of fraud feedbacks loop to update the scores which will be used to predict the risk of the future transactions.

The application of a ML approach in Customs RM brings many advantages over the classical selectivity and random targeting commonly used within Customs RM solutions. Classic examples include:

• Increased objectivity by removing human appreciation and reducing arbitrary controls.
• More accurate results by taking account of the daily controls results, to continuously update the scores of the transactions and consider fluctuating/changing non-compliance patterns.
• Impossibility for decoding by the economic operators or traders.

ML Approach relies on data which should be organised and processed correctly in order to draw robust risk patterns. The lack of systematic and accurate fraud feedback loop may constitute an issue for concern when applying this approach.

B. Selectivity Approach

The selectivity implies systematic orientation of customs declarations to a specific control channel (e.g., physical inspection or fast-track) based on pre-defined rules (risk profiles) applied on declarations submitted in the Custom Management System (CMS). For instance, imports of specific HS codes under certain regimes from Country 1, Country 2 and Country 3 and submitted by High-risk Trader or Authorized Economic Operators.

The selectivity rules are only efficient when they are based on special intelligence (i.e. based on information which has been gathered, analysed, categorised and refined) and when they are regularly updated to reflect the most recent behaviour and fraudulent practices. However, in practice, some Customs Administrations still apply selectivity rules that are based on raw or untested information, often conflicting and most of the time static or only occasionally updated by the Risk Management Unit. These kinds of rules are simply not effective and leaves room for arbitrary controls.

C. Frequency Approach

This approach consists of flagging transaction that involve economic operators, specific type of goods or countries (of supply or origin) which have no prior record in the customs database. This allows customs to ensure that knowledge is created for any new operator, HS code or route which is critical for an effective risk assessment in the future.

A frequency approach requires an effective management of the economic operators database. When the process of creation of Taxpayer Identification Numbers / TIN is not well managed, it generates a lot of identities duplication in the databases for the traders (importers and clearing agents). In this context, the system is unable to clearly identify the “true” new traders and may wrongly flag transactions for controls.

D. Random Approach

This approach applies a pure random selection methodology, used in particular to regulate Customs controls. When this approach is combined with the above-mentioned approaches, it could be configured in the manner to only redirect a small percentage of low-risk transactions for control and therefore dissuade compliant importers from fraud attempts. The positive results obtained from random selection are fed back to refresh the risk scores and are used to evaluate the performance of the others used.

The above approaches are not exclusive and should be combined in an appropriate manner to obtain a robust and accurate risk management system that helps Customs to have effective control on the trade and to reduce clearance time without jeopardizing the national security and the state revenue.

Cotecna has been combining these four approaches for years to power its risk management system – CRMS. For more support and information on implementing artificial intelligence technologies within the Customs environment, contact us at governments@cotecna.com.

For more valuable insight on risk management within Customs environment read our latest articles about the 6 key considerations for adopting AI technologies for Customs and 4 steps for a successful adoption of AI technologies.